1c Copyright (C) 1988-2021 Free Software Foundation, Inc. 2 3@c This is part of the GCC manual. 4@c For copying conditions, see the file gcc.texi. 5 6@node C Extensions 7@chapter Extensions to the C Language Family 8@cindex extensions, C language 9@cindex C language extensions 10 11@opindex pedantic 12GNU C provides several language features not found in ISO standard C@. 13(The @option{-pedantic} option directs GCC to print a warning message if 14any of these features is used.) To test for the availability of these 15features in conditional compilation, check for a predefined macro 16@code{__GNUC__}, which is always defined under GCC@. 17 18These extensions are available in C and Objective-C@. Most of them are 19also available in C++. @xref{C++ Extensions,,Extensions to the 20C++ Language}, for extensions that apply @emph{only} to C++. 21 22Some features that are in ISO C99 but not C90 or C++ are also, as 23extensions, accepted by GCC in C90 mode and in C++. 24 25@menu 26* Statement Exprs:: Putting statements and declarations inside expressions. 27* Local Labels:: Labels local to a block. 28* Labels as Values:: Getting pointers to labels, and computed gotos. 29* Nested Functions:: Nested function in GNU C. 30* Nonlocal Gotos:: Nonlocal gotos. 31* Constructing Calls:: Dispatching a call to another function. 32* Typeof:: @code{typeof}: referring to the type of an expression. 33* Conditionals:: Omitting the middle operand of a @samp{?:} expression. 34* __int128:: 128-bit integers---@code{__int128}. 35* Long Long:: Double-word integers---@code{long long int}. 36* Complex:: Data types for complex numbers. 37* Floating Types:: Additional Floating Types. 38* Half-Precision:: Half-Precision Floating Point. 39* Decimal Float:: Decimal Floating Types. 40* Hex Floats:: Hexadecimal floating-point constants. 41* Fixed-Point:: Fixed-Point Types. 42* Named Address Spaces::Named address spaces. 43* Zero Length:: Zero-length arrays. 44* Empty Structures:: Structures with no members. 45* Variable Length:: Arrays whose length is computed at run time. 46* Variadic Macros:: Macros with a variable number of arguments. 47* Escaped Newlines:: Slightly looser rules for escaped newlines. 48* Subscripting:: Any array can be subscripted, even if not an lvalue. 49* Pointer Arith:: Arithmetic on @code{void}-pointers and function pointers. 50* Variadic Pointer Args:: Pointer arguments to variadic functions. 51* Pointers to Arrays:: Pointers to arrays with qualifiers work as expected. 52* Initializers:: Non-constant initializers. 53* Compound Literals:: Compound literals give structures, unions 54 or arrays as values. 55* Designated Inits:: Labeling elements of initializers. 56* Case Ranges:: `case 1 ... 9' and such. 57* Cast to Union:: Casting to union type from any member of the union. 58* Mixed Labels and Declarations:: Mixing declarations, labels and code. 59* Function Attributes:: Declaring that functions have no side effects, 60 or that they can never return. 61* Variable Attributes:: Specifying attributes of variables. 62* Type Attributes:: Specifying attributes of types. 63* Label Attributes:: Specifying attributes on labels. 64* Enumerator Attributes:: Specifying attributes on enumerators. 65* Statement Attributes:: Specifying attributes on statements. 66* Attribute Syntax:: Formal syntax for attributes. 67* Function Prototypes:: Prototype declarations and old-style definitions. 68* C++ Comments:: C++ comments are recognized. 69* Dollar Signs:: Dollar sign is allowed in identifiers. 70* Character Escapes:: @samp{\e} stands for the character @key{ESC}. 71* Alignment:: Determining the alignment of a function, type or variable. 72* Inline:: Defining inline functions (as fast as macros). 73* Volatiles:: What constitutes an access to a volatile object. 74* Using Assembly Language with C:: Instructions and extensions for interfacing C with assembler. 75* Alternate Keywords:: @code{__const__}, @code{__asm__}, etc., for header files. 76* Incomplete Enums:: @code{enum foo;}, with details to follow. 77* Function Names:: Printable strings which are the name of the current 78 function. 79* Return Address:: Getting the return or frame address of a function. 80* Vector Extensions:: Using vector instructions through built-in functions. 81* Offsetof:: Special syntax for implementing @code{offsetof}. 82* __sync Builtins:: Legacy built-in functions for atomic memory access. 83* __atomic Builtins:: Atomic built-in functions with memory model. 84* Integer Overflow Builtins:: Built-in functions to perform arithmetics and 85 arithmetic overflow checking. 86* x86 specific memory model extensions for transactional memory:: x86 memory models. 87* Object Size Checking:: Built-in functions for limited buffer overflow 88 checking. 89* Other Builtins:: Other built-in functions. 90* Target Builtins:: Built-in functions specific to particular targets. 91* Target Format Checks:: Format checks specific to particular targets. 92* Pragmas:: Pragmas accepted by GCC. 93* Unnamed Fields:: Unnamed struct/union fields within structs/unions. 94* Thread-Local:: Per-thread variables. 95* Binary constants:: Binary constants using the @samp{0b} prefix. 96@end menu 97 98@node Statement Exprs 99@section Statements and Declarations in Expressions 100@cindex statements inside expressions 101@cindex declarations inside expressions 102@cindex expressions containing statements 103@cindex macros, statements in expressions 104 105@c the above section title wrapped and causes an underfull hbox.. i 106@c changed it from "within" to "in". --mew 4feb93 107A compound statement enclosed in parentheses may appear as an expression 108in GNU C@. This allows you to use loops, switches, and local variables 109within an expression. 110 111Recall that a compound statement is a sequence of statements surrounded 112by braces; in this construct, parentheses go around the braces. For 113example: 114 115@smallexample 116(@{ int y = foo (); int z; 117 if (y > 0) z = y; 118 else z = - y; 119 z; @}) 120@end smallexample 121 122@noindent 123is a valid (though slightly more complex than necessary) expression 124for the absolute value of @code{foo ()}. 125 126The last thing in the compound statement should be an expression 127followed by a semicolon; the value of this subexpression serves as the 128value of the entire construct. (If you use some other kind of statement 129last within the braces, the construct has type @code{void}, and thus 130effectively no value.) 131 132This feature is especially useful in making macro definitions ``safe'' (so 133that they evaluate each operand exactly once). For example, the 134``maximum'' function is commonly defined as a macro in standard C as 135follows: 136 137@smallexample 138#define max(a,b) ((a) > (b) ? (a) : (b)) 139@end smallexample 140 141@noindent 142@cindex side effects, macro argument 143But this definition computes either @var{a} or @var{b} twice, with bad 144results if the operand has side effects. In GNU C, if you know the 145type of the operands (here taken as @code{int}), you can avoid this 146problem by defining the macro as follows: 147 148@smallexample 149#define maxint(a,b) \ 150 (@{int _a = (a), _b = (b); _a > _b ? _a : _b; @}) 151@end smallexample 152 153Note that introducing variable declarations (as we do in @code{maxint}) can 154cause variable shadowing, so while this example using the @code{max} macro 155produces correct results: 156@smallexample 157int _a = 1, _b = 2, c; 158c = max (_a, _b); 159@end smallexample 160@noindent 161this example using maxint will not: 162@smallexample 163int _a = 1, _b = 2, c; 164c = maxint (_a, _b); 165@end smallexample 166 167This problem may for instance occur when we use this pattern recursively, like 168so: 169 170@smallexample 171#define maxint3(a, b, c) \ 172 (@{int _a = (a), _b = (b), _c = (c); maxint (maxint (_a, _b), _c); @}) 173@end smallexample 174 175Embedded statements are not allowed in constant expressions, such as 176the value of an enumeration constant, the width of a bit-field, or 177the initial value of a static variable. 178 179If you don't know the type of the operand, you can still do this, but you 180must use @code{typeof} or @code{__auto_type} (@pxref{Typeof}). 181 182In G++, the result value of a statement expression undergoes array and 183function pointer decay, and is returned by value to the enclosing 184expression. For instance, if @code{A} is a class, then 185 186@smallexample 187 A a; 188 189 (@{a;@}).Foo () 190@end smallexample 191 192@noindent 193constructs a temporary @code{A} object to hold the result of the 194statement expression, and that is used to invoke @code{Foo}. 195Therefore the @code{this} pointer observed by @code{Foo} is not the 196address of @code{a}. 197 198In a statement expression, any temporaries created within a statement 199are destroyed at that statement's end. This makes statement 200expressions inside macros slightly different from function calls. In 201the latter case temporaries introduced during argument evaluation are 202destroyed at the end of the statement that includes the function 203call. In the statement expression case they are destroyed during 204the statement expression. For instance, 205 206@smallexample 207#define macro(a) (@{__typeof__(a) b = (a); b + 3; @}) 208template<typename T> T function(T a) @{ T b = a; return b + 3; @} 209 210void foo () 211@{ 212 macro (X ()); 213 function (X ()); 214@} 215@end smallexample 216 217@noindent 218has different places where temporaries are destroyed. For the 219@code{macro} case, the temporary @code{X} is destroyed just after 220the initialization of @code{b}. In the @code{function} case that 221temporary is destroyed when the function returns. 222 223These considerations mean that it is probably a bad idea to use 224statement expressions of this form in header files that are designed to 225work with C++. (Note that some versions of the GNU C Library contained 226header files using statement expressions that lead to precisely this 227bug.) 228 229Jumping into a statement expression with @code{goto} or using a 230@code{switch} statement outside the statement expression with a 231@code{case} or @code{default} label inside the statement expression is 232not permitted. Jumping into a statement expression with a computed 233@code{goto} (@pxref{Labels as Values}) has undefined behavior. 234Jumping out of a statement expression is permitted, but if the 235statement expression is part of a larger expression then it is 236unspecified which other subexpressions of that expression have been 237evaluated except where the language definition requires certain 238subexpressions to be evaluated before or after the statement 239expression. A @code{break} or @code{continue} statement inside of 240a statement expression used in @code{while}, @code{do} or @code{for} 241loop or @code{switch} statement condition 242or @code{for} statement init or increment expressions jumps to an 243outer loop or @code{switch} statement if any (otherwise it is an error), 244rather than to the loop or @code{switch} statement in whose condition 245or init or increment expression it appears. 246In any case, as with a function call, the evaluation of a 247statement expression is not interleaved with the evaluation of other 248parts of the containing expression. For example, 249 250@smallexample 251 foo (), ((@{ bar1 (); goto a; 0; @}) + bar2 ()), baz(); 252@end smallexample 253 254@noindent 255calls @code{foo} and @code{bar1} and does not call @code{baz} but 256may or may not call @code{bar2}. If @code{bar2} is called, it is 257called after @code{foo} and before @code{bar1}. 258 259@node Local Labels 260@section Locally Declared Labels 261@cindex local labels 262@cindex macros, local labels 263 264GCC allows you to declare @dfn{local labels} in any nested block 265scope. A local label is just like an ordinary label, but you can 266only reference it (with a @code{goto} statement, or by taking its 267address) within the block in which it is declared. 268 269A local label declaration looks like this: 270 271@smallexample 272__label__ @var{label}; 273@end smallexample 274 275@noindent 276or 277 278@smallexample 279__label__ @var{label1}, @var{label2}, /* @r{@dots{}} */; 280@end smallexample 281 282Local label declarations must come at the beginning of the block, 283before any ordinary declarations or statements. 284 285The label declaration defines the label @emph{name}, but does not define 286the label itself. You must do this in the usual way, with 287@code{@var{label}:}, within the statements of the statement expression. 288 289The local label feature is useful for complex macros. If a macro 290contains nested loops, a @code{goto} can be useful for breaking out of 291them. However, an ordinary label whose scope is the whole function 292cannot be used: if the macro can be expanded several times in one 293function, the label is multiply defined in that function. A 294local label avoids this problem. For example: 295 296@smallexample 297#define SEARCH(value, array, target) \ 298do @{ \ 299 __label__ found; \ 300 typeof (target) _SEARCH_target = (target); \ 301 typeof (*(array)) *_SEARCH_array = (array); \ 302 int i, j; \ 303 int value; \ 304 for (i = 0; i < max; i++) \ 305 for (j = 0; j < max; j++) \ 306 if (_SEARCH_array[i][j] == _SEARCH_target) \ 307 @{ (value) = i; goto found; @} \ 308 (value) = -1; \ 309 found:; \ 310@} while (0) 311@end smallexample 312 313This could also be written using a statement expression: 314 315@smallexample 316#define SEARCH(array, target) \ 317(@{ \ 318 __label__ found; \ 319 typeof (target) _SEARCH_target = (target); \ 320 typeof (*(array)) *_SEARCH_array = (array); \ 321 int i, j; \ 322 int value; \ 323 for (i = 0; i < max; i++) \ 324 for (j = 0; j < max; j++) \ 325 if (_SEARCH_array[i][j] == _SEARCH_target) \ 326 @{ value = i; goto found; @} \ 327 value = -1; \ 328 found: \ 329 value; \ 330@}) 331@end smallexample 332 333Local label declarations also make the labels they declare visible to 334nested functions, if there are any. @xref{Nested Functions}, for details. 335 336@node Labels as Values 337@section Labels as Values 338@cindex labels as values 339@cindex computed gotos 340@cindex goto with computed label 341@cindex address of a label 342 343You can get the address of a label defined in the current function 344(or a containing function) with the unary operator @samp{&&}. The 345value has type @code{void *}. This value is a constant and can be used 346wherever a constant of that type is valid. For example: 347 348@smallexample 349void *ptr; 350/* @r{@dots{}} */ 351ptr = &&foo; 352@end smallexample 353 354To use these values, you need to be able to jump to one. This is done 355with the computed goto statement@footnote{The analogous feature in 356Fortran is called an assigned goto, but that name seems inappropriate in 357C, where one can do more than simply store label addresses in label 358variables.}, @code{goto *@var{exp};}. For example, 359 360@smallexample 361goto *ptr; 362@end smallexample 363 364@noindent 365Any expression of type @code{void *} is allowed. 366 367One way of using these constants is in initializing a static array that 368serves as a jump table: 369 370@smallexample 371static void *array[] = @{ &&foo, &&bar, &&hack @}; 372@end smallexample 373 374@noindent 375Then you can select a label with indexing, like this: 376 377@smallexample 378goto *array[i]; 379@end smallexample 380 381@noindent 382Note that this does not check whether the subscript is in bounds---array 383indexing in C never does that. 384 385Such an array of label values serves a purpose much like that of the 386@code{switch} statement. The @code{switch} statement is cleaner, so 387use that rather than an array unless the problem does not fit a 388@code{switch} statement very well. 389 390Another use of label values is in an interpreter for threaded code. 391The labels within the interpreter function can be stored in the 392threaded code for super-fast dispatching. 393 394You may not use this mechanism to jump to code in a different function. 395If you do that, totally unpredictable things happen. The best way to 396avoid this is to store the label address only in automatic variables and 397never pass it as an argument. 398 399An alternate way to write the above example is 400 401@smallexample 402static const int array[] = @{ &&foo - &&foo, &&bar - &&foo, 403 &&hack - &&foo @}; 404goto *(&&foo + array[i]); 405@end smallexample 406 407@noindent 408This is more friendly to code living in shared libraries, as it reduces 409the number of dynamic relocations that are needed, and by consequence, 410allows the data to be read-only. 411This alternative with label differences is not supported for the AVR target, 412please use the first approach for AVR programs. 413 414The @code{&&foo} expressions for the same label might have different 415values if the containing function is inlined or cloned. If a program 416relies on them being always the same, 417@code{__attribute__((__noinline__,__noclone__))} should be used to 418prevent inlining and cloning. If @code{&&foo} is used in a static 419variable initializer, inlining and cloning is forbidden. 420 421@node Nested Functions 422@section Nested Functions 423@cindex nested functions 424@cindex downward funargs 425@cindex thunks 426 427A @dfn{nested function} is a function defined inside another function. 428Nested functions are supported as an extension in GNU C, but are not 429supported by GNU C++. 430 431The nested function's name is local to the block where it is defined. 432For example, here we define a nested function named @code{square}, and 433call it twice: 434 435@smallexample 436@group 437foo (double a, double b) 438@{ 439 double square (double z) @{ return z * z; @} 440 441 return square (a) + square (b); 442@} 443@end group 444@end smallexample 445 446The nested function can access all the variables of the containing 447function that are visible at the point of its definition. This is 448called @dfn{lexical scoping}. For example, here we show a nested 449function which uses an inherited variable named @code{offset}: 450 451@smallexample 452@group 453bar (int *array, int offset, int size) 454@{ 455 int access (int *array, int index) 456 @{ return array[index + offset]; @} 457 int i; 458 /* @r{@dots{}} */ 459 for (i = 0; i < size; i++) 460 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */ 461@} 462@end group 463@end smallexample 464 465Nested function definitions are permitted within functions in the places 466where variable definitions are allowed; that is, in any block, mixed 467with the other declarations and statements in the block. 468 469It is possible to call the nested function from outside the scope of its 470name by storing its address or passing the address to another function: 471 472@smallexample 473hack (int *array, int size) 474@{ 475 void store (int index, int value) 476 @{ array[index] = value; @} 477 478 intermediate (store, size); 479@} 480@end smallexample 481 482Here, the function @code{intermediate} receives the address of 483@code{store} as an argument. If @code{intermediate} calls @code{store}, 484the arguments given to @code{store} are used to store into @code{array}. 485But this technique works only so long as the containing function 486(@code{hack}, in this example) does not exit. 487 488If you try to call the nested function through its address after the 489containing function exits, all hell breaks loose. If you try 490to call it after a containing scope level exits, and if it refers 491to some of the variables that are no longer in scope, you may be lucky, 492but it's not wise to take the risk. If, however, the nested function 493does not refer to anything that has gone out of scope, you should be 494safe. 495 496GCC implements taking the address of a nested function using a technique 497called @dfn{trampolines}. This technique was described in 498@cite{Lexical Closures for C++} (Thomas M. Breuel, USENIX 499C++ Conference Proceedings, October 17-21, 1988). 500 501A nested function can jump to a label inherited from a containing 502function, provided the label is explicitly declared in the containing 503function (@pxref{Local Labels}). Such a jump returns instantly to the 504containing function, exiting the nested function that did the 505@code{goto} and any intermediate functions as well. Here is an example: 506 507@smallexample 508@group 509bar (int *array, int offset, int size) 510@{ 511 __label__ failure; 512 int access (int *array, int index) 513 @{ 514 if (index > size) 515 goto failure; 516 return array[index + offset]; 517 @} 518 int i; 519 /* @r{@dots{}} */ 520 for (i = 0; i < size; i++) 521 /* @r{@dots{}} */ access (array, i) /* @r{@dots{}} */ 522 /* @r{@dots{}} */ 523 return 0; 524 525 /* @r{Control comes here from @code{access} 526 if it detects an error.} */ 527 failure: 528 return -1; 529@} 530@end group 531@end smallexample 532 533A nested function always has no linkage. Declaring one with 534@code{extern} or @code{static} is erroneous. If you need to declare the nested function 535before its definition, use @code{auto} (which is otherwise meaningless 536for function declarations). 537 538@smallexample 539bar (int *array, int offset, int size) 540@{ 541 __label__ failure; 542 auto int access (int *, int); 543 /* @r{@dots{}} */ 544 int access (int *array, int index) 545 @{ 546 if (index > size) 547 goto failure; 548 return array[index + offset]; 549 @} 550 /* @r{@dots{}} */ 551@} 552@end smallexample 553 554@node Nonlocal Gotos 555@section Nonlocal Gotos 556@cindex nonlocal gotos 557 558GCC provides the built-in functions @code{__builtin_setjmp} and 559@code{__builtin_longjmp} which are similar to, but not interchangeable 560with, the C library functions @code{setjmp} and @code{longjmp}. 561The built-in versions are used internally by GCC's libraries 562to implement exception handling on some targets. You should use the 563standard C library functions declared in @code{<setjmp.h>} in user code 564instead of the builtins. 565 566The built-in versions of these functions use GCC's normal 567mechanisms to save and restore registers using the stack on function 568entry and exit. The jump buffer argument @var{buf} holds only the 569information needed to restore the stack frame, rather than the entire 570set of saved register values. 571 572An important caveat is that GCC arranges to save and restore only 573those registers known to the specific architecture variant being 574compiled for. This can make @code{__builtin_setjmp} and 575@code{__builtin_longjmp} more efficient than their library 576counterparts in some cases, but it can also cause incorrect and 577mysterious behavior when mixing with code that uses the full register 578set. 579 580You should declare the jump buffer argument @var{buf} to the 581built-in functions as: 582 583@smallexample 584#include <stdint.h> 585intptr_t @var{buf}[5]; 586@end smallexample 587 588@deftypefn {Built-in Function} {int} __builtin_setjmp (intptr_t *@var{buf}) 589This function saves the current stack context in @var{buf}. 590@code{__builtin_setjmp} returns 0 when returning directly, 591and 1 when returning from @code{__builtin_longjmp} using the same 592@var{buf}. 593@end deftypefn 594 595@deftypefn {Built-in Function} {void} __builtin_longjmp (intptr_t *@var{buf}, int @var{val}) 596This function restores the stack context in @var{buf}, 597saved by a previous call to @code{__builtin_setjmp}. After 598@code{__builtin_longjmp} is finished, the program resumes execution as 599if the matching @code{__builtin_setjmp} returns the value @var{val}, 600which must be 1. 601 602Because @code{__builtin_longjmp} depends on the function return 603mechanism to restore the stack context, it cannot be called 604from the same function calling @code{__builtin_setjmp} to 605initialize @var{buf}. It can only be called from a function called 606(directly or indirectly) from the function calling @code{__builtin_setjmp}. 607@end deftypefn 608 609@node Constructing Calls 610@section Constructing Function Calls 611@cindex constructing calls 612@cindex forwarding calls 613 614Using the built-in functions described below, you can record 615the arguments a function received, and call another function 616with the same arguments, without knowing the number or types 617of the arguments. 618 619You can also record the return value of that function call, 620and later return that value, without knowing what data type 621the function tried to return (as long as your caller expects 622that data type). 623 624However, these built-in functions may interact badly with some 625sophisticated features or other extensions of the language. It 626is, therefore, not recommended to use them outside very simple 627functions acting as mere forwarders for their arguments. 628 629@deftypefn {Built-in Function} {void *} __builtin_apply_args () 630This built-in function returns a pointer to data 631describing how to perform a call with the same arguments as are passed 632to the current function. 633 634The function saves the arg pointer register, structure value address, 635and all registers that might be used to pass arguments to a function 636into a block of memory allocated on the stack. Then it returns the 637address of that block. 638@end deftypefn 639 640@deftypefn {Built-in Function} {void *} __builtin_apply (void (*@var{function})(), void *@var{arguments}, size_t @var{size}) 641This built-in function invokes @var{function} 642with a copy of the parameters described by @var{arguments} 643and @var{size}. 644 645The value of @var{arguments} should be the value returned by 646@code{__builtin_apply_args}. The argument @var{size} specifies the size 647of the stack argument data, in bytes. 648 649This function returns a pointer to data describing 650how to return whatever value is returned by @var{function}. The data 651is saved in a block of memory allocated on the stack. 652 653It is not always simple to compute the proper value for @var{size}. The 654value is used by @code{__builtin_apply} to compute the amount of data 655that should be pushed on the stack and copied from the incoming argument 656area. 657@end deftypefn 658 659@deftypefn {Built-in Function} {void} __builtin_return (void *@var{result}) 660This built-in function returns the value described by @var{result} from 661the containing function. You should specify, for @var{result}, a value 662returned by @code{__builtin_apply}. 663@end deftypefn 664 665@deftypefn {Built-in Function} {} __builtin_va_arg_pack () 666This built-in function represents all anonymous arguments of an inline 667function. It can be used only in inline functions that are always 668inlined, never compiled as a separate function, such as those using 669@code{__attribute__ ((__always_inline__))} or 670@code{__attribute__ ((__gnu_inline__))} extern inline functions. 671It must be only passed as last argument to some other function 672with variable arguments. This is useful for writing small wrapper 673inlines for variable argument functions, when using preprocessor 674macros is undesirable. For example: 675@smallexample 676extern int myprintf (FILE *f, const char *format, ...); 677extern inline __attribute__ ((__gnu_inline__)) int 678myprintf (FILE *f, const char *format, ...) 679@{ 680 int r = fprintf (f, "myprintf: "); 681 if (r < 0) 682 return r; 683 int s = fprintf (f, format, __builtin_va_arg_pack ()); 684 if (s < 0) 685 return s; 686 return r + s; 687@} 688@end smallexample 689@end deftypefn 690 691@deftypefn {Built-in Function} {size_t} __builtin_va_arg_pack_len () 692This built-in function returns the number of anonymous arguments of 693an inline function. It can be used only in inline functions that 694are always inlined, never compiled as a separate function, such 695as those using @code{__attribute__ ((__always_inline__))} or 696@code{__attribute__ ((__gnu_inline__))} extern inline functions. 697For example following does link- or run-time checking of open 698arguments for optimized code: 699@smallexample 700#ifdef __OPTIMIZE__ 701extern inline __attribute__((__gnu_inline__)) int 702myopen (const char *path, int oflag, ...) 703@{ 704 if (__builtin_va_arg_pack_len () > 1) 705 warn_open_too_many_arguments (); 706 707 if (__builtin_constant_p (oflag)) 708 @{ 709 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1) 710 @{ 711 warn_open_missing_mode (); 712 return __open_2 (path, oflag); 713 @} 714 return open (path, oflag, __builtin_va_arg_pack ()); 715 @} 716 717 if (__builtin_va_arg_pack_len () < 1) 718 return __open_2 (path, oflag); 719 720 return open (path, oflag, __builtin_va_arg_pack ()); 721@} 722#endif 723@end smallexample 724@end deftypefn 725 726@node Typeof 727@section Referring to a Type with @code{typeof} 728@findex typeof 729@findex sizeof 730@cindex macros, types of arguments 731 732Another way to refer to the type of an expression is with @code{typeof}. 733The syntax of using of this keyword looks like @code{sizeof}, but the 734construct acts semantically like a type name defined with @code{typedef}. 735 736There are two ways of writing the argument to @code{typeof}: with an 737expression or with a type. Here is an example with an expression: 738 739@smallexample 740typeof (x[0](1)) 741@end smallexample 742 743@noindent 744This assumes that @code{x} is an array of pointers to functions; 745the type described is that of the values of the functions. 746 747Here is an example with a typename as the argument: 748 749@smallexample 750typeof (int *) 751@end smallexample 752 753@noindent 754Here the type described is that of pointers to @code{int}. 755 756If you are writing a header file that must work when included in ISO C 757programs, write @code{__typeof__} instead of @code{typeof}. 758@xref{Alternate Keywords}. 759 760A @code{typeof} construct can be used anywhere a typedef name can be 761used. For example, you can use it in a declaration, in a cast, or inside 762of @code{sizeof} or @code{typeof}. 763 764The operand of @code{typeof} is evaluated for its side effects if and 765only if it is an expression of variably modified type or the name of 766such a type. 767 768@code{typeof} is often useful in conjunction with 769statement expressions (@pxref{Statement Exprs}). 770Here is how the two together can 771be used to define a safe ``maximum'' macro which operates on any 772arithmetic type and evaluates each of its arguments exactly once: 773 774@smallexample 775#define max(a,b) \ 776 (@{ typeof (a) _a = (a); \ 777 typeof (b) _b = (b); \ 778 _a > _b ? _a : _b; @}) 779@end smallexample 780 781@cindex underscores in variables in macros 782@cindex @samp{_} in variables in macros 783@cindex local variables in macros 784@cindex variables, local, in macros 785@cindex macros, local variables in 786 787The reason for using names that start with underscores for the local 788variables is to avoid conflicts with variable names that occur within the 789expressions that are substituted for @code{a} and @code{b}. Eventually we 790hope to design a new form of declaration syntax that allows you to declare 791variables whose scopes start only after their initializers; this will be a 792more reliable way to prevent such conflicts. 793 794@noindent 795Some more examples of the use of @code{typeof}: 796 797@itemize @bullet 798@item 799This declares @code{y} with the type of what @code{x} points to. 800 801@smallexample 802typeof (*x) y; 803@end smallexample 804 805@item 806This declares @code{y} as an array of such values. 807 808@smallexample 809typeof (*x) y[4]; 810@end smallexample 811 812@item 813This declares @code{y} as an array of pointers to characters: 814 815@smallexample 816typeof (typeof (char *)[4]) y; 817@end smallexample 818 819@noindent 820It is equivalent to the following traditional C declaration: 821 822@smallexample 823char *y[4]; 824@end smallexample 825 826To see the meaning of the declaration using @code{typeof}, and why it 827might be a useful way to write, rewrite it with these macros: 828 829@smallexample 830#define pointer(T) typeof(T *) 831#define array(T, N) typeof(T [N]) 832@end smallexample 833 834@noindent 835Now the declaration can be rewritten this way: 836 837@smallexample 838array (pointer (char), 4) y; 839@end smallexample 840 841@noindent 842Thus, @code{array (pointer (char), 4)} is the type of arrays of 4 843pointers to @code{char}. 844@end itemize 845 846In GNU C, but not GNU C++, you may also declare the type of a variable 847as @code{__auto_type}. In that case, the declaration must declare 848only one variable, whose declarator must just be an identifier, the 849declaration must be initialized, and the type of the variable is 850determined by the initializer; the name of the variable is not in 851scope until after the initializer. (In C++, you should use C++11 852@code{auto} for this purpose.) Using @code{__auto_type}, the 853``maximum'' macro above could be written as: 854 855@smallexample 856#define max(a,b) \ 857 (@{ __auto_type _a = (a); \ 858 __auto_type _b = (b); \ 859 _a > _b ? _a : _b; @}) 860@end smallexample 861 862Using @code{__auto_type} instead of @code{typeof} has two advantages: 863 864@itemize @bullet 865@item Each argument to the macro appears only once in the expansion of 866the macro. This prevents the size of the macro expansion growing 867exponentially when calls to such macros are nested inside arguments of 868such macros. 869 870@item If the argument to the macro has variably modified type, it is 871evaluated only once when using @code{__auto_type}, but twice if 872@code{typeof} is used. 873@end itemize 874 875@node Conditionals 876@section Conditionals with Omitted Operands 877@cindex conditional expressions, extensions 878@cindex omitted middle-operands 879@cindex middle-operands, omitted 880@cindex extensions, @code{?:} 881@cindex @code{?:} extensions 882 883The middle operand in a conditional expression may be omitted. Then 884if the first operand is nonzero, its value is the value of the conditional 885expression. 886 887Therefore, the expression 888 889@smallexample 890x ? : y 891@end smallexample 892 893@noindent 894has the value of @code{x} if that is nonzero; otherwise, the value of 895@code{y}. 896 897This example is perfectly equivalent to 898 899@smallexample 900x ? x : y 901@end smallexample 902 903@cindex side effect in @code{?:} 904@cindex @code{?:} side effect 905@noindent 906In this simple case, the ability to omit the middle operand is not 907especially useful. When it becomes useful is when the first operand does, 908or may (if it is a macro argument), contain a side effect. Then repeating 909the operand in the middle would perform the side effect twice. Omitting 910the middle operand uses the value already computed without the undesirable 911effects of recomputing it. 912 913@node __int128 914@section 128-bit Integers 915@cindex @code{__int128} data types 916 917As an extension the integer scalar type @code{__int128} is supported for 918targets which have an integer mode wide enough to hold 128 bits. 919Simply write @code{__int128} for a signed 128-bit integer, or 920@code{unsigned __int128} for an unsigned 128-bit integer. There is no 921support in GCC for expressing an integer constant of type @code{__int128} 922for targets with @code{long long} integer less than 128 bits wide. 923 924@node Long Long 925@section Double-Word Integers 926@cindex @code{long long} data types 927@cindex double-word arithmetic 928@cindex multiprecision arithmetic 929@cindex @code{LL} integer suffix 930@cindex @code{ULL} integer suffix 931 932ISO C99 and ISO C++11 support data types for integers that are at least 93364 bits wide, and as an extension GCC supports them in C90 and C++98 modes. 934Simply write @code{long long int} for a signed integer, or 935@code{unsigned long long int} for an unsigned integer. To make an 936integer constant of type @code{long long int}, add the suffix @samp{LL} 937to the integer. To make an integer constant of type @code{unsigned long 938long int}, add the suffix @samp{ULL} to the integer. 939 940You can use these types in arithmetic like any other integer types. 941Addition, subtraction, and bitwise boolean operations on these types 942are open-coded on all types of machines. Multiplication is open-coded 943if the machine supports a fullword-to-doubleword widening multiply 944instruction. Division and shifts are open-coded only on machines that 945provide special support. The operations that are not open-coded use 946special library routines that come with GCC@. 947 948There may be pitfalls when you use @code{long long} types for function 949arguments without function prototypes. If a function 950expects type @code{int} for its argument, and you pass a value of type 951@code{long long int}, confusion results because the caller and the 952subroutine disagree about the number of bytes for the argument. 953Likewise, if the function expects @code{long long int} and you pass 954@code{int}. The best way to avoid such problems is to use prototypes. 955 956@node Complex 957@section Complex Numbers 958@cindex complex numbers 959@cindex @code{_Complex} keyword 960@cindex @code{__complex__} keyword 961 962ISO C99 supports complex floating data types, and as an extension GCC 963supports them in C90 mode and in C++. GCC also supports complex integer data 964types which are not part of ISO C99. You can declare complex types 965using the keyword @code{_Complex}. As an extension, the older GNU 966keyword @code{__complex__} is also supported. 967 968For example, @samp{_Complex double x;} declares @code{x} as a 969variable whose real part and imaginary part are both of type 970@code{double}. @samp{_Complex short int y;} declares @code{y} to 971have real and imaginary parts of type @code{short int}; this is not 972likely to be useful, but it shows that the set of complex types is 973complete. 974 975To write a constant with a complex data type, use the suffix @samp{i} or 976@samp{j} (either one; they are equivalent). For example, @code{2.5fi} 977has type @code{_Complex float} and @code{3i} has type 978@code{_Complex int}. Such a constant always has a pure imaginary 979value, but you can form any complex value you like by adding one to a 980real constant. This is a GNU extension; if you have an ISO C99 981conforming C library (such as the GNU C Library), and want to construct complex 982constants of floating type, you should include @code{<complex.h>} and 983use the macros @code{I} or @code{_Complex_I} instead. 984 985The ISO C++14 library also defines the @samp{i} suffix, so C++14 code 986that includes the @samp{<complex>} header cannot use @samp{i} for the 987GNU extension. The @samp{j} suffix still has the GNU meaning. 988 989@cindex @code{__real__} keyword 990@cindex @code{__imag__} keyword 991To extract the real part of a complex-valued expression @var{exp}, write 992@code{__real__ @var{exp}}. Likewise, use @code{__imag__} to 993extract the imaginary part. This is a GNU extension; for values of 994floating type, you should use the ISO C99 functions @code{crealf}, 995@code{creal}, @code{creall}, @code{cimagf}, @code{cimag} and 996@code{cimagl}, declared in @code{<complex.h>} and also provided as 997built-in functions by GCC@. 998 999@cindex complex conjugation 1000The operator @samp{~} performs complex conjugation when used on a value 1001with a complex type. This is a GNU extension; for values of 1002floating type, you should use the ISO C99 functions @code{conjf}, 1003@code{conj} and @code{conjl}, declared in @code{<complex.h>} and also 1004provided as built-in functions by GCC@. 1005 1006GCC can allocate complex automatic variables in a noncontiguous 1007fashion; it's even possible for the real part to be in a register while 1008the imaginary part is on the stack (or vice versa). Only the DWARF 1009debug info format can represent this, so use of DWARF is recommended. 1010If you are using the stabs debug info format, GCC describes a noncontiguous 1011complex variable as if it were two separate variables of noncomplex type. 1012If the variable's actual name is @code{foo}, the two fictitious 1013variables are named @code{foo$real} and @code{foo$imag}. You can 1014examine and set these two fictitious variables with your debugger. 1015 1016@node Floating Types 1017@section Additional Floating Types 1018@cindex additional floating types 1019@cindex @code{_Float@var{n}} data types 1020@cindex @code{_Float@var{n}x} data types 1021@cindex @code{__float80} data type 1022@cindex @code{__float128} data type 1023@cindex @code{__ibm128} data type 1024@cindex @code{w} floating point suffix 1025@cindex @code{q} floating point suffix 1026@cindex @code{W} floating point suffix 1027@cindex @code{Q} floating point suffix 1028 1029ISO/IEC TS 18661-3:2015 defines C support for additional floating 1030types @code{_Float@var{n}} and @code{_Float@var{n}x}, and GCC supports 1031these type names; the set of types supported depends on the target 1032architecture. These types are not supported when compiling C++. 1033Constants with these types use suffixes @code{f@var{n}} or 1034@code{F@var{n}} and @code{f@var{n}x} or @code{F@var{n}x}. These type 1035names can be used together with @code{_Complex} to declare complex 1036types. 1037 1038As an extension, GNU C and GNU C++ support additional floating 1039types, which are not supported by all targets. 1040@itemize @bullet 1041@item @code{__float128} is available on i386, x86_64, IA-64, and 1042hppa HP-UX, as well as on PowerPC GNU/Linux targets that enable 1043the vector scalar (VSX) instruction set. @code{__float128} supports 1044the 128-bit floating type. On i386, x86_64, PowerPC, and IA-64 1045other than HP-UX, @code{__float128} is an alias for @code{_Float128}. 1046On hppa and IA-64 HP-UX, @code{__float128} is an alias for @code{long 1047double}. 1048 1049@item @code{__float80} is available on the i386, x86_64, and IA-64 1050targets, and supports the 80-bit (@code{XFmode}) floating type. It is 1051an alias for the type name @code{_Float64x} on these targets. 1052 1053@item @code{__ibm128} is available on PowerPC targets, and provides 1054access to the IBM extended double format which is the current format 1055used for @code{long double}. When @code{long double} transitions to 1056@code{__float128} on PowerPC in the future, @code{__ibm128} will remain 1057for use in conversions between the two types. 1058@end itemize 1059 1060Support for these additional types includes the arithmetic operators: 1061add, subtract, multiply, divide; unary arithmetic operators; 1062relational operators; equality operators; and conversions to and from 1063integer and other floating types. Use a suffix @samp{w} or @samp{W} 1064in a literal constant of type @code{__float80} or type 1065@code{__ibm128}. Use a suffix @samp{q} or @samp{Q} for @code{_float128}. 1066 1067In order to use @code{_Float128}, @code{__float128}, and @code{__ibm128} 1068on PowerPC Linux systems, you must use the @option{-mfloat128} option. It is 1069expected in future versions of GCC that @code{_Float128} and @code{__float128} 1070will be enabled automatically. 1071 1072The @code{_Float128} type is supported on all systems where 1073@code{__float128} is supported or where @code{long double} has the 1074IEEE binary128 format. The @code{_Float64x} type is supported on all 1075systems where @code{__float128} is supported. The @code{_Float32} 1076type is supported on all systems supporting IEEE binary32; the 1077@code{_Float64} and @code{_Float32x} types are supported on all systems 1078supporting IEEE binary64. The @code{_Float16} type is supported on AArch64 1079systems by default, and on ARM systems when the IEEE format for 16-bit 1080floating-point types is selected with @option{-mfp16-format=ieee}. 1081GCC does not currently support @code{_Float128x} on any systems. 1082 1083On the i386, x86_64, IA-64, and HP-UX targets, you can declare complex 1084types using the corresponding internal complex type, @code{XCmode} for 1085@code{__float80} type and @code{TCmode} for @code{__float128} type: 1086 1087@smallexample 1088typedef _Complex float __attribute__((mode(TC))) _Complex128; 1089typedef _Complex float __attribute__((mode(XC))) _Complex80; 1090@end smallexample 1091 1092On the PowerPC Linux VSX targets, you can declare complex types using 1093the corresponding internal complex type, @code{KCmode} for 1094@code{__float128} type and @code{ICmode} for @code{__ibm128} type: 1095 1096@smallexample 1097typedef _Complex float __attribute__((mode(KC))) _Complex_float128; 1098typedef _Complex float __attribute__((mode(IC))) _Complex_ibm128; 1099@end smallexample 1100 1101@node Half-Precision 1102@section Half-Precision Floating Point 1103@cindex half-precision floating point 1104@cindex @code{__fp16} data type 1105 1106On ARM and AArch64 targets, GCC supports half-precision (16-bit) floating 1107point via the @code{__fp16} type defined in the ARM C Language Extensions. 1108On ARM systems, you must enable this type explicitly with the 1109@option{-mfp16-format} command-line option in order to use it. 1110 1111ARM targets support two incompatible representations for half-precision 1112floating-point values. You must choose one of the representations and 1113use it consistently in your program. 1114 1115Specifying @option{-mfp16-format=ieee} selects the IEEE 754-2008 format. 1116This format can represent normalized values in the range of @math{2^{-14}} to 65504. 1117There are 11 bits of significand precision, approximately 3 1118decimal digits. 1119 1120Specifying @option{-mfp16-format=alternative} selects the ARM 1121alternative format. This representation is similar to the IEEE 1122format, but does not support infinities or NaNs. Instead, the range 1123of exponents is extended, so that this format can represent normalized 1124values in the range of @math{2^{-14}} to 131008. 1125 1126The GCC port for AArch64 only supports the IEEE 754-2008 format, and does 1127not require use of the @option{-mfp16-format} command-line option. 1128 1129The @code{__fp16} type may only be used as an argument to intrinsics defined 1130in @code{<arm_fp16.h>}, or as a storage format. For purposes of 1131arithmetic and other operations, @code{__fp16} values in C or C++ 1132expressions are automatically promoted to @code{float}. 1133 1134The ARM target provides hardware support for conversions between 1135@code{__fp16} and @code{float} values 1136as an extension to VFP and NEON (Advanced SIMD), and from ARMv8-A provides 1137hardware support for conversions between @code{__fp16} and @code{double} 1138values. GCC generates code using these hardware instructions if you 1139compile with options to select an FPU that provides them; 1140for example, @option{-mfpu=neon-fp16 -mfloat-abi=softfp}, 1141in addition to the @option{-mfp16-format} option to select 1142a half-precision format. 1143 1144Language-level support for the @code{__fp16} data type is 1145independent of whether GCC generates code using hardware floating-point 1146instructions. In cases where hardware support is not specified, GCC 1147implements conversions between @code{__fp16} and other types as library 1148calls. 1149 1150It is recommended that portable code use the @code{_Float16} type defined 1151by ISO/IEC TS 18661-3:2015. @xref{Floating Types}. 1152 1153@node Decimal Float 1154@section Decimal Floating Types 1155@cindex decimal floating types 1156@cindex @code{_Decimal32} data type 1157@cindex @code{_Decimal64} data type 1158@cindex @code{_Decimal128} data type 1159@cindex @code{df} integer suffix 1160@cindex @code{dd} integer suffix 1161@cindex @code{dl} integer suffix 1162@cindex @code{DF} integer suffix 1163@cindex @code{DD} integer suffix 1164@cindex @code{DL} integer suffix 1165 1166As an extension, GNU C supports decimal floating types as 1167defined in the N1312 draft of ISO/IEC WDTR24732. Support for decimal 1168floating types in GCC will evolve as the draft technical report changes. 1169Calling conventions for any target might also change. Not all targets 1170support decimal floating types. 1171 1172The decimal floating types are @code{_Decimal32}, @code{_Decimal64}, and 1173@code{_Decimal128}. They use a radix of ten, unlike the floating types 1174@code{float}, @code{double}, and @code{long double} whose radix is not 1175specified by the C standard but is usually two. 1176 1177Support for decimal floating types includes the arithmetic operators 1178add, subtract, multiply, divide; unary arithmetic operators; 1179relational operators; equality operators; and conversions to and from 1180integer and other floating types. Use a suffix @samp{df} or 1181@samp{DF} in a literal constant of type @code{_Decimal32}, @samp{dd} 1182or @samp{DD} for @code{_Decimal64}, and @samp{dl} or @samp{DL} for 1183@code{_Decimal128}. 1184 1185GCC support of decimal float as specified by the draft technical report 1186is incomplete: 1187 1188@itemize @bullet 1189@item 1190When the value of a decimal floating type cannot be represented in the 1191integer type to which it is being converted, the result is undefined 1192rather than the result value specified by the draft technical report. 1193 1194@item 1195GCC does not provide the C library functionality associated with 1196@file{math.h}, @file{fenv.h}, @file{stdio.h}, @file{stdlib.h}, and 1197@file{wchar.h}, which must come from a separate C library implementation. 1198Because of this the GNU C compiler does not define macro 1199@code{__STDC_DEC_FP__} to indicate that the implementation conforms to 1200the technical report. 1201@end itemize 1202 1203Types @code{_Decimal32}, @code{_Decimal64}, and @code{_Decimal128} 1204are supported by the DWARF debug information format. 1205 1206@node Hex Floats 1207@section Hex Floats 1208@cindex hex floats 1209 1210ISO C99 and ISO C++17 support floating-point numbers written not only in 1211the usual decimal notation, such as @code{1.55e1}, but also numbers such as 1212@code{0x1.fp3} written in hexadecimal format. As a GNU extension, GCC 1213supports this in C90 mode (except in some cases when strictly 1214conforming) and in C++98, C++11 and C++14 modes. In that format the 1215@samp{0x} hex introducer and the @samp{p} or @samp{P} exponent field are 1216mandatory. The exponent is a decimal number that indicates the power of 12172 by which the significant part is multiplied. Thus @samp{0x1.f} is 1218@tex 1219$1 {15\over16}$, 1220@end tex 1221@ifnottex 12221 15/16, 1223@end ifnottex 1224@samp{p3} multiplies it by 8, and the value of @code{0x1.fp3} 1225is the same as @code{1.55e1}. 1226 1227Unlike for floating-point numbers in the decimal notation the exponent 1228is always required in the hexadecimal notation. Otherwise the compiler 1229would not be able to resolve the ambiguity of, e.g., @code{0x1.f}. This 1230could mean @code{1.0f} or @code{1.9375} since @samp{f} is also the 1231extension for floating-point constants of type @code{float}. 1232 1233@node Fixed-Point 1234@section Fixed-Point Types 1235@cindex fixed-point types 1236@cindex @code{_Fract} data type 1237@cindex @code{_Accum} data type 1238@cindex @code{_Sat} data type 1239@cindex @code{hr} fixed-suffix 1240@cindex @code{r} fixed-suffix 1241@cindex @code{lr} fixed-suffix 1242@cindex @code{llr} fixed-suffix 1243@cindex @code{uhr} fixed-suffix 1244@cindex @code{ur} fixed-suffix 1245@cindex @code{ulr} fixed-suffix 1246@cindex @code{ullr} fixed-suffix 1247@cindex @code{hk} fixed-suffix 1248@cindex @code{k} fixed-suffix 1249@cindex @code{lk} fixed-suffix 1250@cindex @code{llk} fixed-suffix 1251@cindex @code{uhk} fixed-suffix 1252@cindex @code{uk} fixed-suffix 1253@cindex @code{ulk} fixed-suffix 1254@cindex @code{ullk} fixed-suffix 1255@cindex @code{HR} fixed-suffix 1256@cindex @code{R} fixed-suffix 1257@cindex @code{LR} fixed-suffix 1258@cindex @code{LLR} fixed-suffix 1259@cindex @code{UHR} fixed-suffix 1260@cindex @code{UR} fixed-suffix 1261@cindex @code{ULR} fixed-suffix 1262@cindex @code{ULLR} fixed-suffix 1263@cindex @code{HK} fixed-suffix 1264@cindex @code{K} fixed-suffix 1265@cindex @code{LK} fixed-suffix 1266@cindex @code{LLK} fixed-suffix 1267@cindex @code{UHK} fixed-suffix 1268@cindex @code{UK} fixed-suffix 1269@cindex @code{ULK} fixed-suffix 1270@cindex @code{ULLK} fixed-suffix 1271 1272As an extension, GNU C supports fixed-point types as 1273defined in the N1169 draft of ISO/IEC DTR 18037. Support for fixed-point 1274types in GCC will evolve as the draft technical report changes. 1275Calling conventions for any target might also change. Not all targets 1276support fixed-point types. 1277 1278The fixed-point types are 1279@code{short _Fract}, 1280@code{_Fract}, 1281@code{long _Fract}, 1282@code{long long _Fract}, 1283@code{unsigned short _Fract}, 1284@code{unsigned _Fract}, 1285@code{unsigned long _Fract}, 1286@code{unsigned long long _Fract}, 1287@code{_Sat short _Fract}, 1288@code{_Sat _Fract}, 1289@code{_Sat long _Fract}, 1290@code{_Sat long long _Fract}, 1291@code{_Sat unsigned short _Fract}, 1292@code{_Sat unsigned _Fract}, 1293@code{_Sat unsigned long _Fract}, 1294@code{_Sat unsigned long long _Fract}, 1295@code{short _Accum}, 1296@code{_Accum}, 1297@code{long _Accum}, 1298@code{long long _Accum}, 1299@code{unsigned short _Accum}, 1300@code{unsigned _Accum}, 1301@code{unsigned long _Accum}, 1302@code{unsigned long long _Accum}, 1303@code{_Sat short _Accum}, 1304@code{_Sat _Accum}, 1305@code{_Sat long _Accum}, 1306@code{_Sat long long _Accum}, 1307@code{_Sat unsigned short _Accum}, 1308@code{_Sat unsigned _Accum}, 1309@code{_Sat unsigned long _Accum}, 1310@code{_Sat unsigned long long _Accum}. 1311 1312Fixed-point data values contain fractional and optional integral parts. 1313The format of fixed-point data varies and depends on the target machine. 1314 1315Support for fixed-point types includes: 1316@itemize @bullet 1317@item 1318prefix and postfix increment and decrement operators (@code{++}, @code{--}) 1319@item 1320unary arithmetic operators (@code{+}, @code{-}, @code{!}) 1321@item 1322binary arithmetic operators (@code{+}, @code{-}, @code{*}, @code{/}) 1323@item 1324binary shift operators (@code{<<}, @code{>>}) 1325@item 1326relational operators (@code{<}, @code{<=}, @code{>=}, @code{>}) 1327@item 1328equality operators (@code{==}, @code{!=}) 1329@item 1330assignment operators (@code{+=}, @code{-=}, @code{*=}, @code{/=}, 1331@code{<<=}, @code{>>=}) 1332@item 1333conversions to and from integer, floating-point, or fixed-point types 1334@end itemize 1335 1336Use a suffix in a fixed-point literal constant: 1337@itemize 1338@item @samp{hr} or @samp{HR} for @code{short _Fract} and 1339@code{_Sat short _Fract} 1340@item @samp{r} or @samp{R} for @code{_Fract} and @code{_Sat _Fract} 1341@item @samp{lr} or @samp{LR} for @code{long _Fract} and 1342@code{_Sat long _Fract} 1343@item @samp{llr} or @samp{LLR} for @code{long long _Fract} and 1344@code{_Sat long long _Fract} 1345@item @samp{uhr} or @samp{UHR} for @code{unsigned short _Fract} and 1346@code{_Sat unsigned short _Fract} 1347@item @samp{ur} or @samp{UR} for @code{unsigned _Fract} and 1348@code{_Sat unsigned _Fract} 1349@item @samp{ulr} or @samp{ULR} for @code{unsigned long _Fract} and 1350@code{_Sat unsigned long _Fract} 1351@item @samp{ullr} or @samp{ULLR} for @code{unsigned long long _Fract} 1352and @code{_Sat unsigned long long _Fract} 1353@item @samp{hk} or @samp{HK} for @code{short _Accum} and 1354@code{_Sat short _Accum} 1355@item @samp{k} or @samp{K} for @code{_Accum} and @code{_Sat _Accum} 1356@item @samp{lk} or @samp{LK} for @code{long _Accum} and 1357@code{_Sat long _Accum} 1358@item @samp{llk} or @samp{LLK} for @code{long long _Accum} and 1359@code{_Sat long long _Accum} 1360@item @samp{uhk} or @samp{UHK} for @code{unsigned short _Accum} and 1361@code{_Sat unsigned short _Accum} 1362@item @samp{uk} or @samp{UK} for @code{unsigned _Accum} and 1363@code{_Sat unsigned _Accum} 1364@item @samp{ulk} or @samp{ULK} for @code{unsigned long _Accum} and 1365@code{_Sat unsigned long _Accum} 1366@item @samp{ullk} or @samp{ULLK} for @code{unsigned long long _Accum} 1367and @code{_Sat unsigned long long _Accum} 1368@end itemize 1369 1370GCC support of fixed-point types as specified by the draft technical report 1371is incomplete: 1372 1373@itemize @bullet 1374@item 1375Pragmas to control overflow and rounding behaviors are not implemented. 1376@end itemize 1377 1378Fixed-point types are supported by the DWARF debug information format. 1379 1380@node Named Address Spaces 1381@section Named Address Spaces 1382@cindex Named Address Spaces 1383 1384As an extension, GNU C supports named address spaces as 1385defined in the N1275 draft of ISO/IEC DTR 18037. Support for named 1386address spaces in GCC will evolve as the draft technical report 1387changes. Calling conventions for any target might also change. At 1388present, only the AVR, M32C, RL78, and x86 targets support 1389address spaces other than the generic address space. 1390 1391Address space identifiers may be used exactly like any other C type 1392qualifier (e.g., @code{const} or @code{volatile}). See the N1275 1393document for more details. 1394 1395@anchor{AVR Named Address Spaces} 1396@subsection AVR Named Address Spaces 1397 1398On the AVR target, there are several address spaces that can be used 1399in order to put read-only data into the flash memory and access that 1400data by means of the special instructions @code{LPM} or @code{ELPM} 1401needed to read from flash. 1402 1403Devices belonging to @code{avrtiny} and @code{avrxmega3} can access 1404flash memory by means of @code{LD*} instructions because the flash 1405memory is mapped into the RAM address space. There is @emph{no need} 1406for language extensions like @code{__flash} or attribute 1407@ref{AVR Variable Attributes,,@code{progmem}}. 1408The default linker description files for these devices cater for that 1409feature and @code{.rodata} stays in flash: The compiler just generates 1410@code{LD*} instructions, and the linker script adds core specific 1411offsets to all @code{.rodata} symbols: @code{0x4000} in the case of 1412@code{avrtiny} and @code{0x8000} in the case of @code{avrxmega3}. 1413See @ref{AVR Options} for a list of respective devices. 1414 1415For devices not in @code{avrtiny} or @code{avrxmega3}, 1416any data including read-only data is located in RAM (the generic 1417address space) because flash memory is not visible in the RAM address 1418space. In order to locate read-only data in flash memory @emph{and} 1419to generate the right instructions to access this data without 1420using (inline) assembler code, special address spaces are needed. 1421 1422@table @code 1423@item __flash 1424@cindex @code{__flash} AVR Named Address Spaces 1425The @code{__flash} qualifier locates data in the 1426@code{.progmem.data} section. Data is read using the @code{LPM} 1427instruction. Pointers to this address space are 16 bits wide. 1428 1429@item __flash1 1430@itemx __flash2 1431@itemx __flash3 1432@itemx __flash4 1433@itemx __flash5 1434@cindex @code{__flash1} AVR Named Address Spaces 1435@cindex @code{__flash2} AVR Named Address Spaces 1436@cindex @code{__flash3} AVR Named Address Spaces 1437@cindex @code{__flash4} AVR Named Address Spaces 1438@cindex @code{__flash5} AVR Named Address Spaces 1439These are 16-bit address spaces locating data in section 1440@code{.progmem@var{N}.data} where @var{N} refers to 1441address space @code{__flash@var{N}}. 1442The compiler sets the @code{RAMPZ} segment register appropriately 1443before reading data by means of the @code{ELPM} instruction. 1444 1445@item __memx 1446@cindex @code{__memx} AVR Named Address Spaces 1447This is a 24-bit address space that linearizes flash and RAM: 1448If the high bit of the address is set, data is read from 1449RAM using the lower two bytes as RAM address. 1450If the high bit of the address is clear, data is read from flash 1451with @code{RAMPZ} set according to the high byte of the address. 1452@xref{AVR Built-in Functions,,@code{__builtin_avr_flash_segment}}. 1453 1454Objects in this address space are located in @code{.progmemx.data}. 1455@end table 1456 1457@b{Example} 1458 1459@smallexample 1460char my_read (const __flash char ** p) 1461@{ 1462 /* p is a pointer to RAM that points to a pointer to flash. 1463 The first indirection of p reads that flash pointer 1464 from RAM and the second indirection reads a char from this 1465 flash address. */ 1466 1467 return **p; 1468@} 1469 1470/* Locate array[] in flash memory */ 1471const __flash int array[] = @{ 3, 5, 7, 11, 13, 17, 19 @}; 1472 1473int i = 1; 1474 1475int main (void) 1476@{ 1477 /* Return 17 by reading from flash memory */ 1478 return array[array[i]]; 1479@} 1480@end smallexample 1481 1482@noindent 1483For each named address space supported by avr-gcc there is an equally 1484named but uppercase built-in macro defined. 1485The purpose is to facilitate testing if respective address space 1486support is available or not: 1487 1488@smallexample 1489#ifdef __FLASH 1490const __flash int var = 1; 1491 1492int read_var (void) 1493@{ 1494 return var; 1495@} 1496#else 1497#include <avr/pgmspace.h> /* From AVR-LibC */ 1498 1499const int var PROGMEM = 1; 1500 1501int read_var (void) 1502@{ 1503 return (int) pgm_read_word (&var); 1504@} 1505#endif /* __FLASH */ 1506@end smallexample 1507 1508@noindent 1509Notice that attribute @ref{AVR Variable Attributes,,@code{progmem}} 1510locates data in flash but 1511accesses to these data read from generic address space, i.e.@: 1512from RAM, 1513so that you need special accessors like @code{pgm_read_byte} 1514from @w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC}} 1515together with attribute @code{progmem}. 1516 1517@noindent 1518@b{Limitations and caveats} 1519 1520@itemize 1521@item 1522Reading across the 64@tie{}KiB section boundary of 1523the @code{__flash} or @code{__flash@var{N}} address spaces 1524shows undefined behavior. The only address space that 1525supports reading across the 64@tie{}KiB flash segment boundaries is 1526@code{__memx}. 1527 1528@item 1529If you use one of the @code{__flash@var{N}} address spaces 1530you must arrange your linker script to locate the 1531@code{.progmem@var{N}.data} sections according to your needs. 1532 1533@item 1534Any data or pointers to the non-generic address spaces must 1535be qualified as @code{const}, i.e.@: as read-only data. 1536This still applies if the data in one of these address 1537spaces like software version number or calibration lookup table are intended to 1538be changed after load time by, say, a boot loader. In this case 1539the right qualification is @code{const} @code{volatile} so that the compiler 1540must not optimize away known values or insert them 1541as immediates into operands of instructions. 1542 1543@item 1544The following code initializes a variable @code{pfoo} 1545located in static storage with a 24-bit address: 1546@smallexample 1547extern const __memx char foo; 1548const __memx void *pfoo = &foo; 1549@end smallexample 1550 1551@item 1552On the reduced Tiny devices like ATtiny40, no address spaces are supported. 1553Just use vanilla C / C++ code without overhead as outlined above. 1554Attribute @code{progmem} is supported but works differently, 1555see @ref{AVR Variable Attributes}. 1556 1557@end itemize 1558 1559@subsection M32C Named Address Spaces 1560@cindex @code{__far} M32C Named Address Spaces 1561 1562On the M32C target, with the R8C and M16C CPU variants, variables 1563qualified with @code{__far} are accessed using 32-bit addresses in 1564order to access memory beyond the first 64@tie{}Ki bytes. If 1565@code{__far} is used with the M32CM or M32C CPU variants, it has no 1566effect. 1567 1568@subsection RL78 Named Address Spaces 1569@cindex @code{__far} RL78 Named Address Spaces 1570 1571On the RL78 target, variables qualified with @code{__far} are accessed 1572with 32-bit pointers (20-bit addresses) rather than the default 16-bit 1573addresses. Non-far variables are assumed to appear in the topmost 157464@tie{}KiB of the address space. 1575 1576@subsection x86 Named Address Spaces 1577@cindex x86 named address spaces 1578 1579On the x86 target, variables may be declared as being relative 1580to the @code{%fs} or @code{%gs} segments. 1581 1582@table @code 1583@item __seg_fs 1584@itemx __seg_gs 1585@cindex @code{__seg_fs} x86 named address space 1586@cindex @code{__seg_gs} x86 named address space 1587The object is accessed with the respective segment override prefix. 1588 1589The respective segment base must be set via some method specific to 1590the operating system. Rather than require an expensive system call 1591to retrieve the segment base, these address spaces are not considered 1592to be subspaces of the generic (flat) address space. This means that 1593explicit casts are required to convert pointers between these address 1594spaces and the generic address space. In practice the application 1595should cast to @code{uintptr_t} and apply the segment base offset 1596that it installed previously. 1597 1598The preprocessor symbols @code{__SEG_FS} and @code{__SEG_GS} are 1599defined when these address spaces are supported. 1600@end table 1601 1602@node Zero Length 1603@section Arrays of Length Zero 1604@cindex arrays of length zero 1605@cindex zero-length arrays 1606@cindex length-zero arrays 1607@cindex flexible array members 1608 1609Declaring zero-length arrays is allowed in GNU C as an extension. 1610A zero-length array can be useful as the last element of a structure 1611that is really a header for a variable-length object: 1612 1613@smallexample 1614struct line @{ 1615 int length; 1616 char contents[0]; 1617@}; 1618 1619struct line *thisline = (struct line *) 1620 malloc (sizeof (struct line) + this_length); 1621thisline->length = this_length; 1622@end smallexample 1623 1624Although the size of a zero-length array is zero, an array member of 1625this kind may increase the size of the enclosing type as a result of tail 1626padding. The offset of a zero-length array member from the beginning 1627of the enclosing structure is the same as the offset of an array with 1628one or more elements of the same type. The alignment of a zero-length 1629array is the same as the alignment of its elements. 1630 1631Declaring zero-length arrays in other contexts, including as interior 1632members of structure objects or as non-member objects, is discouraged. 1633Accessing elements of zero-length arrays declared in such contexts is 1634undefined and may be diagnosed. 1635 1636In the absence of the zero-length array extension, in ISO C90 1637the @code{contents} array in the example above would typically be declared 1638to have a single element. Unlike a zero-length array which only contributes 1639to the size of the enclosing structure for the purposes of alignment, 1640a one-element array always occupies at least as much space as a single 1641object of the type. Although using one-element arrays this way is 1642discouraged, GCC handles accesses to trailing one-element array members 1643analogously to zero-length arrays. 1644 1645The preferred mechanism to declare variable-length types like 1646@code{struct line} above is the ISO C99 @dfn{flexible array member}, 1647with slightly different syntax and semantics: 1648 1649@itemize @bullet 1650@item 1651Flexible array members are written as @code{contents[]} without 1652the @code{0}. 1653 1654@item 1655Flexible array members have incomplete type, and so the @code{sizeof} 1656operator may not be applied. As a quirk of the original implementation 1657of zero-length arrays, @code{sizeof} evaluates to zero. 1658 1659@item 1660Flexible array members may only appear as the last member of a 1661@code{struct} that is otherwise non-empty. 1662 1663@item 1664A structure containing a flexible array member, or a union containing 1665such a structure (possibly recursively), may not be a member of a 1666structure or an element of an array. (However, these uses are 1667permitted by GCC as extensions.) 1668@end itemize 1669 1670Non-empty initialization of zero-length 1671arrays is treated like any case where there are more initializer 1672elements than the array holds, in that a suitable warning about ``excess 1673elements in array'' is given, and the excess elements (all of them, in 1674this case) are ignored. 1675 1676GCC allows static initialization of flexible array members. 1677This is equivalent to defining a new structure containing the original 1678structure followed by an array of sufficient size to contain the data. 1679E.g.@: in the following, @code{f1} is constructed as if it were declared 1680like @code{f2}. 1681 1682@smallexample 1683struct f1 @{ 1684 int x; int y[]; 1685@} f1 = @{ 1, @{ 2, 3, 4 @} @}; 1686 1687struct f2 @{ 1688 struct f1 f1; int data[3]; 1689@} f2 = @{ @{ 1 @}, @{ 2, 3, 4 @} @}; 1690@end smallexample 1691 1692@noindent 1693The convenience of this extension is that @code{f1} has the desired 1694type, eliminating the need to consistently refer to @code{f2.f1}. 1695 1696This has symmetry with normal static arrays, in that an array of 1697unknown size is also written with @code{[]}. 1698 1699Of course, this extension only makes sense if the extra data comes at 1700the end of a top-level object, as otherwise we would be overwriting 1701data at subsequent offsets. To avoid undue complication and confusion 1702with initialization of deeply nested arrays, we simply disallow any 1703non-empty initialization except when the structure is the top-level 1704object. For example: 1705 1706@smallexample 1707struct foo @{ int x; int y[]; @}; 1708struct bar @{ struct foo z; @}; 1709 1710struct foo a = @{ 1, @{ 2, 3, 4 @} @}; // @r{Valid.} 1711struct bar b = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 1712struct bar c = @{ @{ 1, @{ @} @} @}; // @r{Valid.} 1713struct foo d[1] = @{ @{ 1, @{ 2, 3, 4 @} @} @}; // @r{Invalid.} 1714@end smallexample 1715 1716@node Empty Structures 1717@section Structures with No Members 1718@cindex empty structures 1719@cindex zero-size structures 1720 1721GCC permits a C structure to have no members: 1722 1723@smallexample 1724struct empty @{ 1725@}; 1726@end smallexample 1727 1728The structure has size zero. In C++, empty structures are part 1729of the language. G++ treats empty structures as if they had a single 1730member of type @code{char}. 1731 1732@node Variable Length 1733@section Arrays of Variable Length 1734@cindex variable-length arrays 1735@cindex arrays of variable length 1736@cindex VLAs 1737 1738Variable-length automatic arrays are allowed in ISO C99, and as an 1739extension GCC accepts them in C90 mode and in C++. These arrays are 1740declared like any other automatic arrays, but with a length that is not 1741a constant expression. The storage is allocated at the point of 1742declaration and deallocated when the block scope containing the declaration 1743exits. For 1744example: 1745 1746@smallexample 1747FILE * 1748concat_fopen (char *s1, char *s2, char *mode) 1749@{ 1750 char str[strlen (s1) + strlen (s2) + 1]; 1751 strcpy (str, s1); 1752 strcat (str, s2); 1753 return fopen (str, mode); 1754@} 1755@end smallexample 1756 1757@cindex scope of a variable length array 1758@cindex variable-length array scope 1759@cindex deallocating variable length arrays 1760Jumping or breaking out of the scope of the array name deallocates the 1761storage. Jumping into the scope is not allowed; you get an error 1762message for it. 1763 1764@cindex variable-length array in a structure 1765As an extension, GCC accepts variable-length arrays as a member of 1766a structure or a union. For example: 1767 1768@smallexample 1769void 1770foo (int n) 1771@{ 1772 struct S @{ int x[n]; @}; 1773@} 1774@end smallexample 1775 1776@cindex @code{alloca} vs variable-length arrays 1777You can use the function @code{alloca} to get an effect much like 1778variable-length arrays. The function @code{alloca} is available in 1779many other C implementations (but not in all). On the other hand, 1780variable-length arrays are more elegant. 1781 1782There are other differences between these two methods. Space allocated 1783with @code{alloca} exists until the containing @emph{function} returns. 1784The space for a variable-length array is deallocated as soon as the array 1785name's scope ends, unless you also use @code{alloca} in this scope. 1786 1787You can also use variable-length arrays as arguments to functions: 1788 1789@smallexample 1790struct entry 1791tester (int len, char data[len][len]) 1792@{ 1793 /* @r{@dots{}} */ 1794@} 1795@end smallexample 1796 1797The length of an array is computed once when the storage is allocated 1798and is remembered for the scope of the array in case you access it with 1799@code{sizeof}. 1800 1801If you want to pass the array first and the length afterward, you can 1802use a forward declaration in the parameter list---another GNU extension. 1803 1804@smallexample 1805struct entry 1806tester (int len; char data[len][len], int len) 1807@{ 1808 /* @r{@dots{}} */ 1809@} 1810@end smallexample 1811 1812@cindex parameter forward declaration 1813The @samp{int len} before the semicolon is a @dfn{parameter forward 1814declaration}, and it serves the purpose of making the name @code{len} 1815known when the declaration of @code{data} is parsed. 1816 1817You can write any number of such parameter forward declarations in the 1818parameter list. They can be separated by commas or semicolons, but the 1819last one must end with a semicolon, which is followed by the ``real'' 1820parameter declarations. Each forward declaration must match a ``real'' 1821declaration in parameter name and data type. ISO C99 does not support 1822parameter forward declarations. 1823 1824@node Variadic Macros 1825@section Macros with a Variable Number of Arguments. 1826@cindex variable number of arguments 1827@cindex macro with variable arguments 1828@cindex rest argument (in macro) 1829@cindex variadic macros 1830 1831In the ISO C standard of 1999, a macro can be declared to accept a 1832variable number of arguments much as a function can. The syntax for 1833defining the macro is similar to that of a function. Here is an 1834example: 1835 1836@smallexample 1837#define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) 1838@end smallexample 1839 1840@noindent 1841Here @samp{@dots{}} is a @dfn{variable argument}. In the invocation of 1842such a macro, it represents the zero or more tokens until the closing 1843parenthesis that ends the invocation, including any commas. This set of 1844tokens replaces the identifier @code{__VA_ARGS__} in the macro body 1845wherever it appears. See the CPP manual for more information. 1846 1847GCC has long supported variadic macros, and used a different syntax that 1848allowed you to give a name to the variable arguments just like any other 1849argument. Here is an example: 1850 1851@smallexample 1852#define debug(format, args...) fprintf (stderr, format, args) 1853@end smallexample 1854 1855@noindent 1856This is in all ways equivalent to the ISO C example above, but arguably 1857more readable and descriptive. 1858 1859GNU CPP has two further variadic macro extensions, and permits them to 1860be used with either of the above forms of macro definition. 1861 1862In standard C, you are not allowed to leave the variable argument out 1863entirely; but you are allowed to pass an empty argument. For example, 1864this invocation is invalid in ISO C, because there is no comma after 1865the string: 1866 1867@smallexample 1868debug ("A message") 1869@end smallexample 1870 1871GNU CPP permits you to completely omit the variable arguments in this 1872way. In the above examples, the compiler would complain, though since 1873the expansion of the macro still has the extra comma after the format 1874string. 1875 1876To help solve this problem, CPP behaves specially for variable arguments 1877used with the token paste operator, @samp{##}. If instead you write 1878 1879@smallexample 1880#define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) 1881@end smallexample 1882 1883@noindent 1884and if the variable arguments are omitted or empty, the @samp{##} 1885operator causes the preprocessor to remove the comma before it. If you 1886do provide some variable arguments in your macro invocation, GNU CPP 1887does not complain about the paste operation and instead places the 1888variable arguments after the comma. Just like any other pasted macro 1889argument, these arguments are not macro expanded. 1890 1891@node Escaped Newlines 1892@section Slightly Looser Rules for Escaped Newlines 1893@cindex escaped newlines 1894@cindex newlines (escaped) 1895 1896The preprocessor treatment of escaped newlines is more relaxed 1897than that specified by the C90 standard, which requires the newline 1898to immediately follow a backslash. 1899GCC's implementation allows whitespace in the form 1900of spaces, horizontal and vertical tabs, and form feeds between the 1901backslash and the subsequent newline. The preprocessor issues a 1902warning, but treats it as a valid escaped newline and combines the two 1903lines to form a single logical line. This works within comments and 1904tokens, as well as between tokens. Comments are @emph{not} treated as 1905whitespace for the purposes of this relaxation, since they have not 1906yet been replaced with spaces. 1907 1908@node Subscripting 1909@section Non-Lvalue Arrays May Have Subscripts 1910@cindex subscripting 1911@cindex arrays, non-lvalue 1912 1913@cindex subscripting and function values 1914In ISO C99, arrays that are not lvalues still decay to pointers, and 1915may be subscripted, although they may not be modified or used after 1916the next sequence point and the unary @samp{&} operator may not be 1917applied to them. As an extension, GNU C allows such arrays to be 1918subscripted in C90 mode, though otherwise they do not decay to 1919pointers outside C99 mode. For example, 1920this is valid in GNU C though not valid in C90: 1921 1922@smallexample 1923@group 1924struct foo @{int a[4];@}; 1925 1926struct foo f(); 1927 1928bar (int index) 1929@{ 1930 return f().a[index]; 1931@} 1932@end group 1933@end smallexample 1934 1935@node Pointer Arith 1936@section Arithmetic on @code{void}- and Function-Pointers 1937@cindex void pointers, arithmetic 1938@cindex void, size of pointer to 1939@cindex function pointers, arithmetic 1940@cindex function, size of pointer to 1941 1942In GNU C, addition and subtraction operations are supported on pointers to 1943@code{void} and on pointers to functions. This is done by treating the 1944size of a @code{void} or of a function as 1. 1945 1946A consequence of this is that @code{sizeof} is also allowed on @code{void} 1947and on function types, and returns 1. 1948 1949@opindex Wpointer-arith 1950The option @option{-Wpointer-arith} requests a warning if these extensions 1951are used. 1952 1953@node Variadic Pointer Args 1954@section Pointer Arguments in Variadic Functions 1955@cindex pointer arguments in variadic functions 1956@cindex variadic functions, pointer arguments 1957 1958Standard C requires that pointer types used with @code{va_arg} in 1959functions with variable argument lists either must be compatible with 1960that of the actual argument, or that one type must be a pointer to 1961@code{void} and the other a pointer to a character type. GNU C 1962implements the POSIX XSI extension that additionally permits the use 1963of @code{va_arg} with a pointer type to receive arguments of any other 1964pointer type. 1965 1966In particular, in GNU C @samp{va_arg (ap, void *)} can safely be used 1967to consume an argument of any pointer type. 1968 1969@node Pointers to Arrays 1970@section Pointers to Arrays with Qualifiers Work as Expected 1971@cindex pointers to arrays 1972@cindex const qualifier 1973 1974In GNU C, pointers to arrays with qualifiers work similar to pointers 1975to other qualified types. For example, a value of type @code{int (*)[5]} 1976can be used to initialize a variable of type @code{const int (*)[5]}. 1977These types are incompatible in ISO C because the @code{const} qualifier 1978is formally attached to the element type of the array and not the 1979array itself. 1980 1981@smallexample 1982extern void 1983transpose (int N, int M, double out[M][N], const double in[N][M]); 1984double x[3][2]; 1985double y[2][3]; 1986@r{@dots{}} 1987transpose(3, 2, y, x); 1988@end smallexample 1989 1990@node Initializers 1991@section Non-Constant Initializers 1992@cindex initializers, non-constant 1993@cindex non-constant initializers 1994 1995As in standard C++ and ISO C99, the elements of an aggregate initializer for an 1996automatic variable are not required to be constant expressions in GNU C@. 1997Here is an example of an initializer with run-time varying elements: 1998 1999@smallexample 2000foo (float f, float g) 2001@{ 2002 float beat_freqs[2] = @{ f-g, f+g @}; 2003 /* @r{@dots{}} */ 2004@} 2005@end smallexample 2006 2007@node Compound Literals 2008@section Compound Literals 2009@cindex constructor expressions 2010@cindex initializations in expressions 2011@cindex structures, constructor expression 2012@cindex expressions, constructor 2013@cindex compound literals 2014@c The GNU C name for what C99 calls compound literals was "constructor expressions". 2015 2016A compound literal looks like a cast of a brace-enclosed aggregate 2017initializer list. Its value is an object of the type specified in 2018the cast, containing the elements specified in the initializer. 2019Unlike the result of a cast, a compound literal is an lvalue. ISO 2020C99 and later support compound literals. As an extension, GCC 2021supports compound literals also in C90 mode and in C++, although 2022as explained below, the C++ semantics are somewhat different. 2023 2024Usually, the specified type of a compound literal is a structure. Assume 2025that @code{struct foo} and @code{structure} are declared as shown: 2026 2027@smallexample 2028struct foo @{int a; char b[2];@} structure; 2029@end smallexample 2030 2031@noindent 2032Here is an example of constructing a @code{struct foo} with a compound literal: 2033 2034@smallexample 2035structure = ((struct foo) @{x + y, 'a', 0@}); 2036@end smallexample 2037 2038@noindent 2039This is equivalent to writing the following: 2040 2041@smallexample 2042@{ 2043 struct foo temp = @{x + y, 'a', 0@}; 2044 structure = temp; 2045@} 2046@end smallexample 2047 2048You can also construct an array, though this is dangerous in C++, as 2049explained below. If all the elements of the compound literal are 2050(made up of) simple constant expressions suitable for use in 2051initializers of objects of static storage duration, then the compound 2052literal can be coerced to a pointer to its first element and used in 2053such an initializer, as shown here: 2054 2055@smallexample 2056char **foo = (char *[]) @{ "x", "y", "z" @}; 2057@end smallexample 2058 2059Compound literals for scalar types and union types are also allowed. In 2060the following example the variable @code{i} is initialized to the value 2061@code{2}, the result of incrementing the unnamed object created by 2062the compound literal. 2063 2064@smallexample 2065int i = ++(int) @{ 1 @}; 2066@end smallexample 2067 2068As a GNU extension, GCC allows initialization of objects with static storage 2069duration by compound literals (which is not possible in ISO C99 because 2070the initializer is not a constant). 2071It is handled as if the object were initialized only with the brace-enclosed 2072list if the types of the compound literal and the object match. 2073The elements of the compound literal must be constant. 2074If the object being initialized has array type of unknown size, the size is 2075determined by the size of the compound literal. 2076 2077@smallexample 2078static struct foo x = (struct foo) @{1, 'a', 'b'@}; 2079static int y[] = (int []) @{1, 2, 3@}; 2080static int z[] = (int [3]) @{1@}; 2081@end smallexample 2082 2083@noindent 2084The above lines are equivalent to the following: 2085@smallexample 2086static struct foo x = @{1, 'a', 'b'@}; 2087static int y[] = @{1, 2, 3@}; 2088static int z[] = @{1, 0, 0@}; 2089@end smallexample 2090 2091In C, a compound literal designates an unnamed object with static or 2092automatic storage duration. In C++, a compound literal designates a 2093temporary object that only lives until the end of its full-expression. 2094As a result, well-defined C code that takes the address of a subobject 2095of a compound literal can be undefined in C++, so G++ rejects 2096the conversion of a temporary array to a pointer. For instance, if 2097the array compound literal example above appeared inside a function, 2098any subsequent use of @code{foo} in C++ would have undefined behavior 2099because the lifetime of the array ends after the declaration of @code{foo}. 2100 2101As an optimization, G++ sometimes gives array compound literals longer 2102lifetimes: when the array either appears outside a function or has 2103a @code{const}-qualified type. If @code{foo} and its initializer had 2104elements of type @code{char *const} rather than @code{char *}, or if 2105@code{foo} were a global variable, the array would have static storage 2106duration. But it is probably safest just to avoid the use of array 2107compound literals in C++ code. 2108 2109@node Designated Inits 2110@section Designated Initializers 2111@cindex initializers with labeled elements 2112@cindex labeled elements in initializers 2113@cindex case labels in initializers 2114@cindex designated initializers 2115 2116Standard C90 requires the elements of an initializer to appear in a fixed 2117order, the same as the order of the elements in the array or structure 2118being initialized. 2119 2120In ISO C99 you can give the elements in any order, specifying the array 2121indices or structure field names they apply to, and GNU C allows this as 2122an extension in C90 mode as well. This extension is not 2123implemented in GNU C++. 2124 2125To specify an array index, write 2126@samp{[@var{index}] =} before the element value. For example, 2127 2128@smallexample 2129int a[6] = @{ [4] = 29, [2] = 15 @}; 2130@end smallexample 2131 2132@noindent 2133is equivalent to 2134 2135@smallexample 2136int a[6] = @{ 0, 0, 15, 0, 29, 0 @}; 2137@end smallexample 2138 2139@noindent 2140The index values must be constant expressions, even if the array being 2141initialized is automatic. 2142 2143An alternative syntax for this that has been obsolete since GCC 2.5 but 2144GCC still accepts is to write @samp{[@var{index}]} before the element 2145value, with no @samp{=}. 2146 2147To initialize a range of elements to the same value, write 2148@samp{[@var{first} ... @var{last}] = @var{value}}. This is a GNU 2149extension. For example, 2150 2151@smallexample 2152int widths[] = @{ [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 @}; 2153@end smallexample 2154 2155@noindent 2156If the value in it has side effects, the side effects happen only once, 2157not for each initialized field by the range initializer. 2158 2159@noindent 2160Note that the length of the array is the highest value specified 2161plus one. 2162 2163In a structure initializer, specify the name of a field to initialize 2164with @samp{.@var{fieldname} =} before the element value. For example, 2165given the following structure, 2166 2167@smallexample 2168struct point @{ int x, y; @}; 2169@end smallexample 2170 2171@noindent 2172the following initialization 2173 2174@smallexample 2175struct point p = @{ .y = yvalue, .x = xvalue @}; 2176@end smallexample 2177 2178@noindent 2179is equivalent to 2180 2181@smallexample 2182struct point p = @{ xvalue, yvalue @}; 2183@end smallexample 2184 2185Another syntax that has the same meaning, obsolete since GCC 2.5, is 2186@samp{@var{fieldname}:}, as shown here: 2187 2188@smallexample 2189struct point p = @{ y: yvalue, x: xvalue @}; 2190@end smallexample 2191 2192Omitted fields are implicitly initialized the same as for objects 2193that have static storage duration. 2194 2195@cindex designators 2196The @samp{[@var{index}]} or @samp{.@var{fieldname}} is known as a 2197@dfn{designator}. You can also use a designator (or the obsolete colon 2198syntax) when initializing a union, to specify which element of the union 2199should be used. For example, 2200 2201@smallexample 2202union foo @{ int i; double d; @}; 2203 2204union foo f = @{ .d = 4 @}; 2205@end smallexample 2206 2207@noindent 2208converts 4 to a @code{double} to store it in the union using 2209the second element. By contrast, casting 4 to type @code{union foo} 2210stores it into the union as the integer @code{i}, since it is 2211an integer. @xref{Cast to Union}. 2212 2213You can combine this technique of naming elements with ordinary C 2214initialization of successive elements. Each initializer element that 2215does not have a designator applies to the next consecutive element of the 2216array or structure. For example, 2217 2218@smallexample 2219int a[6] = @{ [1] = v1, v2, [4] = v4 @}; 2220@end smallexample 2221 2222@noindent 2223is equivalent to 2224 2225@smallexample 2226int a[6] = @{ 0, v1, v2, 0, v4, 0 @}; 2227@end smallexample 2228 2229Labeling the elements of an array initializer is especially useful 2230when the indices are characters or belong to an @code{enum} type. 2231For example: 2232 2233@smallexample 2234int whitespace[256] 2235 = @{ [' '] = 1, ['\t'] = 1, ['\h'] = 1, 2236 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 @}; 2237@end smallexample 2238 2239@cindex designator lists 2240You can also write a series of @samp{.@var{fieldname}} and 2241@samp{[@var{index}]} designators before an @samp{=} to specify a 2242nested subobject to initialize; the list is taken relative to the 2243subobject corresponding to the closest surrounding brace pair. For 2244example, with the @samp{struct point} declaration above: 2245 2246@smallexample 2247struct point ptarray[10] = @{ [2].y = yv2, [2].x = xv2, [0].x = xv0 @}; 2248@end smallexample 2249 2250If the same field is initialized multiple times, or overlapping 2251fields of a union are initialized, the value from the last 2252initialization is used. When a field of a union is itself a structure, 2253the entire structure from the last field initialized is used. If any previous 2254initializer has side effect, it is unspecified whether the side effect 2255happens or not. Currently, GCC discards the side-effecting 2256initializer expressions and issues a warning. 2257 2258@node Case Ranges 2259@section Case Ranges 2260@cindex case ranges 2261@cindex ranges in case statements 2262 2263You can specify a range of consecutive values in a single @code{case} label, 2264like this: 2265 2266@smallexample 2267case @var{low} ... @var{high}: 2268@end smallexample 2269 2270@noindent 2271This has the same effect as the proper number of individual @code{case} 2272labels, one for each integer value from @var{low} to @var{high}, inclusive. 2273 2274This feature is especially useful for ranges of ASCII character codes: 2275 2276@smallexample 2277case 'A' ... 'Z': 2278@end smallexample 2279 2280@strong{Be careful:} Write spaces around the @code{...}, for otherwise 2281it may be parsed wrong when you use it with integer values. For example, 2282write this: 2283 2284@smallexample 2285case 1 ... 5: 2286@end smallexample 2287 2288@noindent 2289rather than this: 2290 2291@smallexample 2292case 1...5: 2293@end smallexample 2294 2295@node Cast to Union 2296@section Cast to a Union Type 2297@cindex cast to a union 2298@cindex union, casting to a 2299 2300A cast to a union type is a C extension not available in C++. It looks 2301just like ordinary casts with the constraint that the type specified is 2302a union type. You can specify the type either with the @code{union} 2303keyword or with a @code{typedef} name that refers to a union. The result 2304of a cast to a union is a temporary rvalue of the union type with a member 2305whose type matches that of the operand initialized to the value of 2306the operand. The effect of a cast to a union is similar to a compound 2307literal except that it yields an rvalue like standard casts do. 2308@xref{Compound Literals}. 2309 2310Expressions that may be cast to the union type are those whose type matches 2311at least one of the members of the union. Thus, given the following union 2312and variables: 2313 2314@smallexample 2315union foo @{ int i; double d; @}; 2316int x; 2317double y; 2318union foo z; 2319@end smallexample 2320 2321@noindent 2322both @code{x} and @code{y} can be cast to type @code{union foo} and 2323the following assignments 2324@smallexample 2325 z = (union foo) x; 2326 z = (union foo) y; 2327@end smallexample 2328are shorthand equivalents of these 2329@smallexample 2330 z = (union foo) @{ .i = x @}; 2331 z = (union foo) @{ .d = y @}; 2332@end smallexample 2333 2334However, @code{(union foo) FLT_MAX;} is not a valid cast because the union 2335has no member of type @code{float}. 2336 2337Using the cast as the right-hand side of an assignment to a variable of 2338union type is equivalent to storing in a member of the union with 2339the same type 2340 2341@smallexample 2342union foo u; 2343/* @r{@dots{}} */ 2344u = (union foo) x @equiv{} u.i = x 2345u = (union foo) y @equiv{} u.d = y 2346@end smallexample 2347 2348You can also use the union cast as a function argument: 2349 2350@smallexample 2351void hack (union foo); 2352/* @r{@dots{}} */ 2353hack ((union foo) x); 2354@end smallexample 2355 2356@node Mixed Labels and Declarations 2357@section Mixed Declarations, Labels and Code 2358@cindex mixed declarations and code 2359@cindex declarations, mixed with code 2360@cindex code, mixed with declarations 2361 2362ISO C99 and ISO C++ allow declarations and code to be freely mixed 2363within compound statements. ISO C2X allows labels to be 2364placed before declarations and at the end of a compound statement. 2365As an extension, GNU C also allows all this in C90 mode. For example, 2366you could do: 2367 2368@smallexample 2369int i; 2370/* @r{@dots{}} */ 2371i++; 2372int j = i + 2; 2373@end smallexample 2374 2375Each identifier is visible from where it is declared until the end of 2376the enclosing block. 2377 2378@node Function Attributes 2379@section Declaring Attributes of Functions 2380@cindex function attributes 2381@cindex declaring attributes of functions 2382@cindex @code{volatile} applied to function 2383@cindex @code{const} applied to function 2384 2385In GNU C and C++, you can use function attributes to specify certain 2386function properties that may help the compiler optimize calls or 2387check code more carefully for correctness. For example, you 2388can use attributes to specify that a function never returns 2389(@code{noreturn}), returns a value depending only on the values of 2390its arguments (@code{const}), or has @code{printf}-style arguments 2391(@code{format}). 2392 2393You can also use attributes to control memory placement, code 2394generation options or call/return conventions within the function 2395being annotated. Many of these attributes are target-specific. For 2396example, many targets support attributes for defining interrupt 2397handler functions, which typically must follow special register usage 2398and return conventions. Such attributes are described in the subsection 2399for each target. However, a considerable number of attributes are 2400supported by most, if not all targets. Those are described in 2401the @ref{Common Function Attributes} section. 2402 2403Function attributes are introduced by the @code{__attribute__} keyword 2404in the declaration of a function, followed by an attribute specification 2405enclosed in double parentheses. You can specify multiple attributes in 2406a declaration by separating them by commas within the double parentheses 2407or by immediately following one attribute specification with another. 2408@xref{Attribute Syntax}, for the exact rules on attribute syntax and 2409placement. Compatible attribute specifications on distinct declarations 2410of the same function are merged. An attribute specification that is not 2411compatible with attributes already applied to a declaration of the same 2412function is ignored with a warning. 2413 2414Some function attributes take one or more arguments that refer to 2415the function's parameters by their positions within the function parameter 2416list. Such attribute arguments are referred to as @dfn{positional arguments}. 2417Unless specified otherwise, positional arguments that specify properties 2418of parameters with pointer types can also specify the same properties of 2419the implicit C++ @code{this} argument in non-static member functions, and 2420of parameters of reference to a pointer type. For ordinary functions, 2421position one refers to the first parameter on the list. In C++ non-static 2422member functions, position one refers to the implicit @code{this} pointer. 2423The same restrictions and effects apply to function attributes used with 2424ordinary functions or C++ member functions. 2425 2426GCC also supports attributes on 2427variable declarations (@pxref{Variable Attributes}), 2428labels (@pxref{Label Attributes}), 2429enumerators (@pxref{Enumerator Attributes}), 2430statements (@pxref{Statement Attributes}), 2431and types (@pxref{Type Attributes}). 2432 2433There is some overlap between the purposes of attributes and pragmas 2434(@pxref{Pragmas,,Pragmas Accepted by GCC}). It has been 2435found convenient to use @code{__attribute__} to achieve a natural 2436attachment of attributes to their corresponding declarations, whereas 2437@code{#pragma} is of use for compatibility with other compilers 2438or constructs that do not naturally form part of the grammar. 2439 2440In addition to the attributes documented here, 2441GCC plugins may provide their own attributes. 2442 2443@menu 2444* Common Function Attributes:: 2445* AArch64 Function Attributes:: 2446* AMD GCN Function Attributes:: 2447* ARC Function Attributes:: 2448* ARM Function Attributes:: 2449* AVR Function Attributes:: 2450* Blackfin Function Attributes:: 2451* BPF Function Attributes:: 2452* CR16 Function Attributes:: 2453* C-SKY Function Attributes:: 2454* Epiphany Function Attributes:: 2455* H8/300 Function Attributes:: 2456* IA-64 Function Attributes:: 2457* M32C Function Attributes:: 2458* M32R/D Function Attributes:: 2459* m68k Function Attributes:: 2460* MCORE Function Attributes:: 2461* MeP Function Attributes:: 2462* MicroBlaze Function Attributes:: 2463* Microsoft Windows Function Attributes:: 2464* MIPS Function Attributes:: 2465* MSP430 Function Attributes:: 2466* NDS32 Function Attributes:: 2467* Nios II Function Attributes:: 2468* Nvidia PTX Function Attributes:: 2469* PowerPC Function Attributes:: 2470* RISC-V Function Attributes:: 2471* RL78 Function Attributes:: 2472* RX Function Attributes:: 2473* S/390 Function Attributes:: 2474* SH Function Attributes:: 2475* Symbian OS Function Attributes:: 2476* V850 Function Attributes:: 2477* Visium Function Attributes:: 2478* x86 Function Attributes:: 2479* Xstormy16 Function Attributes:: 2480@end menu 2481 2482@node Common Function Attributes 2483@subsection Common Function Attributes 2484 2485The following attributes are supported on most targets. 2486 2487@table @code 2488@c Keep this table alphabetized by attribute name. Treat _ as space. 2489 2490@item access 2491@itemx access (@var{access-mode}, @var{ref-index}) 2492@itemx access (@var{access-mode}, @var{ref-index}, @var{size-index}) 2493 2494The @code{access} attribute enables the detection of invalid or unsafe 2495accesses by functions to which they apply or their callers, as well as 2496write-only accesses to objects that are never read from. Such accesses 2497may be diagnosed by warnings such as @option{-Wstringop-overflow}, 2498@option{-Wuninitialized}, @option{-Wunused}, and others. 2499 2500The @code{access} attribute specifies that a function to whose by-reference 2501arguments the attribute applies accesses the referenced object according to 2502@var{access-mode}. The @var{access-mode} argument is required and must be 2503one of four names: @code{read_only}, @code{read_write}, @code{write_only}, 2504or @code{none}. The remaining two are positional arguments. 2505 2506The required @var{ref-index} positional argument denotes a function 2507argument of pointer (or in C++, reference) type that is subject to 2508the access. The same pointer argument can be referenced by at most one 2509distinct @code{access} attribute. 2510 2511The optional @var{size-index} positional argument denotes a function 2512argument of integer type that specifies the maximum size of the access. 2513The size is the number of elements of the type referenced by @var{ref-index}, 2514or the number of bytes when the pointer type is @code{void*}. When no 2515@var{size-index} argument is specified, the pointer argument must be either 2516null or point to a space that is suitably aligned and large for at least one 2517object of the referenced type (this implies that a past-the-end pointer is 2518not a valid argument). The actual size of the access may be less but it 2519must not be more. 2520 2521The @code{read_only} access mode specifies that the pointer to which it 2522applies is used to read the referenced object but not write to it. Unless 2523the argument specifying the size of the access denoted by @var{size-index} 2524is zero, the referenced object must be initialized. The mode implies 2525a stronger guarantee than the @code{const} qualifier which, when cast away 2526from a pointer, does not prevent the pointed-to object from being modified. 2527Examples of the use of the @code{read_only} access mode is the argument to 2528the @code{puts} function, or the second and third arguments to 2529the @code{memcpy} function. 2530 2531@smallexample 2532__attribute__ ((access (read_only, 1))) int puts (const char*); 2533__attribute__ ((access (read_only, 2, 3))) void* memcpy (void*, const void*, size_t); 2534@end smallexample 2535 2536The @code{read_write} access mode applies to arguments of pointer types 2537without the @code{const} qualifier. It specifies that the pointer to which 2538it applies is used to both read and write the referenced object. Unless 2539the argument specifying the size of the access denoted by @var{size-index} 2540is zero, the object referenced by the pointer must be initialized. An example 2541of the use of the @code{read_write} access mode is the first argument to 2542the @code{strcat} function. 2543 2544@smallexample 2545__attribute__ ((access (read_write, 1), access (read_only, 2))) char* strcat (char*, const char*); 2546@end smallexample 2547 2548The @code{write_only} access mode applies to arguments of pointer types 2549without the @code{const} qualifier. It specifies that the pointer to which 2550it applies is used to write to the referenced object but not read from it. 2551The object referenced by the pointer need not be initialized. An example 2552of the use of the @code{write_only} access mode is the first argument to 2553the @code{strcpy} function, or the first two arguments to the @code{fgets} 2554function. 2555 2556@smallexample 2557__attribute__ ((access (write_only, 1), access (read_only, 2))) char* strcpy (char*, const char*); 2558__attribute__ ((access (write_only, 1, 2), access (read_write, 3))) int fgets (char*, int, FILE*); 2559@end smallexample 2560 2561The access mode @code{none} specifies that the pointer to which it applies 2562is not used to access the referenced object at all. Unless the pointer is 2563null the pointed-to object must exist and have at least the size as denoted 2564by the @var{size-index} argument. The object need not be initialized. 2565The mode is intended to be used as a means to help validate the expected 2566object size, for example in functions that call @code{__builtin_object_size}. 2567@xref{Object Size Checking}. 2568 2569@item alias ("@var{target}") 2570@cindex @code{alias} function attribute 2571The @code{alias} attribute causes the declaration to be emitted as an alias 2572for another symbol, which must have been previously declared with the same 2573type, and for variables, also the same size and alignment. Declaring an alias 2574with a different type than the target is undefined and may be diagnosed. As 2575an example, the following declarations: 2576 2577@smallexample 2578void __f () @{ /* @r{Do something.} */; @} 2579void f () __attribute__ ((weak, alias ("__f"))); 2580@end smallexample 2581 2582@noindent 2583define @samp{f} to be a weak alias for @samp{__f}. In C++, the mangled name 2584for the target must be used. It is an error if @samp{__f} is not defined in 2585the same translation unit. 2586 2587This attribute requires assembler and object file support, 2588and may not be available on all targets. 2589 2590@item aligned 2591@itemx aligned (@var{alignment}) 2592@cindex @code{aligned} function attribute 2593The @code{aligned} attribute specifies a minimum alignment for 2594the first instruction of the function, measured in bytes. When specified, 2595@var{alignment} must be an integer constant power of 2. Specifying no 2596@var{alignment} argument implies the ideal alignment for the target. 2597The @code{__alignof__} operator can be used to determine what that is 2598(@pxref{Alignment}). The attribute has no effect when a definition for 2599the function is not provided in the same translation unit. 2600 2601The attribute cannot be used to decrease the alignment of a function 2602previously declared with a more restrictive alignment; only to increase 2603it. Attempts to do otherwise are diagnosed. Some targets specify 2604a minimum default alignment for functions that is greater than 1. On 2605such targets, specifying a less restrictive alignment is silently ignored. 2606Using the attribute overrides the effect of the @option{-falign-functions} 2607(@pxref{Optimize Options}) option for this function. 2608 2609Note that the effectiveness of @code{aligned} attributes may be 2610limited by inherent limitations in the system linker 2611and/or object file format. On some systems, the 2612linker is only able to arrange for functions to be aligned up to a 2613certain maximum alignment. (For some linkers, the maximum supported 2614alignment may be very very small.) See your linker documentation for 2615further information. 2616 2617The @code{aligned} attribute can also be used for variables and fields 2618(@pxref{Variable Attributes}.) 2619 2620@item alloc_align (@var{position}) 2621@cindex @code{alloc_align} function attribute 2622The @code{alloc_align} attribute may be applied to a function that 2623returns a pointer and takes at least one argument of an integer or 2624enumerated type. 2625It indicates that the returned pointer is aligned on a boundary given 2626by the function argument at @var{position}. Meaningful alignments are 2627powers of 2 greater than one. GCC uses this information to improve 2628pointer alignment analysis. 2629 2630The function parameter denoting the allocated alignment is specified by 2631one constant integer argument whose number is the argument of the attribute. 2632Argument numbering starts at one. 2633 2634For instance, 2635 2636@smallexample 2637void* my_memalign (size_t, size_t) __attribute__ ((alloc_align (1))); 2638@end smallexample 2639 2640@noindent 2641declares that @code{my_memalign} returns memory with minimum alignment 2642given by parameter 1. 2643 2644@item alloc_size (@var{position}) 2645@itemx alloc_size (@var{position-1}, @var{position-2}) 2646@cindex @code{alloc_size} function attribute 2647The @code{alloc_size} attribute may be applied to a function that 2648returns a pointer and takes at least one argument of an integer or 2649enumerated type. 2650It indicates that the returned pointer points to memory whose size is 2651given by the function argument at @var{position-1}, or by the product 2652of the arguments at @var{position-1} and @var{position-2}. Meaningful 2653sizes are positive values less than @code{PTRDIFF_MAX}. GCC uses this 2654information to improve the results of @code{__builtin_object_size}. 2655 2656The function parameter(s) denoting the allocated size are specified by 2657one or two integer arguments supplied to the attribute. The allocated size 2658is either the value of the single function argument specified or the product 2659of the two function arguments specified. Argument numbering starts at 2660one for ordinary functions, and at two for C++ non-static member functions. 2661 2662For instance, 2663 2664@smallexample 2665void* my_calloc (size_t, size_t) __attribute__ ((alloc_size (1, 2))); 2666void* my_realloc (void*, size_t) __attribute__ ((alloc_size (2))); 2667@end smallexample 2668 2669@noindent 2670declares that @code{my_calloc} returns memory of the size given by 2671the product of parameter 1 and 2 and that @code{my_realloc} returns memory 2672of the size given by parameter 2. 2673 2674@item always_inline 2675@cindex @code{always_inline} function attribute 2676Generally, functions are not inlined unless optimization is specified. 2677For functions declared inline, this attribute inlines the function 2678independent of any restrictions that otherwise apply to inlining. 2679Failure to inline such a function is diagnosed as an error. 2680Note that if such a function is called indirectly the compiler may 2681or may not inline it depending on optimization level and a failure 2682to inline an indirect call may or may not be diagnosed. 2683 2684@item artificial 2685@cindex @code{artificial} function attribute 2686This attribute is useful for small inline wrappers that if possible 2687should appear during debugging as a unit. Depending on the debug 2688info format it either means marking the function as artificial 2689or using the caller location for all instructions within the inlined 2690body. 2691 2692@item assume_aligned (@var{alignment}) 2693@itemx assume_aligned (@var{alignment}, @var{offset}) 2694@cindex @code{assume_aligned} function attribute 2695The @code{assume_aligned} attribute may be applied to a function that 2696returns a pointer. It indicates that the returned pointer is aligned 2697on a boundary given by @var{alignment}. If the attribute has two 2698arguments, the second argument is misalignment @var{offset}. Meaningful 2699values of @var{alignment} are powers of 2 greater than one. Meaningful 2700values of @var{offset} are greater than zero and less than @var{alignment}. 2701 2702For instance 2703 2704@smallexample 2705void* my_alloc1 (size_t) __attribute__((assume_aligned (16))); 2706void* my_alloc2 (size_t) __attribute__((assume_aligned (32, 8))); 2707@end smallexample 2708 2709@noindent 2710declares that @code{my_alloc1} returns 16-byte aligned pointers and 2711that @code{my_alloc2} returns a pointer whose value modulo 32 is equal 2712to 8. 2713 2714@item cold 2715@cindex @code{cold} function attribute 2716The @code{cold} attribute on functions is used to inform the compiler that 2717the function is unlikely to be executed. The function is optimized for 2718size rather than speed and on many targets it is placed into a special 2719subsection of the text section so all cold functions appear close together, 2720improving code locality of non-cold parts of program. The paths leading 2721to calls of cold functions within code are marked as unlikely by the branch 2722prediction mechanism. It is thus useful to mark functions used to handle 2723unlikely conditions, such as @code{perror}, as cold to improve optimization 2724of hot functions that do call marked functions in rare occasions. 2725 2726When profile feedback is available, via @option{-fprofile-use}, cold functions 2727are automatically detected and this attribute is ignored. 2728 2729@item const 2730@cindex @code{const} function attribute 2731@cindex functions that have no side effects 2732Calls to functions whose return value is not affected by changes to 2733the observable state of the program and that have no observable effects 2734on such state other than to return a value may lend themselves to 2735optimizations such as common subexpression elimination. Declaring such 2736functions with the @code{const} attribute allows GCC to avoid emitting 2737some calls in repeated invocations of the function with the same argument 2738values. 2739 2740For example, 2741 2742@smallexample 2743int square (int) __attribute__ ((const)); 2744@end smallexample 2745 2746@noindent 2747tells GCC that subsequent calls to function @code{square} with the same 2748argument value can be replaced by the result of the first call regardless 2749of the statements in between. 2750 2751The @code{const} attribute prohibits a function from reading objects 2752that affect its return value between successive invocations. However, 2753functions declared with the attribute can safely read objects that do 2754not change their return value, such as non-volatile constants. 2755 2756The @code{const} attribute imposes greater restrictions on a function's 2757definition than the similar @code{pure} attribute. Declaring the same 2758function with both the @code{const} and the @code{pure} attribute is 2759diagnosed. Because a const function cannot have any observable side 2760effects it does not make sense for it to return @code{void}. Declaring 2761such a function is diagnosed. 2762 2763@cindex pointer arguments 2764Note that a function that has pointer arguments and examines the data 2765pointed to must @emph{not} be declared @code{const} if the pointed-to 2766data might change between successive invocations of the function. In 2767general, since a function cannot distinguish data that might change 2768from data that cannot, const functions should never take pointer or, 2769in C++, reference arguments. Likewise, a function that calls a non-const 2770function usually must not be const itself. 2771 2772@item constructor 2773@itemx destructor 2774@itemx constructor (@var{priority}) 2775@itemx destructor (@var{priority}) 2776@cindex @code{constructor} function attribute 2777@cindex @code{destructor} function attribute 2778The @code{constructor} attribute causes the function to be called 2779automatically before execution enters @code{main ()}. Similarly, the 2780@code{destructor} attribute causes the function to be called 2781automatically after @code{main ()} completes or @code{exit ()} is 2782called. Functions with these attributes are useful for 2783initializing data that is used implicitly during the execution of 2784the program. 2785 2786On some targets the attributes also accept an integer argument to 2787specify a priority to control the order in which constructor and 2788destructor functions are run. A constructor 2789with a smaller priority number runs before a constructor with a larger 2790priority number; the opposite relationship holds for destructors. So, 2791if you have a constructor that allocates a resource and a destructor 2792that deallocates the same resource, both functions typically have the 2793same priority. The priorities for constructor and destructor 2794functions are the same as those specified for namespace-scope C++ 2795objects (@pxref{C++ Attributes}). However, at present, the order in which 2796constructors for C++ objects with static storage duration and functions 2797decorated with attribute @code{constructor} are invoked is unspecified. 2798In mixed declarations, attribute @code{init_priority} can be used to 2799impose a specific ordering. 2800 2801Using the argument forms of the @code{constructor} and @code{destructor} 2802attributes on targets where the feature is not supported is rejected with 2803an error. 2804 2805@item copy 2806@itemx copy (@var{function}) 2807@cindex @code{copy} function attribute 2808The @code{copy} attribute applies the set of attributes with which 2809@var{function} has been declared to the declaration of the function 2810to which the attribute is applied. The attribute is designed for 2811libraries that define aliases or function resolvers that are expected 2812to specify the same set of attributes as their targets. The @code{copy} 2813attribute can be used with functions, variables, or types. However, 2814the kind of symbol to which the attribute is applied (either function 2815or variable) must match the kind of symbol to which the argument refers. 2816The @code{copy} attribute copies only syntactic and semantic attributes 2817but not attributes that affect a symbol's linkage or visibility such as 2818@code{alias}, @code{visibility}, or @code{weak}. The @code{deprecated} 2819and @code{target_clones} attribute are also not copied. 2820@xref{Common Type Attributes}. 2821@xref{Common Variable Attributes}. 2822 2823For example, the @var{StrongAlias} macro below makes use of the @code{alias} 2824and @code{copy} attributes to define an alias named @var{alloc} for function 2825@var{allocate} declared with attributes @var{alloc_size}, @var{malloc}, and 2826@var{nothrow}. Thanks to the @code{__typeof__} operator the alias has 2827the same type as the target function. As a result of the @code{copy} 2828attribute the alias also shares the same attributes as the target. 2829 2830@smallexample 2831#define StrongAlias(TargetFunc, AliasDecl) \ 2832 extern __typeof__ (TargetFunc) AliasDecl \ 2833 __attribute__ ((alias (#TargetFunc), copy (TargetFunc))); 2834 2835extern __attribute__ ((alloc_size (1), malloc, nothrow)) 2836 void* allocate (size_t); 2837StrongAlias (allocate, alloc); 2838@end smallexample 2839 2840@item deprecated 2841@itemx deprecated (@var{msg}) 2842@cindex @code{deprecated} function attribute 2843The @code{deprecated} attribute results in a warning if the function 2844is used anywhere in the source file. This is useful when identifying 2845functions that are expected to be removed in a future version of a 2846program. The warning also includes the location of the declaration 2847of the deprecated function, to enable users to easily find further 2848information about why the function is deprecated, or what they should 2849do instead. Note that the warnings only occurs for uses: 2850 2851@smallexample 2852int old_fn () __attribute__ ((deprecated)); 2853int old_fn (); 2854int (*fn_ptr)() = old_fn; 2855@end smallexample 2856 2857@noindent 2858results in a warning on line 3 but not line 2. The optional @var{msg} 2859argument, which must be a string, is printed in the warning if 2860present. 2861 2862The @code{deprecated} attribute can also be used for variables and 2863types (@pxref{Variable Attributes}, @pxref{Type Attributes}.) 2864 2865The message attached to the attribute is affected by the setting of 2866the @option{-fmessage-length} option. 2867 2868@item error ("@var{message}") 2869@itemx warning ("@var{message}") 2870@cindex @code{error} function attribute 2871@cindex @code{warning} function attribute 2872If the @code{error} or @code{warning} attribute 2873is used on a function declaration and a call to such a function 2874is not eliminated through dead code elimination or other optimizations, 2875an error or warning (respectively) that includes @var{message} is diagnosed. 2876This is useful 2877for compile-time checking, especially together with @code{__builtin_constant_p} 2878and inline functions where checking the inline function arguments is not 2879possible through @code{extern char [(condition) ? 1 : -1];} tricks. 2880 2881While it is possible to leave the function undefined and thus invoke 2882a link failure (to define the function with 2883a message in @code{.gnu.warning*} section), 2884when using these attributes the problem is diagnosed 2885earlier and with exact location of the call even in presence of inline 2886functions or when not emitting debugging information. 2887 2888@item externally_visible 2889@cindex @code{externally_visible} function attribute 2890This attribute, attached to a global variable or function, nullifies 2891the effect of the @option{-fwhole-program} command-line option, so the 2892object remains visible outside the current compilation unit. 2893 2894If @option{-fwhole-program} is used together with @option{-flto} and 2895@command{gold} is used as the linker plugin, 2896@code{externally_visible} attributes are automatically added to functions 2897(not variable yet due to a current @command{gold} issue) 2898that are accessed outside of LTO objects according to resolution file 2899produced by @command{gold}. 2900For other linkers that cannot generate resolution file, 2901explicit @code{externally_visible} attributes are still necessary. 2902 2903@item flatten 2904@cindex @code{flatten} function attribute 2905Generally, inlining into a function is limited. For a function marked with 2906this attribute, every call inside this function is inlined, if possible. 2907Functions declared with attribute @code{noinline} and similar are not 2908inlined. Whether the function itself is considered for inlining depends 2909on its size and the current inlining parameters. 2910 2911@item format (@var{archetype}, @var{string-index}, @var{first-to-check}) 2912@cindex @code{format} function attribute 2913@cindex functions with @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style arguments 2914@opindex Wformat 2915The @code{format} attribute specifies that a function takes @code{printf}, 2916@code{scanf}, @code{strftime} or @code{strfmon} style arguments that 2917should be type-checked against a format string. For example, the 2918declaration: 2919 2920@smallexample 2921extern int 2922my_printf (void *my_object, const char *my_format, ...) 2923 __attribute__ ((format (printf, 2, 3))); 2924@end smallexample 2925 2926@noindent 2927causes the compiler to check the arguments in calls to @code{my_printf} 2928for consistency with the @code{printf} style format string argument 2929@code{my_format}. 2930 2931The parameter @var{archetype} determines how the format string is 2932interpreted, and should be @code{printf}, @code{scanf}, @code{strftime}, 2933@code{gnu_printf}, @code{gnu_scanf}, @code{gnu_strftime} or 2934@code{strfmon}. (You can also use @code{__printf__}, 2935@code{__scanf__}, @code{__strftime__} or @code{__strfmon__}.) On 2936MinGW targets, @code{ms_printf}, @code{ms_scanf}, and 2937@code{ms_strftime} are also present. 2938@var{archetype} values such as @code{printf} refer to the formats accepted 2939by the system's C runtime library, 2940while values prefixed with @samp{gnu_} always refer 2941to the formats accepted by the GNU C Library. On Microsoft Windows 2942targets, values prefixed with @samp{ms_} refer to the formats accepted by the 2943@file{msvcrt.dll} library. 2944The parameter @var{string-index} 2945specifies which argument is the format string argument (starting 2946from 1), while @var{first-to-check} is the number of the first 2947argument to check against the format string. For functions 2948where the arguments are not available to be checked (such as 2949@code{vprintf}), specify the third parameter as zero. In this case the 2950compiler only checks the format string for consistency. For 2951@code{strftime} formats, the third parameter is required to be zero. 2952Since non-static C++ methods have an implicit @code{this} argument, the 2953arguments of such methods should be counted from two, not one, when 2954giving values for @var{string-index} and @var{first-to-check}. 2955 2956In the example above, the format string (@code{my_format}) is the second 2957argument of the function @code{my_print}, and the arguments to check 2958start with the third argument, so the correct parameters for the format 2959attribute are 2 and 3. 2960 2961@opindex ffreestanding 2962@opindex fno-builtin 2963The @code{format} attribute allows you to identify your own functions 2964that take format strings as arguments, so that GCC can check the 2965calls to these functions for errors. The compiler always (unless 2966@option{-ffreestanding} or @option{-fno-builtin} is used) checks formats 2967for the standard library functions @code{printf}, @code{fprintf}, 2968@code{sprintf}, @code{scanf}, @code{fscanf}, @code{sscanf}, @code{strftime}, 2969@code{vprintf}, @code{vfprintf} and @code{vsprintf} whenever such 2970warnings are requested (using @option{-Wformat}), so there is no need to 2971modify the header file @file{stdio.h}. In C99 mode, the functions 2972@code{snprintf}, @code{vsnprintf}, @code{vscanf}, @code{vfscanf} and 2973@code{vsscanf} are also checked. Except in strictly conforming C 2974standard modes, the X/Open function @code{strfmon} is also checked as 2975are @code{printf_unlocked} and @code{fprintf_unlocked}. 2976@xref{C Dialect Options,,Options Controlling C Dialect}. 2977 2978For Objective-C dialects, @code{NSString} (or @code{__NSString__}) is 2979recognized in the same context. Declarations including these format attributes 2980are parsed for correct syntax, however the result of checking of such format 2981strings is not yet defined, and is not carried out by this version of the 2982compiler. 2983 2984The target may also provide additional types of format checks. 2985@xref{Target Format Checks,,Format Checks Specific to Particular 2986Target Machines}. 2987 2988@item format_arg (@var{string-index}) 2989@cindex @code{format_arg} function attribute 2990@opindex Wformat-nonliteral 2991The @code{format_arg} attribute specifies that a function takes one or 2992more format strings for a @code{printf}, @code{scanf}, @code{strftime} or 2993@code{strfmon} style function and modifies it (for example, to translate 2994it into another language), so the result can be passed to a 2995@code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} style 2996function (with the remaining arguments to the format function the same 2997as they would have been for the unmodified string). Multiple 2998@code{format_arg} attributes may be applied to the same function, each 2999designating a distinct parameter as a format string. For example, the 3000declaration: 3001 3002@smallexample 3003extern char * 3004my_dgettext (char *my_domain, const char *my_format) 3005 __attribute__ ((format_arg (2))); 3006@end smallexample 3007 3008@noindent 3009causes the compiler to check the arguments in calls to a @code{printf}, 3010@code{scanf}, @code{strftime} or @code{strfmon} type function, whose 3011format string argument is a call to the @code{my_dgettext} function, for 3012consistency with the format string argument @code{my_format}. If the 3013@code{format_arg} attribute had not been specified, all the compiler 3014could tell in such calls to format functions would be that the format 3015string argument is not constant; this would generate a warning when 3016@option{-Wformat-nonliteral} is used, but the calls could not be checked 3017without the attribute. 3018 3019In calls to a function declared with more than one @code{format_arg} 3020attribute, each with a distinct argument value, the corresponding 3021actual function arguments are checked against all format strings 3022designated by the attributes. This capability is designed to support 3023the GNU @code{ngettext} family of functions. 3024 3025The parameter @var{string-index} specifies which argument is the format 3026string argument (starting from one). Since non-static C++ methods have 3027an implicit @code{this} argument, the arguments of such methods should 3028be counted from two. 3029 3030The @code{format_arg} attribute allows you to identify your own 3031functions that modify format strings, so that GCC can check the 3032calls to @code{printf}, @code{scanf}, @code{strftime} or @code{strfmon} 3033type function whose operands are a call to one of your own function. 3034The compiler always treats @code{gettext}, @code{dgettext}, and 3035@code{dcgettext} in this manner except when strict ISO C support is 3036requested by @option{-ansi} or an appropriate @option{-std} option, or 3037@option{-ffreestanding} or @option{-fno-builtin} 3038is used. @xref{C Dialect Options,,Options 3039Controlling C Dialect}. 3040 3041For Objective-C dialects, the @code{format-arg} attribute may refer to an 3042@code{NSString} reference for compatibility with the @code{format} attribute 3043above. 3044 3045The target may also allow additional types in @code{format-arg} attributes. 3046@xref{Target Format Checks,,Format Checks Specific to Particular 3047Target Machines}. 3048 3049@item gnu_inline 3050@cindex @code{gnu_inline} function attribute 3051This attribute should be used with a function that is also declared 3052with the @code{inline} keyword. It directs GCC to treat the function 3053as if it were defined in gnu90 mode even when compiling in C99 or 3054gnu99 mode. 3055 3056If the function is declared @code{extern}, then this definition of the 3057function is used only for inlining. In no case is the function 3058compiled as a standalone function, not even if you take its address 3059explicitly. Such an address becomes an external reference, as if you 3060had only declared the function, and had not defined it. This has 3061almost the effect of a macro. The way to use this is to put a 3062function definition in a header file with this attribute, and put 3063another copy of the function, without @code{extern}, in a library 3064file. The definition in the header file causes most calls to the 3065function to be inlined. If any uses of the function remain, they 3066refer to the single copy in the library. Note that the two 3067definitions of the functions need not be precisely the same, although 3068if they do not have the same effect your program may behave oddly. 3069 3070In C, if the function is neither @code{extern} nor @code{static}, then 3071the function is compiled as a standalone function, as well as being 3072inlined where possible. 3073 3074This is how GCC traditionally handled functions declared 3075@code{inline}. Since ISO C99 specifies a different semantics for 3076@code{inline}, this function attribute is provided as a transition 3077measure and as a useful feature in its own right. This attribute is 3078available in GCC 4.1.3 and later. It is available if either of the 3079preprocessor macros @code{__GNUC_GNU_INLINE__} or 3080@code{__GNUC_STDC_INLINE__} are defined. @xref{Inline,,An Inline 3081Function is As Fast As a Macro}. 3082 3083In C++, this attribute does not depend on @code{extern} in any way, 3084but it still requires the @code{inline} keyword to enable its special 3085behavior. 3086 3087@item hot 3088@cindex @code{hot} function attribute 3089The @code{hot} attribute on a function is used to inform the compiler that 3090the function is a hot spot of the compiled program. The function is 3091optimized more aggressively and on many targets it is placed into a special 3092subsection of the text section so all hot functions appear close together, 3093improving locality. 3094 3095When profile feedback is available, via @option{-fprofile-use}, hot functions 3096are automatically detected and this attribute is ignored. 3097 3098@item ifunc ("@var{resolver}") 3099@cindex @code{ifunc} function attribute 3100@cindex indirect functions 3101@cindex functions that are dynamically resolved 3102The @code{ifunc} attribute is used to mark a function as an indirect 3103function using the STT_GNU_IFUNC symbol type extension to the ELF 3104standard. This allows the resolution of the symbol value to be 3105determined dynamically at load time, and an optimized version of the 3106routine to be selected for the particular processor or other system 3107characteristics determined then. To use this attribute, first define 3108the implementation functions available, and a resolver function that 3109returns a pointer to the selected implementation function. The 3110implementation functions' declarations must match the API of the 3111function being implemented. The resolver should be declared to 3112be a function taking no arguments and returning a pointer to 3113a function of the same type as the implementation. For example: 3114 3115@smallexample 3116void *my_memcpy (void *dst, const void *src, size_t len) 3117@{ 3118 @dots{} 3119 return dst; 3120@} 3121 3122static void * (*resolve_memcpy (void))(void *, const void *, size_t) 3123@{ 3124 return my_memcpy; // we will just always select this routine 3125@} 3126@end smallexample 3127 3128@noindent 3129The exported header file declaring the function the user calls would 3130contain: 3131 3132@smallexample 3133extern void *memcpy (void *, const void *, size_t); 3134@end smallexample 3135 3136@noindent 3137allowing the user to call @code{memcpy} as a regular function, unaware of 3138the actual implementation. Finally, the indirect function needs to be 3139defined in the same translation unit as the resolver function: 3140 3141@smallexample 3142void *memcpy (void *, const void *, size_t) 3143 __attribute__ ((ifunc ("resolve_memcpy"))); 3144@end smallexample 3145 3146In C++, the @code{ifunc} attribute takes a string that is the mangled name 3147of the resolver function. A C++ resolver for a non-static member function 3148of class @code{C} should be declared to return a pointer to a non-member 3149function taking pointer to @code{C} as the first argument, followed by 3150the same arguments as of the implementation function. G++ checks 3151the signatures of the two functions and issues 3152a @option{-Wattribute-alias} warning for mismatches. To suppress a warning 3153for the necessary cast from a pointer to the implementation member function 3154to the type of the corresponding non-member function use 3155the @option{-Wno-pmf-conversions} option. For example: 3156 3157@smallexample 3158class S 3159@{ 3160private: 3161 int debug_impl (int); 3162 int optimized_impl (int); 3163 3164 typedef int Func (S*, int); 3165 3166 static Func* resolver (); 3167public: 3168 3169 int interface (int); 3170@}; 3171 3172int S::debug_impl (int) @{ /* @r{@dots{}} */ @} 3173int S::optimized_impl (int) @{ /* @r{@dots{}} */ @} 3174 3175S::Func* S::resolver () 3176@{ 3177 int (S::*pimpl) (int) 3178 = getenv ("DEBUG") ? &S::debug_impl : &S::optimized_impl; 3179 3180 // Cast triggers -Wno-pmf-conversions. 3181 return reinterpret_cast<Func*>(pimpl); 3182@} 3183 3184int S::interface (int) __attribute__ ((ifunc ("_ZN1S8resolverEv"))); 3185@end smallexample 3186 3187Indirect functions cannot be weak. Binutils version 2.20.1 or higher 3188and GNU C Library version 2.11.1 are required to use this feature. 3189 3190@item interrupt 3191@itemx interrupt_handler 3192Many GCC back ends support attributes to indicate that a function is 3193an interrupt handler, which tells the compiler to generate function 3194entry and exit sequences that differ from those from regular 3195functions. The exact syntax and behavior are target-specific; 3196refer to the following subsections for details. 3197 3198@item leaf 3199@cindex @code{leaf} function attribute 3200Calls to external functions with this attribute must return to the 3201current compilation unit only by return or by exception handling. In 3202particular, a leaf function is not allowed to invoke callback functions 3203passed to it from the current compilation unit, directly call functions 3204exported by the unit, or @code{longjmp} into the unit. Leaf functions 3205might still call functions from other compilation units and thus they 3206are not necessarily leaf in the sense that they contain no function 3207calls at all. 3208 3209The attribute is intended for library functions to improve dataflow 3210analysis. The compiler takes the hint that any data not escaping the 3211current compilation unit cannot be used or modified by the leaf 3212function. For example, the @code{sin} function is a leaf function, but 3213@code{qsort} is not. 3214 3215Note that leaf functions might indirectly run a signal handler defined 3216in the current compilation unit that uses static variables. Similarly, 3217when lazy symbol resolution is in effect, leaf functions might invoke 3218indirect functions whose resolver function or implementation function is 3219defined in the current compilation unit and uses static variables. There 3220is no standard-compliant way to write such a signal handler, resolver 3221function, or implementation function, and the best that you can do is to 3222remove the @code{leaf} attribute or mark all such static variables 3223@code{volatile}. Lastly, for ELF-based systems that support symbol 3224interposition, care should be taken that functions defined in the 3225current compilation unit do not unexpectedly interpose other symbols 3226based on the defined standards mode and defined feature test macros; 3227otherwise an inadvertent callback would be added. 3228 3229The attribute has no effect on functions defined within the current 3230compilation unit. This is to allow easy merging of multiple compilation 3231units into one, for example, by using the link-time optimization. For 3232this reason the attribute is not allowed on types to annotate indirect 3233calls. 3234 3235@item malloc 3236@item malloc (@var{deallocator}) 3237@item malloc (@var{deallocator}, @var{ptr-index}) 3238@cindex @code{malloc} function attribute 3239@cindex functions that behave like malloc 3240Attribute @code{malloc} indicates that a function is @code{malloc}-like, 3241i.e., that the pointer @var{P} returned by the function cannot alias any 3242other pointer valid when the function returns, and moreover no 3243pointers to valid objects occur in any storage addressed by @var{P}. In 3244addition, the GCC predicts that a function with the attribute returns 3245non-null in most cases. 3246 3247Independently, the form of the attribute with one or two arguments 3248associates @code{deallocator} as a suitable deallocation function for 3249pointers returned from the @code{malloc}-like function. @var{ptr-index} 3250denotes the positional argument to which when the pointer is passed in 3251calls to @code{deallocator} has the effect of deallocating it. 3252 3253Using the attribute with no arguments is designed to improve optimization 3254by relying on the aliasing property it implies. Functions like @code{malloc} 3255and @code{calloc} have this property because they return a pointer to 3256uninitialized or zeroed-out, newly obtained storage. However, functions 3257like @code{realloc} do not have this property, as they may return pointers 3258to storage containing pointers to existing objects. Additionally, since 3259all such functions are assumed to return null only infrequently, callers 3260can be optimized based on that assumption. 3261 3262Associating a function with a @var{deallocator} helps detect calls to 3263mismatched allocation and deallocation functions and diagnose them under 3264the control of options such as @option{-Wmismatched-dealloc}. It also 3265makes it possible to diagnose attempts to deallocate objects that were not 3266allocated dynamically, by @option{-Wfree-nonheap-object}. To indicate 3267that an allocation function both satisifies the nonaliasing property and 3268has a deallocator associated with it, both the plain form of the attribute 3269and the one with the @var{deallocator} argument must be used. The same 3270function can be both an allocator and a deallocator. Since inlining one 3271of the associated functions but not the other could result in apparent 3272mismatches, this form of attribute @code{malloc} is not accepted on inline 3273functions. For the same reason, using the attribute prevents both 3274the allocation and deallocation functions from being expanded inline. 3275 3276For example, besides stating that the functions return pointers that do 3277not alias any others, the following declarations make @code{fclose} 3278a suitable deallocator for pointers returned from all functions except 3279@code{popen}, and @code{pclose} as the only suitable deallocator for 3280pointers returned from @code{popen}. The deallocator functions must 3281be declared before they can be referenced in the attribute. 3282 3283@smallexample 3284int fclose (FILE*); 3285int pclose (FILE*); 3286 3287__attribute__ ((malloc, malloc (fclose, 1))) 3288 FILE* fdopen (int, const char*); 3289__attribute__ ((malloc, malloc (fclose, 1))) 3290 FILE* fopen (const char*, const char*); 3291__attribute__ ((malloc, malloc (fclose, 1))) 3292 FILE* fmemopen(void *, size_t, const char *); 3293__attribute__ ((malloc, malloc (pclose, 1))) 3294 FILE* popen (const char*, const char*); 3295__attribute__ ((malloc, malloc (fclose, 1))) 3296 FILE* tmpfile (void); 3297@end smallexample 3298 3299The warnings guarded by @option{-fanalyzer} respect allocation and 3300deallocation pairs marked with the @code{malloc}. In particular: 3301 3302@itemize @bullet 3303 3304@item 3305The analyzer will emit a @option{-Wanalyzer-mismatching-deallocation} 3306diagnostic if there is an execution path in which the result of an 3307allocation call is passed to a different deallocator. 3308 3309@item 3310The analyzer will emit a @option{-Wanalyzer-double-free} 3311diagnostic if there is an execution path in which a value is passed 3312more than once to a deallocation call. 3313 3314@item 3315The analyzer will consider the possibility that an allocation function 3316could fail and return NULL. It will emit 3317@option{-Wanalyzer-possible-null-dereference} and 3318@option{-Wanalyzer-possible-null-argument} diagnostics if there are 3319execution paths in which an unchecked result of an allocation call is 3320dereferenced or passed to a function requiring a non-null argument. 3321If the allocator always returns non-null, use 3322@code{__attribute__ ((returns_nonnull))} to suppress these warnings. 3323For example: 3324@smallexample 3325char *xstrdup (const char *) 3326 __attribute__((malloc (free), returns_nonnull)); 3327@end smallexample 3328 3329@item 3330The analyzer will emit a @option{-Wanalyzer-use-after-free} 3331diagnostic if there is an execution path in which the memory passed 3332by pointer to a deallocation call is used after the deallocation. 3333 3334@item 3335The analyzer will emit a @option{-Wanalyzer-malloc-leak} diagnostic if 3336there is an execution path in which the result of an allocation call 3337is leaked (without being passed to the deallocation function). 3338 3339@item 3340The analyzer will emit a @option{-Wanalyzer-free-of-non-heap} diagnostic 3341if a deallocation function is used on a global or on-stack variable. 3342 3343@end itemize 3344 3345The analyzer assumes that deallocators can gracefully handle the @code{NULL} 3346pointer. If this is not the case, the deallocator can be marked with 3347@code{__attribute__((nonnull))} so that @option{-fanalyzer} can emit 3348a @option{-Wanalyzer-possible-null-argument} diagnostic for code paths 3349in which the deallocator is called with NULL. 3350 3351@item no_icf 3352@cindex @code{no_icf} function attribute 3353This function attribute prevents a functions from being merged with another 3354semantically equivalent function. 3355 3356@item no_instrument_function 3357@cindex @code{no_instrument_function} function attribute 3358@opindex finstrument-functions 3359@opindex p 3360@opindex pg 3361If any of @option{-finstrument-functions}, @option{-p}, or @option{-pg} are 3362given, profiling function calls are 3363generated at entry and exit of most user-compiled functions. 3364Functions with this attribute are not so instrumented. 3365 3366@item no_profile_instrument_function 3367@cindex @code{no_profile_instrument_function} function attribute 3368The @code{no_profile_instrument_function} attribute on functions is used 3369to inform the compiler that it should not process any profile feedback based 3370optimization code instrumentation. 3371 3372@item no_reorder 3373@cindex @code{no_reorder} function attribute 3374Do not reorder functions or variables marked @code{no_reorder} 3375against each other or top level assembler statements the executable. 3376The actual order in the program will depend on the linker command 3377line. Static variables marked like this are also not removed. 3378This has a similar effect 3379as the @option{-fno-toplevel-reorder} option, but only applies to the 3380marked symbols. 3381 3382@item no_sanitize ("@var{sanitize_option}") 3383@cindex @code{no_sanitize} function attribute 3384The @code{no_sanitize} attribute on functions is used 3385to inform the compiler that it should not do sanitization of any option 3386mentioned in @var{sanitize_option}. A list of values acceptable by 3387the @option{-fsanitize} option can be provided. 3388 3389@smallexample 3390void __attribute__ ((no_sanitize ("alignment", "object-size"))) 3391f () @{ /* @r{Do something.} */; @} 3392void __attribute__ ((no_sanitize ("alignment,object-size"))) 3393g () @{ /* @r{Do something.} */; @} 3394@end smallexample 3395 3396@item no_sanitize_address 3397@itemx no_address_safety_analysis 3398@cindex @code{no_sanitize_address} function attribute 3399The @code{no_sanitize_address} attribute on functions is used 3400to inform the compiler that it should not instrument memory accesses 3401in the function when compiling with the @option{-fsanitize=address} option. 3402The @code{no_address_safety_analysis} is a deprecated alias of the 3403@code{no_sanitize_address} attribute, new code should use 3404@code{no_sanitize_address}. 3405 3406@item no_sanitize_thread 3407@cindex @code{no_sanitize_thread} function attribute 3408The @code{no_sanitize_thread} attribute on functions is used 3409to inform the compiler that it should not instrument memory accesses 3410in the function when compiling with the @option{-fsanitize=thread} option. 3411 3412@item no_sanitize_undefined 3413@cindex @code{no_sanitize_undefined} function attribute 3414The @code{no_sanitize_undefined} attribute on functions is used 3415to inform the compiler that it should not check for undefined behavior 3416in the function when compiling with the @option{-fsanitize=undefined} option. 3417 3418@item no_split_stack 3419@cindex @code{no_split_stack} function attribute 3420@opindex fsplit-stack 3421If @option{-fsplit-stack} is given, functions have a small 3422prologue which decides whether to split the stack. Functions with the 3423@code{no_split_stack} attribute do not have that prologue, and thus 3424may run with only a small amount of stack space available. 3425 3426@item no_stack_limit 3427@cindex @code{no_stack_limit} function attribute 3428This attribute locally overrides the @option{-fstack-limit-register} 3429and @option{-fstack-limit-symbol} command-line options; it has the effect 3430of disabling stack limit checking in the function it applies to. 3431 3432@item noclone 3433@cindex @code{noclone} function attribute 3434This function attribute prevents a function from being considered for 3435cloning---a mechanism that produces specialized copies of functions 3436and which is (currently) performed by interprocedural constant 3437propagation. 3438 3439@item noinline 3440@cindex @code{noinline} function attribute 3441This function attribute prevents a function from being considered for 3442inlining. 3443@c Don't enumerate the optimizations by name here; we try to be 3444@c future-compatible with this mechanism. 3445If the function does not have side effects, there are optimizations 3446other than inlining that cause function calls to be optimized away, 3447although the function call is live. To keep such calls from being 3448optimized away, put 3449@smallexample 3450asm (""); 3451@end smallexample 3452 3453@noindent 3454(@pxref{Extended Asm}) in the called function, to serve as a special 3455side effect. 3456 3457@item noipa 3458@cindex @code{noipa} function attribute 3459Disable interprocedural optimizations between the function with this 3460attribute and its callers, as if the body of the function is not available 3461when optimizing callers and the callers are unavailable when optimizing 3462the body. This attribute implies @code{noinline}, @code{noclone} and 3463@code{no_icf} attributes. However, this attribute is not equivalent 3464to a combination of other attributes, because its purpose is to suppress 3465existing and future optimizations employing interprocedural analysis, 3466including those that do not have an attribute suitable for disabling 3467them individually. This attribute is supported mainly for the purpose 3468of testing the compiler. 3469 3470@item nonnull 3471@itemx nonnull (@var{arg-index}, @dots{}) 3472@cindex @code{nonnull} function attribute 3473@cindex functions with non-null pointer arguments 3474The @code{nonnull} attribute may be applied to a function that takes at 3475least one argument of a pointer type. It indicates that the referenced 3476arguments must be non-null pointers. For instance, the declaration: 3477 3478@smallexample 3479extern void * 3480my_memcpy (void *dest, const void *src, size_t len) 3481 __attribute__((nonnull (1, 2))); 3482@end smallexample 3483 3484@noindent 3485causes the compiler to check that, in calls to @code{my_memcpy}, 3486arguments @var{dest} and @var{src} are non-null. If the compiler 3487determines that a null pointer is passed in an argument slot marked 3488as non-null, and the @option{-Wnonnull} option is enabled, a warning 3489is issued. @xref{Warning Options}. Unless disabled by 3490the @option{-fno-delete-null-pointer-checks} option the compiler may 3491also perform optimizations based on the knowledge that certain function 3492arguments cannot be null. In addition, 3493the @option{-fisolate-erroneous-paths-attribute} option can be specified 3494to have GCC transform calls with null arguments to non-null functions 3495into traps. @xref{Optimize Options}. 3496 3497If no @var{arg-index} is given to the @code{nonnull} attribute, 3498all pointer arguments are marked as non-null. To illustrate, the 3499following declaration is equivalent to the previous example: 3500 3501@smallexample 3502extern void * 3503my_memcpy (void *dest, const void *src, size_t len) 3504 __attribute__((nonnull)); 3505@end smallexample 3506 3507@item noplt 3508@cindex @code{noplt} function attribute 3509The @code{noplt} attribute is the counterpart to option @option{-fno-plt}. 3510Calls to functions marked with this attribute in position-independent code 3511do not use the PLT. 3512 3513@smallexample 3514@group 3515/* Externally defined function foo. */ 3516int foo () __attribute__ ((noplt)); 3517 3518int 3519main (/* @r{@dots{}} */) 3520@{ 3521 /* @r{@dots{}} */ 3522 foo (); 3523 /* @r{@dots{}} */ 3524@} 3525@end group 3526@end smallexample 3527 3528The @code{noplt} attribute on function @code{foo} 3529tells the compiler to assume that 3530the function @code{foo} is externally defined and that the call to 3531@code{foo} must avoid the PLT 3532in position-independent code. 3533 3534In position-dependent code, a few targets also convert calls to 3535functions that are marked to not use the PLT to use the GOT instead. 3536 3537@item noreturn 3538@cindex @code{noreturn} function attribute 3539@cindex functions that never return 3540A few standard library functions, such as @code{abort} and @code{exit}, 3541cannot return. GCC knows this automatically. Some programs define 3542their own functions that never return. You can declare them 3543@code{noreturn} to tell the compiler this fact. For example, 3544 3545@smallexample 3546@group 3547void fatal () __attribute__ ((noreturn)); 3548 3549void 3550fatal (/* @r{@dots{}} */) 3551@{ 3552 /* @r{@dots{}} */ /* @r{Print error message.} */ /* @r{@dots{}} */ 3553 exit (1); 3554@} 3555@end group 3556@end smallexample 3557 3558The @code{noreturn} keyword tells the compiler to assume that 3559@code{fatal} cannot return. It can then optimize without regard to what 3560would happen if @code{fatal} ever did return. This makes slightly 3561better code. More importantly, it helps avoid spurious warnings of 3562uninitialized variables. 3563 3564The @code{noreturn} keyword does not affect the exceptional path when that 3565applies: a @code{noreturn}-marked function may still return to the caller 3566by throwing an exception or calling @code{longjmp}. 3567 3568In order to preserve backtraces, GCC will never turn calls to 3569@code{noreturn} functions into tail calls. 3570 3571Do not assume that registers saved by the calling function are 3572restored before calling the @code{noreturn} function. 3573 3574It does not make sense for a @code{noreturn} function to have a return 3575type other than @code{void}. 3576 3577@item nothrow 3578@cindex @code{nothrow} function attribute 3579The @code{nothrow} attribute is used to inform the compiler that a 3580function cannot throw an exception. For example, most functions in 3581the standard C library can be guaranteed not to throw an exception 3582with the notable exceptions of @code{qsort} and @code{bsearch} that 3583take function pointer arguments. 3584 3585@item optimize (@var{level}, @dots{}) 3586@item optimize (@var{string}, @dots{}) 3587@cindex @code{optimize} function attribute 3588The @code{optimize} attribute is used to specify that a function is to 3589be compiled with different optimization options than specified on the 3590command line. Valid arguments are constant non-negative integers and 3591strings. Each numeric argument specifies an optimization @var{level}. 3592Each @var{string} argument consists of one or more comma-separated 3593substrings. Each substring that begins with the letter @code{O} refers 3594to an optimization option such as @option{-O0} or @option{-Os}. Other 3595substrings are taken as suffixes to the @code{-f} prefix jointly 3596forming the name of an optimization option. @xref{Optimize Options}. 3597 3598@samp{#pragma GCC optimize} can be used to set optimization options 3599for more than one function. @xref{Function Specific Option Pragmas}, 3600for details about the pragma. 3601 3602Providing multiple strings as arguments separated by commas to specify 3603multiple options is equivalent to separating the option suffixes with 3604a comma (@samp{,}) within a single string. Spaces are not permitted 3605within the strings. 3606 3607Not every optimization option that starts with the @var{-f} prefix 3608specified by the attribute necessarily has an effect on the function. 3609The @code{optimize} attribute should be used for debugging purposes only. 3610It is not suitable in production code. 3611 3612@item patchable_function_entry 3613@cindex @code{patchable_function_entry} function attribute 3614@cindex extra NOP instructions at the function entry point 3615In case the target's text segment can be made writable at run time by 3616any means, padding the function entry with a number of NOPs can be 3617used to provide a universal tool for instrumentation. 3618 3619The @code{patchable_function_entry} function attribute can be used to 3620change the number of NOPs to any desired value. The two-value syntax 3621is the same as for the command-line switch 3622@option{-fpatchable-function-entry=N,M}, generating @var{N} NOPs, with 3623the function entry point before the @var{M}th NOP instruction. 3624@var{M} defaults to 0 if omitted e.g.@: function entry point is before 3625the first NOP. 3626 3627If patchable function entries are enabled globally using the command-line 3628option @option{-fpatchable-function-entry=N,M}, then you must disable 3629instrumentation on all functions that are part of the instrumentation 3630framework with the attribute @code{patchable_function_entry (0)} 3631to prevent recursion. 3632 3633@item pure 3634@cindex @code{pure} function attribute 3635@cindex functions that have no side effects 3636 3637Calls to functions that have no observable effects on the state of 3638the program other than to return a value may lend themselves to optimizations 3639such as common subexpression elimination. Declaring such functions with 3640the @code{pure} attribute allows GCC to avoid emitting some calls in repeated 3641invocations of the function with the same argument values. 3642 3643The @code{pure} attribute prohibits a function from modifying the state 3644of the program that is observable by means other than inspecting 3645the function's return value. However, functions declared with the @code{pure} 3646attribute can safely read any non-volatile objects, and modify the value of 3647objects in a way that does not affect their return value or the observable 3648state of the program. 3649 3650For example, 3651 3652@smallexample 3653int hash (char *) __attribute__ ((pure)); 3654@end smallexample 3655 3656@noindent 3657tells GCC that subsequent calls to the function @code{hash} with the same 3658string can be replaced by the result of the first call provided the state 3659of the program observable by @code{hash}, including the contents of the array 3660itself, does not change in between. Even though @code{hash} takes a non-const 3661pointer argument it must not modify the array it points to, or any other object 3662whose value the rest of the program may depend on. However, the caller may 3663safely change the contents of the array between successive calls to 3664the function (doing so disables the optimization). The restriction also 3665applies to member objects referenced by the @code{this} pointer in C++ 3666non-static member functions. 3667 3668Some common examples of pure functions are @code{strlen} or @code{memcmp}. 3669Interesting non-pure functions are functions with infinite loops or those 3670depending on volatile memory or other system resource, that may change between 3671consecutive calls (such as the standard C @code{feof} function in 3672a multithreading environment). 3673 3674The @code{pure} attribute imposes similar but looser restrictions on 3675a function's definition than the @code{const} attribute: @code{pure} 3676allows the function to read any non-volatile memory, even if it changes 3677in between successive invocations of the function. Declaring the same 3678function with both the @code{pure} and the @code{const} attribute is 3679diagnosed. Because a pure function cannot have any observable side 3680effects it does not make sense for such a function to return @code{void}. 3681Declaring such a function is diagnosed. 3682 3683@item returns_nonnull 3684@cindex @code{returns_nonnull} function attribute 3685The @code{returns_nonnull} attribute specifies that the function 3686return value should be a non-null pointer. For instance, the declaration: 3687 3688@smallexample 3689extern void * 3690mymalloc (size_t len) __attribute__((returns_nonnull)); 3691@end smallexample 3692 3693@noindent 3694lets the compiler optimize callers based on the knowledge 3695that the return value will never be null. 3696 3697@item returns_twice 3698@cindex @code{returns_twice} function attribute 3699@cindex functions that return more than once 3700The @code{returns_twice} attribute tells the compiler that a function may 3701return more than one time. The compiler ensures that all registers 3702are dead before calling such a function and emits a warning about 3703the variables that may be clobbered after the second return from the 3704function. Examples of such functions are @code{setjmp} and @code{vfork}. 3705The @code{longjmp}-like counterpart of such function, if any, might need 3706to be marked with the @code{noreturn} attribute. 3707 3708@item section ("@var{section-name}") 3709@cindex @code{section} function attribute 3710@cindex functions in arbitrary sections 3711Normally, the compiler places the code it generates in the @code{text} section. 3712Sometimes, however, you need additional sections, or you need certain 3713particular functions to appear in special sections. The @code{section} 3714attribute specifies that a function lives in a particular section. 3715For example, the declaration: 3716 3717@smallexample 3718extern void foobar (void) __attribute__ ((section ("bar"))); 3719@end smallexample 3720 3721@noindent 3722puts the function @code{foobar} in the @code{bar} section. 3723 3724Some file formats do not support arbitrary sections so the @code{section} 3725attribute is not available on all platforms. 3726If you need to map the entire contents of a module to a particular 3727section, consider using the facilities of the linker instead. 3728 3729@item sentinel 3730@itemx sentinel (@var{position}) 3731@cindex @code{sentinel} function attribute 3732This function attribute indicates that an argument in a call to the function 3733is expected to be an explicit @code{NULL}. The attribute is only valid on 3734variadic functions. By default, the sentinel is expected to be the last 3735argument of the function call. If the optional @var{position} argument 3736is specified to the attribute, the sentinel must be located at 3737@var{position} counting backwards from the end of the argument list. 3738 3739@smallexample 3740__attribute__ ((sentinel)) 3741is equivalent to 3742__attribute__ ((sentinel(0))) 3743@end smallexample 3744 3745The attribute is automatically set with a position of 0 for the built-in 3746functions @code{execl} and @code{execlp}. The built-in function 3747@code{execle} has the attribute set with a position of 1. 3748 3749A valid @code{NULL} in this context is defined as zero with any object 3750pointer type. If your system defines the @code{NULL} macro with 3751an integer type then you need to add an explicit cast. During 3752installation GCC replaces the system @code{<stddef.h>} header with 3753a copy that redefines NULL appropriately. 3754 3755The warnings for missing or incorrect sentinels are enabled with 3756@option{-Wformat}. 3757 3758@item simd 3759@itemx simd("@var{mask}") 3760@cindex @code{simd} function attribute 3761This attribute enables creation of one or more function versions that 3762can process multiple arguments using SIMD instructions from a 3763single invocation. Specifying this attribute allows compiler to 3764assume that such versions are available at link time (provided 3765in the same or another translation unit). Generated versions are 3766target-dependent and described in the corresponding Vector ABI document. For 3767x86_64 target this document can be found 3768@w{@uref{https://sourceware.org/glibc/wiki/libmvec?action=AttachFile&do=view&target=VectorABI.txt,here}}. 3769 3770The optional argument @var{mask} may have the value 3771@code{notinbranch} or @code{inbranch}, 3772and instructs the compiler to generate non-masked or masked 3773clones correspondingly. By default, all clones are generated. 3774 3775If the attribute is specified and @code{#pragma omp declare simd} is 3776present on a declaration and the @option{-fopenmp} or @option{-fopenmp-simd} 3777switch is specified, then the attribute is ignored. 3778 3779@item stack_protect 3780@cindex @code{stack_protect} function attribute 3781This attribute adds stack protection code to the function if 3782flags @option{-fstack-protector}, @option{-fstack-protector-strong} 3783or @option{-fstack-protector-explicit} are set. 3784 3785@item no_stack_protector 3786@cindex @code{no_stack_protector} function attribute 3787This attribute prevents stack protection code for the function. 3788 3789@item target (@var{string}, @dots{}) 3790@cindex @code{target} function attribute 3791Multiple target back ends implement the @code{target} attribute 3792to specify that a function is to 3793be compiled with different target options than specified on the 3794command line. One or more strings can be provided as arguments. 3795Each string consists of one or more comma-separated suffixes to 3796the @code{-m} prefix jointly forming the name of a machine-dependent 3797option. @xref{Submodel Options,,Machine-Dependent Options}. 3798 3799The @code{target} attribute can be used for instance to have a function 3800compiled with a different ISA (instruction set architecture) than the 3801default. @samp{#pragma GCC target} can be used to specify target-specific 3802options for more than one function. @xref{Function Specific Option Pragmas}, 3803for details about the pragma. 3804 3805For instance, on an x86, you could declare one function with the 3806@code{target("sse4.1,arch=core2")} attribute and another with 3807@code{target("sse4a,arch=amdfam10")}. This is equivalent to 3808compiling the first function with @option{-msse4.1} and 3809@option{-march=core2} options, and the second function with 3810@option{-msse4a} and @option{-march=amdfam10} options. It is up to you 3811to make sure that a function is only invoked on a machine that 3812supports the particular ISA it is compiled for (for example by using 3813@code{cpuid} on x86 to determine what feature bits and architecture 3814family are used). 3815 3816@smallexample 3817int core2_func (void) __attribute__ ((__target__ ("arch=core2"))); 3818int sse3_func (void) __attribute__ ((__target__ ("sse3"))); 3819@end smallexample 3820 3821Providing multiple strings as arguments separated by commas to specify 3822multiple options is equivalent to separating the option suffixes with 3823a comma (@samp{,}) within a single string. Spaces are not permitted 3824within the strings. 3825 3826The options supported are specific to each target; refer to @ref{x86 3827Function Attributes}, @ref{PowerPC Function Attributes}, 3828@ref{ARM Function Attributes}, @ref{AArch64 Function Attributes}, 3829@ref{Nios II Function Attributes}, and @ref{S/390 Function Attributes} 3830for details. 3831 3832@item symver ("@var{name2}@@@var{nodename}") 3833@cindex @code{symver} function attribute 3834On ELF targets this attribute creates a symbol version. The @var{name2} part 3835of the parameter is the actual name of the symbol by which it will be 3836externally referenced. The @code{nodename} portion should be the name of a 3837node specified in the version script supplied to the linker when building a 3838shared library. Versioned symbol must be defined and must be exported with 3839default visibility. 3840 3841@smallexample 3842__attribute__ ((__symver__ ("foo@@VERS_1"))) int 3843foo_v1 (void) 3844@{ 3845@} 3846@end smallexample 3847 3848Will produce a @code{.symver foo_v1, foo@@VERS_1} directive in the assembler 3849output. 3850 3851One can also define multiple version for a given symbol 3852(starting from binutils 2.35). 3853 3854@smallexample 3855__attribute__ ((__symver__ ("foo@@VERS_2"), __symver__ ("foo@@VERS_3"))) 3856int symver_foo_v1 (void) 3857@{ 3858@} 3859@end smallexample 3860 3861This example creates a symbol name @code{symver_foo_v1} 3862which will be version @code{VERS_2} and @code{VERS_3} of @code{foo}. 3863 3864If you have an older release of binutils, then symbol alias needs to 3865be used: 3866 3867@smallexample 3868__attribute__ ((__symver__ ("foo@@VERS_2"))) 3869int foo_v1 (void) 3870@{ 3871 return 0; 3872@} 3873 3874__attribute__ ((__symver__ ("foo@@VERS_3"))) 3875__attribute__ ((alias ("foo_v1"))) 3876int symver_foo_v1 (void); 3877@end smallexample 3878 3879Finally if the parameter is @code{"@var{name2}@@@@@var{nodename}"} then in 3880addition to creating a symbol version (as if 3881@code{"@var{name2}@@@var{nodename}"} was used) the version will be also used 3882to resolve @var{name2} by the linker. 3883 3884@item target_clones (@var{options}) 3885@cindex @code{target_clones} function attribute 3886The @code{target_clones} attribute is used to specify that a function 3887be cloned into multiple versions compiled with different target options 3888than specified on the command line. The supported options and restrictions 3889are the same as for @code{target} attribute. 3890 3891For instance, on an x86, you could compile a function with 3892@code{target_clones("sse4.1,avx")}. GCC creates two function clones, 3893one compiled with @option{-msse4.1} and another with @option{-mavx}. 3894 3895On a PowerPC, you can compile a function with 3896@code{target_clones("cpu=power9,default")}. GCC will create two 3897function clones, one compiled with @option{-mcpu=power9} and another 3898with the default options. GCC must be configured to use GLIBC 2.23 or 3899newer in order to use the @code{target_clones} attribute. 3900 3901It also creates a resolver function (see 3902the @code{ifunc} attribute above) that dynamically selects a clone 3903suitable for current architecture. The resolver is created only if there 3904is a usage of a function with @code{target_clones} attribute. 3905 3906Note that any subsequent call of a function without @code{target_clone} 3907from a @code{target_clone} caller will not lead to copying 3908(target clone) of the called function. 3909If you want to enforce such behaviour, 3910we recommend declaring the calling function with the @code{flatten} attribute? 3911 3912@item unused 3913@cindex @code{unused} function attribute 3914This attribute, attached to a function, means that the function is meant 3915to be possibly unused. GCC does not produce a warning for this 3916function. 3917 3918@item used 3919@cindex @code{used} function attribute 3920This attribute, attached to a function, means that code must be emitted 3921for the function even if it appears that the function is not referenced. 3922This is useful, for example, when the function is referenced only in 3923inline assembly. 3924 3925When applied to a member function of a C++ class template, the 3926attribute also means that the function is instantiated if the 3927class itself is instantiated. 3928 3929@item retain 3930@cindex @code{retain} function attribute 3931For ELF targets that support the GNU or FreeBSD OSABIs, this attribute 3932will save the function from linker garbage collection. To support 3933this behavior, functions that have not been placed in specific sections 3934(e.g. by the @code{section} attribute, or the @code{-ffunction-sections} 3935option), will be placed in new, unique sections. 3936 3937This additional functionality requires Binutils version 2.36 or later. 3938 3939@item visibility ("@var{visibility_type}") 3940@cindex @code{visibility} function attribute 3941This attribute affects the linkage of the declaration to which it is attached. 3942It can be applied to variables (@pxref{Common Variable Attributes}) and types 3943(@pxref{Common Type Attributes}) as well as functions. 3944 3945There are four supported @var{visibility_type} values: default, 3946hidden, protected or internal visibility. 3947 3948@smallexample 3949void __attribute__ ((visibility ("protected"))) 3950f () @{ /* @r{Do something.} */; @} 3951int i __attribute__ ((visibility ("hidden"))); 3952@end smallexample 3953 3954The possible values of @var{visibility_type} correspond to the 3955visibility settings in the ELF gABI. 3956 3957@table @code 3958@c keep this list of visibilities in alphabetical order. 3959 3960@item default 3961Default visibility is the normal case for the object file format. 3962This value is available for the visibility attribute to override other 3963options that may change the assumed visibility of entities. 3964 3965On ELF, default visibility means that the declaration is visible to other 3966modules and, in shared libraries, means that the declared entity may be 3967overridden. 3968 3969On Darwin, default visibility means that the declaration is visible to 3970other modules. 3971 3972Default visibility corresponds to ``external linkage'' in the language. 3973 3974@item hidden 3975Hidden visibility indicates that the entity declared has a new 3976form of linkage, which we call ``hidden linkage''. Two 3977declarations of an object with hidden linkage refer to the same object 3978if they are in the same shared object. 3979 3980@item internal 3981Internal visibility is like hidden visibility, but with additional 3982processor specific semantics. Unless otherwise specified by the 3983psABI, GCC defines internal visibility to mean that a function is 3984@emph{never} called from another module. Compare this with hidden 3985functions which, while they cannot be referenced directly by other 3986modules, can be referenced indirectly via function pointers. By 3987indicating that a function cannot be called from outside the module, 3988GCC may for instance omit the load of a PIC register since it is known 3989that the calling function loaded the correct value. 3990 3991@item protected 3992Protected visibility is like default visibility except that it 3993indicates that references within the defining module bind to the 3994definition in that module. That is, the declared entity cannot be 3995overridden by another module. 3996 3997@end table 3998 3999All visibilities are supported on many, but not all, ELF targets 4000(supported when the assembler supports the @samp{.visibility} 4001pseudo-op). Default visibility is supported everywhere. Hidden 4002visibility is supported on Darwin targets. 4003 4004The visibility attribute should be applied only to declarations that 4005would otherwise have external linkage. The attribute should be applied 4006consistently, so that the same entity should not be declared with 4007different settings of the attribute. 4008 4009In C++, the visibility attribute applies to types as well as functions 4010and objects, because in C++ types have linkage. A class must not have 4011greater visibility than its non-static data member types and bases, 4012and class members default to the visibility of their class. Also, a 4013declaration without explicit visibility is limited to the visibility 4014of its type. 4015 4016In C++, you can mark member functions and static member variables of a 4017class with the visibility attribute. This is useful if you know a 4018particular method or static member variable should only be used from 4019one shared object; then you can mark it hidden while the rest of the 4020class has default visibility. Care must be taken to avoid breaking 4021the One Definition Rule; for example, it is usually not useful to mark 4022an inline method as hidden without marking the whole class as hidden. 4023 4024A C++ namespace declaration can also have the visibility attribute. 4025 4026@smallexample 4027namespace nspace1 __attribute__ ((visibility ("protected"))) 4028@{ /* @r{Do something.} */; @} 4029@end smallexample 4030 4031This attribute applies only to the particular namespace body, not to 4032other definitions of the same namespace; it is equivalent to using 4033@samp{#pragma GCC visibility} before and after the namespace 4034definition (@pxref{Visibility Pragmas}). 4035 4036In C++, if a template argument has limited visibility, this 4037restriction is implicitly propagated to the template instantiation. 4038Otherwise, template instantiations and specializations default to the 4039visibility of their template. 4040 4041If both the template and enclosing class have explicit visibility, the 4042visibility from the template is used. 4043 4044@item warn_unused_result 4045@cindex @code{warn_unused_result} function attribute 4046The @code{warn_unused_result} attribute causes a warning to be emitted 4047if a caller of the function with this attribute does not use its 4048return value. This is useful for functions where not checking 4049the result is either a security problem or always a bug, such as 4050@code{realloc}. 4051 4052@smallexample 4053int fn () __attribute__ ((warn_unused_result)); 4054int foo () 4055@{ 4056 if (fn () < 0) return -1; 4057 fn (); 4058 return 0; 4059@} 4060@end smallexample 4061 4062@noindent 4063results in warning on line 5. 4064 4065@item weak 4066@cindex @code{weak} function attribute 4067The @code{weak} attribute causes a declaration of an external symbol 4068to be emitted as a weak symbol rather than a global. This is primarily 4069useful in defining library functions that can be overridden in user code, 4070though it can also be used with non-function declarations. The overriding 4071symbol must have the same type as the weak symbol. In addition, if it 4072designates a variable it must also have the same size and alignment as 4073the weak symbol. Weak symbols are supported for ELF targets, and also 4074for a.out targets when using the GNU assembler and linker. 4075 4076@item weakref 4077@itemx weakref ("@var{target}") 4078@cindex @code{weakref} function attribute 4079The @code{weakref} attribute marks a declaration as a weak reference. 4080Without arguments, it should be accompanied by an @code{alias} attribute 4081naming the target symbol. Alternatively, @var{target} may be given as 4082an argument to @code{weakref} itself, naming the target definition of 4083the alias. The @var{target} must have the same type as the declaration. 4084In addition, if it designates a variable it must also have the same size 4085and alignment as the declaration. In either form of the declaration 4086@code{weakref} implicitly marks the declared symbol as @code{weak}. Without 4087a @var{target} given as an argument to @code{weakref} or to @code{alias}, 4088@code{weakref} is equivalent to @code{weak} (in that case the declaration 4089may be @code{extern}). 4090 4091@smallexample 4092/* Given the declaration: */ 4093extern int y (void); 4094 4095/* the following... */ 4096static int x (void) __attribute__ ((weakref ("y"))); 4097 4098/* is equivalent to... */ 4099static int x (void) __attribute__ ((weakref, alias ("y"))); 4100 4101/* or, alternatively, to... */ 4102static int x (void) __attribute__ ((weakref)); 4103static int x (void) __attribute__ ((alias ("y"))); 4104@end smallexample 4105 4106A weak reference is an alias that does not by itself require a 4107definition to be given for the target symbol. If the target symbol is 4108only referenced through weak references, then it becomes a @code{weak} 4109undefined symbol. If it is directly referenced, however, then such 4110strong references prevail, and a definition is required for the 4111symbol, not necessarily in the same translation unit. 4112 4113The effect is equivalent to moving all references to the alias to a 4114separate translation unit, renaming the alias to the aliased symbol, 4115declaring it as weak, compiling the two separate translation units and 4116performing a link with relocatable output (i.e.@: @code{ld -r}) on them. 4117 4118A declaration to which @code{weakref} is attached and that is associated 4119with a named @code{target} must be @code{static}. 4120 4121@item zero_call_used_regs ("@var{choice}") 4122@cindex @code{zero_call_used_regs} function attribute 4123 4124The @code{zero_call_used_regs} attribute causes the compiler to zero 4125a subset of all call-used registers@footnote{A ``call-used'' register 4126is a register whose contents can be changed by a function call; 4127therefore, a caller cannot assume that the register has the same contents 4128on return from the function as it had before calling the function. Such 4129registers are also called ``call-clobbered'', ``caller-saved'', or 4130``volatile''.} at function return. 4131This is used to increase program security by either mitigating 4132Return-Oriented Programming (ROP) attacks or preventing information leakage 4133through registers. 4134 4135In order to satisfy users with different security needs and control the 4136run-time overhead at the same time, the @var{choice} parameter provides a 4137flexible way to choose the subset of the call-used registers to be zeroed. 4138The three basic values of @var{choice} are: 4139 4140@itemize @bullet 4141@item 4142@samp{skip} doesn't zero any call-used registers. 4143 4144@item 4145@samp{used} only zeros call-used registers that are used in the function. 4146A ``used'' register is one whose content has been set or referenced in 4147the function. 4148 4149@item 4150@samp{all} zeros all call-used registers. 4151@end itemize 4152 4153In addition to these three basic choices, it is possible to modify 4154@samp{used} or @samp{all} as follows: 4155 4156@itemize @bullet 4157@item 4158Adding @samp{-gpr} restricts the zeroing to general-purpose registers. 4159 4160@item 4161Adding @samp{-arg} restricts the zeroing to registers that can sometimes 4162be used to pass function arguments. This includes all argument registers 4163defined by the platform's calling conversion, regardless of whether the 4164function uses those registers for function arguments or not. 4165@end itemize 4166 4167The modifiers can be used individually or together. If they are used 4168together, they must appear in the order above. 4169 4170The full list of @var{choice}s is therefore: 4171 4172@table @code 4173@item skip 4174doesn't zero any call-used register. 4175 4176@item used 4177only zeros call-used registers that are used in the function. 4178 4179@item used-gpr 4180only zeros call-used general purpose registers that are used in the function. 4181 4182@item used-arg 4183only zeros call-used registers that are used in the function and pass arguments. 4184 4185@item used-gpr-arg 4186only zeros call-used general purpose registers that are used in the function 4187and pass arguments. 4188 4189@item all 4190zeros all call-used registers. 4191 4192@item all-gpr 4193zeros all call-used general purpose registers. 4194 4195@item all-arg 4196zeros all call-used registers that pass arguments. 4197 4198@item all-gpr-arg 4199zeros all call-used general purpose registers that pass 4200arguments. 4201@end table 4202 4203Of this list, @samp{used-arg}, @samp{used-gpr-arg}, @samp{all-arg}, 4204and @samp{all-gpr-arg} are mainly used for ROP mitigation. 4205 4206The default for the attribute is controlled by @option{-fzero-call-used-regs}. 4207@end table 4208 4209@c This is the end of the target-independent attribute table 4210 4211@node AArch64 Function Attributes 4212@subsection AArch64 Function Attributes 4213 4214The following target-specific function attributes are available for the 4215AArch64 target. For the most part, these options mirror the behavior of 4216similar command-line options (@pxref{AArch64 Options}), but on a 4217per-function basis. 4218 4219@table @code 4220@item general-regs-only 4221@cindex @code{general-regs-only} function attribute, AArch64 4222Indicates that no floating-point or Advanced SIMD registers should be 4223used when generating code for this function. If the function explicitly 4224uses floating-point code, then the compiler gives an error. This is 4225the same behavior as that of the command-line option 4226@option{-mgeneral-regs-only}. 4227 4228@item fix-cortex-a53-835769 4229@cindex @code{fix-cortex-a53-835769} function attribute, AArch64 4230Indicates that the workaround for the Cortex-A53 erratum 835769 should be 4231applied to this function. To explicitly disable the workaround for this 4232function specify the negated form: @code{no-fix-cortex-a53-835769}. 4233This corresponds to the behavior of the command line options 4234@option{-mfix-cortex-a53-835769} and @option{-mno-fix-cortex-a53-835769}. 4235 4236@item cmodel= 4237@cindex @code{cmodel=} function attribute, AArch64 4238Indicates that code should be generated for a particular code model for 4239this function. The behavior and permissible arguments are the same as 4240for the command line option @option{-mcmodel=}. 4241 4242@item strict-align 4243@itemx no-strict-align 4244@cindex @code{strict-align} function attribute, AArch64 4245@code{strict-align} indicates that the compiler should not assume that unaligned 4246memory references are handled by the system. To allow the compiler to assume 4247that aligned memory references are handled by the system, the inverse attribute 4248@code{no-strict-align} can be specified. The behavior is same as for the 4249command-line option @option{-mstrict-align} and @option{-mno-strict-align}. 4250 4251@item omit-leaf-frame-pointer 4252@cindex @code{omit-leaf-frame-pointer} function attribute, AArch64 4253Indicates that the frame pointer should be omitted for a leaf function call. 4254To keep the frame pointer, the inverse attribute 4255@code{no-omit-leaf-frame-pointer} can be specified. These attributes have 4256the same behavior as the command-line options @option{-momit-leaf-frame-pointer} 4257and @option{-mno-omit-leaf-frame-pointer}. 4258 4259@item tls-dialect= 4260@cindex @code{tls-dialect=} function attribute, AArch64 4261Specifies the TLS dialect to use for this function. The behavior and 4262permissible arguments are the same as for the command-line option 4263@option{-mtls-dialect=}. 4264 4265@item arch= 4266@cindex @code{arch=} function attribute, AArch64 4267Specifies the architecture version and architectural extensions to use 4268for this function. The behavior and permissible arguments are the same as 4269for the @option{-march=} command-line option. 4270 4271@item tune= 4272@cindex @code{tune=} function attribute, AArch64 4273Specifies the core for which to tune the performance of this function. 4274The behavior and permissible arguments are the same as for the @option{-mtune=} 4275command-line option. 4276 4277@item cpu= 4278@cindex @code{cpu=} function attribute, AArch64 4279Specifies the core for which to tune the performance of this function and also 4280whose architectural features to use. The behavior and valid arguments are the 4281same as for the @option{-mcpu=} command-line option. 4282 4283@item sign-return-address 4284@cindex @code{sign-return-address} function attribute, AArch64 4285Select the function scope on which return address signing will be applied. The 4286behavior and permissible arguments are the same as for the command-line option 4287@option{-msign-return-address=}. The default value is @code{none}. This 4288attribute is deprecated. The @code{branch-protection} attribute should 4289be used instead. 4290 4291@item branch-protection 4292@cindex @code{branch-protection} function attribute, AArch64 4293Select the function scope on which branch protection will be applied. The 4294behavior and permissible arguments are the same as for the command-line option 4295@option{-mbranch-protection=}. The default value is @code{none}. 4296 4297@item outline-atomics 4298@cindex @code{outline-atomics} function attribute, AArch64 4299Enable or disable calls to out-of-line helpers to implement atomic operations. 4300This corresponds to the behavior of the command line options 4301@option{-moutline-atomics} and @option{-mno-outline-atomics}. 4302 4303@end table 4304 4305The above target attributes can be specified as follows: 4306 4307@smallexample 4308__attribute__((target("@var{attr-string}"))) 4309int 4310f (int a) 4311@{ 4312 return a + 5; 4313@} 4314@end smallexample 4315 4316where @code{@var{attr-string}} is one of the attribute strings specified above. 4317 4318Additionally, the architectural extension string may be specified on its 4319own. This can be used to turn on and off particular architectural extensions 4320without having to specify a particular architecture version or core. Example: 4321 4322@smallexample 4323__attribute__((target("+crc+nocrypto"))) 4324int 4325foo (int a) 4326@{ 4327 return a + 5; 4328@} 4329@end smallexample 4330 4331In this example @code{target("+crc+nocrypto")} enables the @code{crc} 4332extension and disables the @code{crypto} extension for the function @code{foo} 4333without modifying an existing @option{-march=} or @option{-mcpu} option. 4334 4335Multiple target function attributes can be specified by separating them with 4336a comma. For example: 4337@smallexample 4338__attribute__((target("arch=armv8-a+crc+crypto,tune=cortex-a53"))) 4339int 4340foo (int a) 4341@{ 4342 return a + 5; 4343@} 4344@end smallexample 4345 4346is valid and compiles function @code{foo} for ARMv8-A with @code{crc} 4347and @code{crypto} extensions and tunes it for @code{cortex-a53}. 4348 4349@subsubsection Inlining rules 4350Specifying target attributes on individual functions or performing link-time 4351optimization across translation units compiled with different target options 4352can affect function inlining rules: 4353 4354In particular, a caller function can inline a callee function only if the 4355architectural features available to the callee are a subset of the features 4356available to the caller. 4357For example: A function @code{foo} compiled with @option{-march=armv8-a+crc}, 4358or tagged with the equivalent @code{arch=armv8-a+crc} attribute, 4359can inline a function @code{bar} compiled with @option{-march=armv8-a+nocrc} 4360because the all the architectural features that function @code{bar} requires 4361are available to function @code{foo}. Conversely, function @code{bar} cannot 4362inline function @code{foo}. 4363 4364Additionally inlining a function compiled with @option{-mstrict-align} into a 4365function compiled without @code{-mstrict-align} is not allowed. 4366However, inlining a function compiled without @option{-mstrict-align} into a 4367function compiled with @option{-mstrict-align} is allowed. 4368 4369Note that CPU tuning options and attributes such as the @option{-mcpu=}, 4370@option{-mtune=} do not inhibit inlining unless the CPU specified by the 4371@option{-mcpu=} option or the @code{cpu=} attribute conflicts with the 4372architectural feature rules specified above. 4373 4374@node AMD GCN Function Attributes 4375@subsection AMD GCN Function Attributes 4376 4377These function attributes are supported by the AMD GCN back end: 4378 4379@table @code 4380@item amdgpu_hsa_kernel 4381@cindex @code{amdgpu_hsa_kernel} function attribute, AMD GCN 4382This attribute indicates that the corresponding function should be compiled as 4383a kernel function, that is an entry point that can be invoked from the host 4384via the HSA runtime library. By default functions are only callable only from 4385other GCN functions. 4386 4387This attribute is implicitly applied to any function named @code{main}, using 4388default parameters. 4389 4390Kernel functions may return an integer value, which will be written to a 4391conventional place within the HSA "kernargs" region. 4392 4393The attribute parameters configure what values are passed into the kernel 4394function by the GPU drivers, via the initial register state. Some values are 4395used by the compiler, and therefore forced on. Enabling other options may 4396break assumptions in the compiler and/or run-time libraries. 4397 4398@table @code 4399@item private_segment_buffer 4400Set @code{enable_sgpr_private_segment_buffer} flag. Always on (required to 4401locate the stack). 4402 4403@item dispatch_ptr 4404Set @code{enable_sgpr_dispatch_ptr} flag. Always on (required to locate the 4405launch dimensions). 4406 4407@item queue_ptr 4408Set @code{enable_sgpr_queue_ptr} flag. Always on (required to convert address 4409spaces). 4410 4411@item kernarg_segment_ptr 4412Set @code{enable_sgpr_kernarg_segment_ptr} flag. Always on (required to 4413locate the kernel arguments, "kernargs"). 4414 4415@item dispatch_id 4416Set @code{enable_sgpr_dispatch_id} flag. 4417 4418@item flat_scratch_init 4419Set @code{enable_sgpr_flat_scratch_init} flag. 4420 4421@item private_segment_size 4422Set @code{enable_sgpr_private_segment_size} flag. 4423 4424@item grid_workgroup_count_X 4425Set @code{enable_sgpr_grid_workgroup_count_x} flag. Always on (required to 4426use OpenACC/OpenMP). 4427 4428@item grid_workgroup_count_Y 4429Set @code{enable_sgpr_grid_workgroup_count_y} flag. 4430 4431@item grid_workgroup_count_Z 4432Set @code{enable_sgpr_grid_workgroup_count_z} flag. 4433 4434@item workgroup_id_X 4435Set @code{enable_sgpr_workgroup_id_x} flag. 4436 4437@item workgroup_id_Y 4438Set @code{enable_sgpr_workgroup_id_y} flag. 4439 4440@item workgroup_id_Z 4441Set @code{enable_sgpr_workgroup_id_z} flag. 4442 4443@item workgroup_info 4444Set @code{enable_sgpr_workgroup_info} flag. 4445 4446@item private_segment_wave_offset 4447Set @code{enable_sgpr_private_segment_wave_byte_offset} flag. Always on 4448(required to locate the stack). 4449 4450@item work_item_id_X 4451Set @code{enable_vgpr_workitem_id} parameter. Always on (can't be disabled). 4452 4453@item work_item_id_Y 4454Set @code{enable_vgpr_workitem_id} parameter. Always on (required to enable 4455vectorization.) 4456 4457@item work_item_id_Z 4458Set @code{enable_vgpr_workitem_id} parameter. Always on (required to use 4459OpenACC/OpenMP). 4460 4461@end table 4462@end table 4463 4464@node ARC Function Attributes 4465@subsection ARC Function Attributes 4466 4467These function attributes are supported by the ARC back end: 4468 4469@table @code 4470@item interrupt 4471@cindex @code{interrupt} function attribute, ARC 4472Use this attribute to indicate 4473that the specified function is an interrupt handler. The compiler generates 4474function entry and exit sequences suitable for use in an interrupt handler 4475when this attribute is present. 4476 4477On the ARC, you must specify the kind of interrupt to be handled 4478in a parameter to the interrupt attribute like this: 4479 4480@smallexample 4481void f () __attribute__ ((interrupt ("ilink1"))); 4482@end smallexample 4483 4484Permissible values for this parameter are: @w{@code{ilink1}} and 4485@w{@code{ilink2}} for ARCv1 architecture, and @w{@code{ilink}} and 4486@w{@code{firq}} for ARCv2 architecture. 4487 4488@item long_call 4489@itemx medium_call 4490@itemx short_call 4491@cindex @code{long_call} function attribute, ARC 4492@cindex @code{medium_call} function attribute, ARC 4493@cindex @code{short_call} function attribute, ARC 4494@cindex indirect calls, ARC 4495These attributes specify how a particular function is called. 4496These attributes override the 4497@option{-mlong-calls} and @option{-mmedium-calls} (@pxref{ARC Options}) 4498command-line switches and @code{#pragma long_calls} settings. 4499 4500For ARC, a function marked with the @code{long_call} attribute is 4501always called using register-indirect jump-and-link instructions, 4502thereby enabling the called function to be placed anywhere within the 450332-bit address space. A function marked with the @code{medium_call} 4504attribute will always be close enough to be called with an unconditional 4505branch-and-link instruction, which has a 25-bit offset from 4506the call site. A function marked with the @code{short_call} 4507attribute will always be close enough to be called with a conditional 4508branch-and-link instruction, which has a 21-bit offset from 4509the call site. 4510 4511@item jli_always 4512@cindex @code{jli_always} function attribute, ARC 4513Forces a particular function to be called using @code{jli} 4514instruction. The @code{jli} instruction makes use of a table stored 4515into @code{.jlitab} section, which holds the location of the functions 4516which are addressed using this instruction. 4517 4518@item jli_fixed 4519@cindex @code{jli_fixed} function attribute, ARC 4520Identical like the above one, but the location of the function in the 4521@code{jli} table is known and given as an attribute parameter. 4522 4523@item secure_call 4524@cindex @code{secure_call} function attribute, ARC 4525This attribute allows one to mark secure-code functions that are 4526callable from normal mode. The location of the secure call function 4527into the @code{sjli} table needs to be passed as argument. 4528 4529@item naked 4530@cindex @code{naked} function attribute, ARC 4531This attribute allows the compiler to construct the requisite function 4532declaration, while allowing the body of the function to be assembly 4533code. The specified function will not have prologue/epilogue 4534sequences generated by the compiler. Only basic @code{asm} statements 4535can safely be included in naked functions (@pxref{Basic Asm}). While 4536using extended @code{asm} or a mixture of basic @code{asm} and C code 4537may appear to work, they cannot be depended upon to work reliably and 4538are not supported. 4539 4540@end table 4541 4542@node ARM Function Attributes 4543@subsection ARM Function Attributes 4544 4545These function attributes are supported for ARM targets: 4546 4547@table @code 4548 4549@item general-regs-only 4550@cindex @code{general-regs-only} function attribute, ARM 4551Indicates that no floating-point or Advanced SIMD registers should be 4552used when generating code for this function. If the function explicitly 4553uses floating-point code, then the compiler gives an error. This is 4554the same behavior as that of the command-line option 4555@option{-mgeneral-regs-only}. 4556 4557@item interrupt 4558@cindex @code{interrupt} function attribute, ARM 4559Use this attribute to indicate 4560that the specified function is an interrupt handler. The compiler generates 4561function entry and exit sequences suitable for use in an interrupt handler 4562when this attribute is present. 4563 4564You can specify the kind of interrupt to be handled by 4565adding an optional parameter to the interrupt attribute like this: 4566 4567@smallexample 4568void f () __attribute__ ((interrupt ("IRQ"))); 4569@end smallexample 4570 4571@noindent 4572Permissible values for this parameter are: @code{IRQ}, @code{FIQ}, 4573@code{SWI}, @code{ABORT} and @code{UNDEF}. 4574 4575On ARMv7-M the interrupt type is ignored, and the attribute means the function 4576may be called with a word-aligned stack pointer. 4577 4578@item isr 4579@cindex @code{isr} function attribute, ARM 4580Use this attribute on ARM to write Interrupt Service Routines. This is an 4581alias to the @code{interrupt} attribute above. 4582 4583@item long_call 4584@itemx short_call 4585@cindex @code{long_call} function attribute, ARM 4586@cindex @code{short_call} function attribute, ARM 4587@cindex indirect calls, ARM 4588These attributes specify how a particular function is called. 4589These attributes override the 4590@option{-mlong-calls} (@pxref{ARM Options}) 4591command-line switch and @code{#pragma long_calls} settings. For ARM, the 4592@code{long_call} attribute indicates that the function might be far 4593away from the call site and require a different (more expensive) 4594calling sequence. The @code{short_call} attribute always places 4595the offset to the function from the call site into the @samp{BL} 4596instruction directly. 4597 4598@item naked 4599@cindex @code{naked} function attribute, ARM 4600This attribute allows the compiler to construct the 4601requisite function declaration, while allowing the body of the 4602function to be assembly code. The specified function will not have 4603prologue/epilogue sequences generated by the compiler. Only basic 4604@code{asm} statements can safely be included in naked functions 4605(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 4606basic @code{asm} and C code may appear to work, they cannot be 4607depended upon to work reliably and are not supported. 4608 4609@item pcs 4610@cindex @code{pcs} function attribute, ARM 4611 4612The @code{pcs} attribute can be used to control the calling convention 4613used for a function on ARM. The attribute takes an argument that specifies 4614the calling convention to use. 4615 4616When compiling using the AAPCS ABI (or a variant of it) then valid 4617values for the argument are @code{"aapcs"} and @code{"aapcs-vfp"}. In 4618order to use a variant other than @code{"aapcs"} then the compiler must 4619be permitted to use the appropriate co-processor registers (i.e., the 4620VFP registers must be available in order to use @code{"aapcs-vfp"}). 4621For example, 4622 4623@smallexample 4624/* Argument passed in r0, and result returned in r0+r1. */ 4625double f2d (float) __attribute__((pcs("aapcs"))); 4626@end smallexample 4627 4628Variadic functions always use the @code{"aapcs"} calling convention and 4629the compiler rejects attempts to specify an alternative. 4630 4631@item target (@var{options}) 4632@cindex @code{target} function attribute 4633As discussed in @ref{Common Function Attributes}, this attribute 4634allows specification of target-specific compilation options. 4635 4636On ARM, the following options are allowed: 4637 4638@table @samp 4639@item thumb 4640@cindex @code{target("thumb")} function attribute, ARM 4641Force code generation in the Thumb (T16/T32) ISA, depending on the 4642architecture level. 4643 4644@item arm 4645@cindex @code{target("arm")} function attribute, ARM 4646Force code generation in the ARM (A32) ISA. 4647 4648Functions from different modes can be inlined in the caller's mode. 4649 4650@item fpu= 4651@cindex @code{target("fpu=")} function attribute, ARM 4652Specifies the fpu for which to tune the performance of this function. 4653The behavior and permissible arguments are the same as for the @option{-mfpu=} 4654command-line option. 4655 4656@item arch= 4657@cindex @code{arch=} function attribute, ARM 4658Specifies the architecture version and architectural extensions to use 4659for this function. The behavior and permissible arguments are the same as 4660for the @option{-march=} command-line option. 4661 4662The above target attributes can be specified as follows: 4663 4664@smallexample 4665__attribute__((target("arch=armv8-a+crc"))) 4666int 4667f (int a) 4668@{ 4669 return a + 5; 4670@} 4671@end smallexample 4672 4673Additionally, the architectural extension string may be specified on its 4674own. This can be used to turn on and off particular architectural extensions 4675without having to specify a particular architecture version or core. Example: 4676 4677@smallexample 4678__attribute__((target("+crc+nocrypto"))) 4679int 4680foo (int a) 4681@{ 4682 return a + 5; 4683@} 4684@end smallexample 4685 4686In this example @code{target("+crc+nocrypto")} enables the @code{crc} 4687extension and disables the @code{crypto} extension for the function @code{foo} 4688without modifying an existing @option{-march=} or @option{-mcpu} option. 4689 4690@end table 4691 4692@end table 4693 4694@node AVR Function Attributes 4695@subsection AVR Function Attributes 4696 4697These function attributes are supported by the AVR back end: 4698 4699@table @code 4700@item interrupt 4701@cindex @code{interrupt} function attribute, AVR 4702Use this attribute to indicate 4703that the specified function is an interrupt handler. The compiler generates 4704function entry and exit sequences suitable for use in an interrupt handler 4705when this attribute is present. 4706 4707On the AVR, the hardware globally disables interrupts when an 4708interrupt is executed. The first instruction of an interrupt handler 4709declared with this attribute is a @code{SEI} instruction to 4710re-enable interrupts. See also the @code{signal} function attribute 4711that does not insert a @code{SEI} instruction. If both @code{signal} and 4712@code{interrupt} are specified for the same function, @code{signal} 4713is silently ignored. 4714 4715@item naked 4716@cindex @code{naked} function attribute, AVR 4717This attribute allows the compiler to construct the 4718requisite function declaration, while allowing the body of the 4719function to be assembly code. The specified function will not have 4720prologue/epilogue sequences generated by the compiler. Only basic 4721@code{asm} statements can safely be included in naked functions 4722(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 4723basic @code{asm} and C code may appear to work, they cannot be 4724depended upon to work reliably and are not supported. 4725 4726@item no_gccisr 4727@cindex @code{no_gccisr} function attribute, AVR 4728Do not use @code{__gcc_isr} pseudo instructions in a function with 4729the @code{interrupt} or @code{signal} attribute aka. interrupt 4730service routine (ISR). 4731Use this attribute if the preamble of the ISR prologue should always read 4732@example 4733push __zero_reg__ 4734push __tmp_reg__ 4735in __tmp_reg__, __SREG__ 4736push __tmp_reg__ 4737clr __zero_reg__ 4738@end example 4739and accordingly for the postamble of the epilogue --- no matter whether 4740the mentioned registers are actually used in the ISR or not. 4741Situations where you might want to use this attribute include: 4742@itemize @bullet 4743@item 4744Code that (effectively) clobbers bits of @code{SREG} other than the 4745@code{I}-flag by writing to the memory location of @code{SREG}. 4746@item 4747Code that uses inline assembler to jump to a different function which 4748expects (parts of) the prologue code as outlined above to be present. 4749@end itemize 4750To disable @code{__gcc_isr} generation for the whole compilation unit, 4751there is option @option{-mno-gas-isr-prologues}, @pxref{AVR Options}. 4752 4753@item OS_main 4754@itemx OS_task 4755@cindex @code{OS_main} function attribute, AVR 4756@cindex @code{OS_task} function attribute, AVR 4757On AVR, functions with the @code{OS_main} or @code{OS_task} attribute 4758do not save/restore any call-saved register in their prologue/epilogue. 4759 4760The @code{OS_main} attribute can be used when there @emph{is 4761guarantee} that interrupts are disabled at the time when the function 4762is entered. This saves resources when the stack pointer has to be 4763changed to set up a frame for local variables. 4764 4765The @code{OS_task} attribute can be used when there is @emph{no 4766guarantee} that interrupts are disabled at that time when the function 4767is entered like for, e@.g@. task functions in a multi-threading operating 4768system. In that case, changing the stack pointer register is 4769guarded by save/clear/restore of the global interrupt enable flag. 4770 4771The differences to the @code{naked} function attribute are: 4772@itemize @bullet 4773@item @code{naked} functions do not have a return instruction whereas 4774@code{OS_main} and @code{OS_task} functions have a @code{RET} or 4775@code{RETI} return instruction. 4776@item @code{naked} functions do not set up a frame for local variables 4777or a frame pointer whereas @code{OS_main} and @code{OS_task} do this 4778as needed. 4779@end itemize 4780 4781@item signal 4782@cindex @code{signal} function attribute, AVR 4783Use this attribute on the AVR to indicate that the specified 4784function is an interrupt handler. The compiler generates function 4785entry and exit sequences suitable for use in an interrupt handler when this 4786attribute is present. 4787 4788See also the @code{interrupt} function attribute. 4789 4790The AVR hardware globally disables interrupts when an interrupt is executed. 4791Interrupt handler functions defined with the @code{signal} attribute 4792do not re-enable interrupts. It is save to enable interrupts in a 4793@code{signal} handler. This ``save'' only applies to the code 4794generated by the compiler and not to the IRQ layout of the 4795application which is responsibility of the application. 4796 4797If both @code{signal} and @code{interrupt} are specified for the same 4798function, @code{signal} is silently ignored. 4799@end table 4800 4801@node Blackfin Function Attributes 4802@subsection Blackfin Function Attributes 4803 4804These function attributes are supported by the Blackfin back end: 4805 4806@table @code 4807 4808@item exception_handler 4809@cindex @code{exception_handler} function attribute 4810@cindex exception handler functions, Blackfin 4811Use this attribute on the Blackfin to indicate that the specified function 4812is an exception handler. The compiler generates function entry and 4813exit sequences suitable for use in an exception handler when this 4814attribute is present. 4815 4816@item interrupt_handler 4817@cindex @code{interrupt_handler} function attribute, Blackfin 4818Use this attribute to 4819indicate that the specified function is an interrupt handler. The compiler 4820generates function entry and exit sequences suitable for use in an 4821interrupt handler when this attribute is present. 4822 4823@item kspisusp 4824@cindex @code{kspisusp} function attribute, Blackfin 4825@cindex User stack pointer in interrupts on the Blackfin 4826When used together with @code{interrupt_handler}, @code{exception_handler} 4827or @code{nmi_handler}, code is generated to load the stack pointer 4828from the USP register in the function prologue. 4829 4830@item l1_text 4831@cindex @code{l1_text} function attribute, Blackfin 4832This attribute specifies a function to be placed into L1 Instruction 4833SRAM@. The function is put into a specific section named @code{.l1.text}. 4834With @option{-mfdpic}, function calls with a such function as the callee 4835or caller uses inlined PLT. 4836 4837@item l2 4838@cindex @code{l2} function attribute, Blackfin 4839This attribute specifies a function to be placed into L2 4840SRAM. The function is put into a specific section named 4841@code{.l2.text}. With @option{-mfdpic}, callers of such functions use 4842an inlined PLT. 4843 4844@item longcall 4845@itemx shortcall 4846@cindex indirect calls, Blackfin 4847@cindex @code{longcall} function attribute, Blackfin 4848@cindex @code{shortcall} function attribute, Blackfin 4849The @code{longcall} attribute 4850indicates that the function might be far away from the call site and 4851require a different (more expensive) calling sequence. The 4852@code{shortcall} attribute indicates that the function is always close 4853enough for the shorter calling sequence to be used. These attributes 4854override the @option{-mlongcall} switch. 4855 4856@item nesting 4857@cindex @code{nesting} function attribute, Blackfin 4858@cindex Allow nesting in an interrupt handler on the Blackfin processor 4859Use this attribute together with @code{interrupt_handler}, 4860@code{exception_handler} or @code{nmi_handler} to indicate that the function 4861entry code should enable nested interrupts or exceptions. 4862 4863@item nmi_handler 4864@cindex @code{nmi_handler} function attribute, Blackfin 4865@cindex NMI handler functions on the Blackfin processor 4866Use this attribute on the Blackfin to indicate that the specified function 4867is an NMI handler. The compiler generates function entry and 4868exit sequences suitable for use in an NMI handler when this 4869attribute is present. 4870 4871@item saveall 4872@cindex @code{saveall} function attribute, Blackfin 4873@cindex save all registers on the Blackfin 4874Use this attribute to indicate that 4875all registers except the stack pointer should be saved in the prologue 4876regardless of whether they are used or not. 4877@end table 4878 4879@node BPF Function Attributes 4880@subsection BPF Function Attributes 4881 4882These function attributes are supported by the BPF back end: 4883 4884@table @code 4885@item kernel_helper 4886@cindex @code{kernel helper}, function attribute, BPF 4887use this attribute to indicate the specified function declaration is a 4888kernel helper. The helper function is passed as an argument to the 4889attribute. Example: 4890 4891@smallexample 4892int bpf_probe_read (void *dst, int size, const void *unsafe_ptr) 4893 __attribute__ ((kernel_helper (4))); 4894@end smallexample 4895@end table 4896 4897@node CR16 Function Attributes 4898@subsection CR16 Function Attributes 4899 4900These function attributes are supported by the CR16 back end: 4901 4902@table @code 4903@item interrupt 4904@cindex @code{interrupt} function attribute, CR16 4905Use this attribute to indicate 4906that the specified function is an interrupt handler. The compiler generates 4907function entry and exit sequences suitable for use in an interrupt handler 4908when this attribute is present. 4909@end table 4910 4911@node C-SKY Function Attributes 4912@subsection C-SKY Function Attributes 4913 4914These function attributes are supported by the C-SKY back end: 4915 4916@table @code 4917@item interrupt 4918@itemx isr 4919@cindex @code{interrupt} function attribute, C-SKY 4920@cindex @code{isr} function attribute, C-SKY 4921Use these attributes to indicate that the specified function 4922is an interrupt handler. 4923The compiler generates function entry and exit sequences suitable for 4924use in an interrupt handler when either of these attributes are present. 4925 4926Use of these options requires the @option{-mistack} command-line option 4927to enable support for the necessary interrupt stack instructions. They 4928are ignored with a warning otherwise. @xref{C-SKY Options}. 4929 4930@item naked 4931@cindex @code{naked} function attribute, C-SKY 4932This attribute allows the compiler to construct the 4933requisite function declaration, while allowing the body of the 4934function to be assembly code. The specified function will not have 4935prologue/epilogue sequences generated by the compiler. Only basic 4936@code{asm} statements can safely be included in naked functions 4937(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 4938basic @code{asm} and C code may appear to work, they cannot be 4939depended upon to work reliably and are not supported. 4940@end table 4941 4942 4943@node Epiphany Function Attributes 4944@subsection Epiphany Function Attributes 4945 4946These function attributes are supported by the Epiphany back end: 4947 4948@table @code 4949@item disinterrupt 4950@cindex @code{disinterrupt} function attribute, Epiphany 4951This attribute causes the compiler to emit 4952instructions to disable interrupts for the duration of the given 4953function. 4954 4955@item forwarder_section 4956@cindex @code{forwarder_section} function attribute, Epiphany 4957This attribute modifies the behavior of an interrupt handler. 4958The interrupt handler may be in external memory which cannot be 4959reached by a branch instruction, so generate a local memory trampoline 4960to transfer control. The single parameter identifies the section where 4961the trampoline is placed. 4962 4963@item interrupt 4964@cindex @code{interrupt} function attribute, Epiphany 4965Use this attribute to indicate 4966that the specified function is an interrupt handler. The compiler generates 4967function entry and exit sequences suitable for use in an interrupt handler 4968when this attribute is present. It may also generate 4969a special section with code to initialize the interrupt vector table. 4970 4971On Epiphany targets one or more optional parameters can be added like this: 4972 4973@smallexample 4974void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler (); 4975@end smallexample 4976 4977Permissible values for these parameters are: @w{@code{reset}}, 4978@w{@code{software_exception}}, @w{@code{page_miss}}, 4979@w{@code{timer0}}, @w{@code{timer1}}, @w{@code{message}}, 4980@w{@code{dma0}}, @w{@code{dma1}}, @w{@code{wand}} and @w{@code{swi}}. 4981Multiple parameters indicate that multiple entries in the interrupt 4982vector table should be initialized for this function, i.e.@: for each 4983parameter @w{@var{name}}, a jump to the function is emitted in 4984the section @w{ivt_entry_@var{name}}. The parameter(s) may be omitted 4985entirely, in which case no interrupt vector table entry is provided. 4986 4987Note that interrupts are enabled inside the function 4988unless the @code{disinterrupt} attribute is also specified. 4989 4990The following examples are all valid uses of these attributes on 4991Epiphany targets: 4992@smallexample 4993void __attribute__ ((interrupt)) universal_handler (); 4994void __attribute__ ((interrupt ("dma1"))) dma1_handler (); 4995void __attribute__ ((interrupt ("dma0, dma1"))) 4996 universal_dma_handler (); 4997void __attribute__ ((interrupt ("timer0"), disinterrupt)) 4998 fast_timer_handler (); 4999void __attribute__ ((interrupt ("dma0, dma1"), 5000 forwarder_section ("tramp"))) 5001 external_dma_handler (); 5002@end smallexample 5003 5004@item long_call 5005@itemx short_call 5006@cindex @code{long_call} function attribute, Epiphany 5007@cindex @code{short_call} function attribute, Epiphany 5008@cindex indirect calls, Epiphany 5009These attributes specify how a particular function is called. 5010These attributes override the 5011@option{-mlong-calls} (@pxref{Adapteva Epiphany Options}) 5012command-line switch and @code{#pragma long_calls} settings. 5013@end table 5014 5015 5016@node H8/300 Function Attributes 5017@subsection H8/300 Function Attributes 5018 5019These function attributes are available for H8/300 targets: 5020 5021@table @code 5022@item function_vector 5023@cindex @code{function_vector} function attribute, H8/300 5024Use this attribute on the H8/300, H8/300H, and H8S to indicate 5025that the specified function should be called through the function vector. 5026Calling a function through the function vector reduces code size; however, 5027the function vector has a limited size (maximum 128 entries on the H8/300 5028and 64 entries on the H8/300H and H8S) 5029and shares space with the interrupt vector. 5030 5031@item interrupt_handler 5032@cindex @code{interrupt_handler} function attribute, H8/300 5033Use this attribute on the H8/300, H8/300H, and H8S to 5034indicate that the specified function is an interrupt handler. The compiler 5035generates function entry and exit sequences suitable for use in an 5036interrupt handler when this attribute is present. 5037 5038@item saveall 5039@cindex @code{saveall} function attribute, H8/300 5040@cindex save all registers on the H8/300, H8/300H, and H8S 5041Use this attribute on the H8/300, H8/300H, and H8S to indicate that 5042all registers except the stack pointer should be saved in the prologue 5043regardless of whether they are used or not. 5044@end table 5045 5046@node IA-64 Function Attributes 5047@subsection IA-64 Function Attributes 5048 5049These function attributes are supported on IA-64 targets: 5050 5051@table @code 5052@item syscall_linkage 5053@cindex @code{syscall_linkage} function attribute, IA-64 5054This attribute is used to modify the IA-64 calling convention by marking 5055all input registers as live at all function exits. This makes it possible 5056to restart a system call after an interrupt without having to save/restore 5057the input registers. This also prevents kernel data from leaking into 5058application code. 5059 5060@item version_id 5061@cindex @code{version_id} function attribute, IA-64 5062This IA-64 HP-UX attribute, attached to a global variable or function, renames a 5063symbol to contain a version string, thus allowing for function level 5064versioning. HP-UX system header files may use function level versioning 5065for some system calls. 5066 5067@smallexample 5068extern int foo () __attribute__((version_id ("20040821"))); 5069@end smallexample 5070 5071@noindent 5072Calls to @code{foo} are mapped to calls to @code{foo@{20040821@}}. 5073@end table 5074 5075@node M32C Function Attributes 5076@subsection M32C Function Attributes 5077 5078These function attributes are supported by the M32C back end: 5079 5080@table @code 5081@item bank_switch 5082@cindex @code{bank_switch} function attribute, M32C 5083When added to an interrupt handler with the M32C port, causes the 5084prologue and epilogue to use bank switching to preserve the registers 5085rather than saving them on the stack. 5086 5087@item fast_interrupt 5088@cindex @code{fast_interrupt} function attribute, M32C 5089Use this attribute on the M32C port to indicate that the specified 5090function is a fast interrupt handler. This is just like the 5091@code{interrupt} attribute, except that @code{freit} is used to return 5092instead of @code{reit}. 5093 5094@item function_vector 5095@cindex @code{function_vector} function attribute, M16C/M32C 5096On M16C/M32C targets, the @code{function_vector} attribute declares a 5097special page subroutine call function. Use of this attribute reduces 5098the code size by 2 bytes for each call generated to the 5099subroutine. The argument to the attribute is the vector number entry 5100from the special page vector table which contains the 16 low-order 5101bits of the subroutine's entry address. Each vector table has special 5102page number (18 to 255) that is used in @code{jsrs} instructions. 5103Jump addresses of the routines are generated by adding 0x0F0000 (in 5104case of M16C targets) or 0xFF0000 (in case of M32C targets), to the 51052-byte addresses set in the vector table. Therefore you need to ensure 5106that all the special page vector routines should get mapped within the 5107address range 0x0F0000 to 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF 5108(for M32C). 5109 5110In the following example 2 bytes are saved for each call to 5111function @code{foo}. 5112 5113@smallexample 5114void foo (void) __attribute__((function_vector(0x18))); 5115void foo (void) 5116@{ 5117@} 5118 5119void bar (void) 5120@{ 5121 foo(); 5122@} 5123@end smallexample 5124 5125If functions are defined in one file and are called in another file, 5126then be sure to write this declaration in both files. 5127 5128This attribute is ignored for R8C target. 5129 5130@item interrupt 5131@cindex @code{interrupt} function attribute, M32C 5132Use this attribute to indicate 5133that the specified function is an interrupt handler. The compiler generates 5134function entry and exit sequences suitable for use in an interrupt handler 5135when this attribute is present. 5136@end table 5137 5138@node M32R/D Function Attributes 5139@subsection M32R/D Function Attributes 5140 5141These function attributes are supported by the M32R/D back end: 5142 5143@table @code 5144@item interrupt 5145@cindex @code{interrupt} function attribute, M32R/D 5146Use this attribute to indicate 5147that the specified function is an interrupt handler. The compiler generates 5148function entry and exit sequences suitable for use in an interrupt handler 5149when this attribute is present. 5150 5151@item model (@var{model-name}) 5152@cindex @code{model} function attribute, M32R/D 5153@cindex function addressability on the M32R/D 5154 5155On the M32R/D, use this attribute to set the addressability of an 5156object, and of the code generated for a function. The identifier 5157@var{model-name} is one of @code{small}, @code{medium}, or 5158@code{large}, representing each of the code models. 5159 5160Small model objects live in the lower 16MB of memory (so that their 5161addresses can be loaded with the @code{ld24} instruction), and are 5162callable with the @code{bl} instruction. 5163 5164Medium model objects may live anywhere in the 32-bit address space (the 5165compiler generates @code{seth/add3} instructions to load their addresses), 5166and are callable with the @code{bl} instruction. 5167 5168Large model objects may live anywhere in the 32-bit address space (the 5169compiler generates @code{seth/add3} instructions to load their addresses), 5170and may not be reachable with the @code{bl} instruction (the compiler 5171generates the much slower @code{seth/add3/jl} instruction sequence). 5172@end table 5173 5174@node m68k Function Attributes 5175@subsection m68k Function Attributes 5176 5177These function attributes are supported by the m68k back end: 5178 5179@table @code 5180@item interrupt 5181@itemx interrupt_handler 5182@cindex @code{interrupt} function attribute, m68k 5183@cindex @code{interrupt_handler} function attribute, m68k 5184Use this attribute to 5185indicate that the specified function is an interrupt handler. The compiler 5186generates function entry and exit sequences suitable for use in an 5187interrupt handler when this attribute is present. Either name may be used. 5188 5189@item interrupt_thread 5190@cindex @code{interrupt_thread} function attribute, fido 5191Use this attribute on fido, a subarchitecture of the m68k, to indicate 5192that the specified function is an interrupt handler that is designed 5193to run as a thread. The compiler omits generate prologue/epilogue 5194sequences and replaces the return instruction with a @code{sleep} 5195instruction. This attribute is available only on fido. 5196@end table 5197 5198@node MCORE Function Attributes 5199@subsection MCORE Function Attributes 5200 5201These function attributes are supported by the MCORE back end: 5202 5203@table @code 5204@item naked 5205@cindex @code{naked} function attribute, MCORE 5206This attribute allows the compiler to construct the 5207requisite function declaration, while allowing the body of the 5208function to be assembly code. The specified function will not have 5209prologue/epilogue sequences generated by the compiler. Only basic 5210@code{asm} statements can safely be included in naked functions 5211(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 5212basic @code{asm} and C code may appear to work, they cannot be 5213depended upon to work reliably and are not supported. 5214@end table 5215 5216@node MeP Function Attributes 5217@subsection MeP Function Attributes 5218 5219These function attributes are supported by the MeP back end: 5220 5221@table @code 5222@item disinterrupt 5223@cindex @code{disinterrupt} function attribute, MeP 5224On MeP targets, this attribute causes the compiler to emit 5225instructions to disable interrupts for the duration of the given 5226function. 5227 5228@item interrupt 5229@cindex @code{interrupt} function attribute, MeP 5230Use this attribute to indicate 5231that the specified function is an interrupt handler. The compiler generates 5232function entry and exit sequences suitable for use in an interrupt handler 5233when this attribute is present. 5234 5235@item near 5236@cindex @code{near} function attribute, MeP 5237This attribute causes the compiler to assume the called 5238function is close enough to use the normal calling convention, 5239overriding the @option{-mtf} command-line option. 5240 5241@item far 5242@cindex @code{far} function attribute, MeP 5243On MeP targets this causes the compiler to use a calling convention 5244that assumes the called function is too far away for the built-in 5245addressing modes. 5246 5247@item vliw 5248@cindex @code{vliw} function attribute, MeP 5249The @code{vliw} attribute tells the compiler to emit 5250instructions in VLIW mode instead of core mode. Note that this 5251attribute is not allowed unless a VLIW coprocessor has been configured 5252and enabled through command-line options. 5253@end table 5254 5255@node MicroBlaze Function Attributes 5256@subsection MicroBlaze Function Attributes 5257 5258These function attributes are supported on MicroBlaze targets: 5259 5260@table @code 5261@item save_volatiles 5262@cindex @code{save_volatiles} function attribute, MicroBlaze 5263Use this attribute to indicate that the function is 5264an interrupt handler. All volatile registers (in addition to non-volatile 5265registers) are saved in the function prologue. If the function is a leaf 5266function, only volatiles used by the function are saved. A normal function 5267return is generated instead of a return from interrupt. 5268 5269@item break_handler 5270@cindex @code{break_handler} function attribute, MicroBlaze 5271@cindex break handler functions 5272Use this attribute to indicate that 5273the specified function is a break handler. The compiler generates function 5274entry and exit sequences suitable for use in an break handler when this 5275attribute is present. The return from @code{break_handler} is done through 5276the @code{rtbd} instead of @code{rtsd}. 5277 5278@smallexample 5279void f () __attribute__ ((break_handler)); 5280@end smallexample 5281 5282@item interrupt_handler 5283@itemx fast_interrupt 5284@cindex @code{interrupt_handler} function attribute, MicroBlaze 5285@cindex @code{fast_interrupt} function attribute, MicroBlaze 5286These attributes indicate that the specified function is an interrupt 5287handler. Use the @code{fast_interrupt} attribute to indicate handlers 5288used in low-latency interrupt mode, and @code{interrupt_handler} for 5289interrupts that do not use low-latency handlers. In both cases, GCC 5290emits appropriate prologue code and generates a return from the handler 5291using @code{rtid} instead of @code{rtsd}. 5292@end table 5293 5294@node Microsoft Windows Function Attributes 5295@subsection Microsoft Windows Function Attributes 5296 5297The following attributes are available on Microsoft Windows and Symbian OS 5298targets. 5299 5300@table @code 5301@item dllexport 5302@cindex @code{dllexport} function attribute 5303@cindex @code{__declspec(dllexport)} 5304On Microsoft Windows targets and Symbian OS targets the 5305@code{dllexport} attribute causes the compiler to provide a global 5306pointer to a pointer in a DLL, so that it can be referenced with the 5307@code{dllimport} attribute. On Microsoft Windows targets, the pointer 5308name is formed by combining @code{_imp__} and the function or variable 5309name. 5310 5311You can use @code{__declspec(dllexport)} as a synonym for 5312@code{__attribute__ ((dllexport))} for compatibility with other 5313compilers. 5314 5315On systems that support the @code{visibility} attribute, this 5316attribute also implies ``default'' visibility. It is an error to 5317explicitly specify any other visibility. 5318 5319GCC's default behavior is to emit all inline functions with the 5320@code{dllexport} attribute. Since this can cause object file-size bloat, 5321you can use @option{-fno-keep-inline-dllexport}, which tells GCC to 5322ignore the attribute for inlined functions unless the 5323@option{-fkeep-inline-functions} flag is used instead. 5324 5325The attribute is ignored for undefined symbols. 5326 5327When applied to C++ classes, the attribute marks defined non-inlined 5328member functions and static data members as exports. Static consts 5329initialized in-class are not marked unless they are also defined 5330out-of-class. 5331 5332For Microsoft Windows targets there are alternative methods for 5333including the symbol in the DLL's export table such as using a 5334@file{.def} file with an @code{EXPORTS} section or, with GNU ld, using 5335the @option{--export-all} linker flag. 5336 5337@item dllimport 5338@cindex @code{dllimport} function attribute 5339@cindex @code{__declspec(dllimport)} 5340On Microsoft Windows and Symbian OS targets, the @code{dllimport} 5341attribute causes the compiler to reference a function or variable via 5342a global pointer to a pointer that is set up by the DLL exporting the 5343symbol. The attribute implies @code{extern}. On Microsoft Windows 5344targets, the pointer name is formed by combining @code{_imp__} and the 5345function or variable name. 5346 5347You can use @code{__declspec(dllimport)} as a synonym for 5348@code{__attribute__ ((dllimport))} for compatibility with other 5349compilers. 5350 5351On systems that support the @code{visibility} attribute, this 5352attribute also implies ``default'' visibility. It is an error to 5353explicitly specify any other visibility. 5354 5355Currently, the attribute is ignored for inlined functions. If the 5356attribute is applied to a symbol @emph{definition}, an error is reported. 5357If a symbol previously declared @code{dllimport} is later defined, the 5358attribute is ignored in subsequent references, and a warning is emitted. 5359The attribute is also overridden by a subsequent declaration as 5360@code{dllexport}. 5361 5362When applied to C++ classes, the attribute marks non-inlined 5363member functions and static data members as imports. However, the 5364attribute is ignored for virtual methods to allow creation of vtables 5365using thunks. 5366 5367On the SH Symbian OS target the @code{dllimport} attribute also has 5368another affect---it can cause the vtable and run-time type information 5369for a class to be exported. This happens when the class has a 5370dllimported constructor or a non-inline, non-pure virtual function 5371and, for either of those two conditions, the class also has an inline 5372constructor or destructor and has a key function that is defined in 5373the current translation unit. 5374 5375For Microsoft Windows targets the use of the @code{dllimport} 5376attribute on functions is not necessary, but provides a small 5377performance benefit by eliminating a thunk in the DLL@. The use of the 5378@code{dllimport} attribute on imported variables can be avoided by passing the 5379@option{--enable-auto-import} switch to the GNU linker. As with 5380functions, using the attribute for a variable eliminates a thunk in 5381the DLL@. 5382 5383One drawback to using this attribute is that a pointer to a 5384@emph{variable} marked as @code{dllimport} cannot be used as a constant 5385address. However, a pointer to a @emph{function} with the 5386@code{dllimport} attribute can be used as a constant initializer; in 5387this case, the address of a stub function in the import lib is 5388referenced. On Microsoft Windows targets, the attribute can be disabled 5389for functions by setting the @option{-mnop-fun-dllimport} flag. 5390@end table 5391 5392@node MIPS Function Attributes 5393@subsection MIPS Function Attributes 5394 5395These function attributes are supported by the MIPS back end: 5396 5397@table @code 5398@item interrupt 5399@cindex @code{interrupt} function attribute, MIPS 5400Use this attribute to indicate that the specified function is an interrupt 5401handler. The compiler generates function entry and exit sequences suitable 5402for use in an interrupt handler when this attribute is present. 5403An optional argument is supported for the interrupt attribute which allows 5404the interrupt mode to be described. By default GCC assumes the external 5405interrupt controller (EIC) mode is in use, this can be explicitly set using 5406@code{eic}. When interrupts are non-masked then the requested Interrupt 5407Priority Level (IPL) is copied to the current IPL which has the effect of only 5408enabling higher priority interrupts. To use vectored interrupt mode use 5409the argument @code{vector=[sw0|sw1|hw0|hw1|hw2|hw3|hw4|hw5]}, this will change 5410the behavior of the non-masked interrupt support and GCC will arrange to mask 5411all interrupts from sw0 up to and including the specified interrupt vector. 5412 5413You can use the following attributes to modify the behavior 5414of an interrupt handler: 5415@table @code 5416@item use_shadow_register_set 5417@cindex @code{use_shadow_register_set} function attribute, MIPS 5418Assume that the handler uses a shadow register set, instead of 5419the main general-purpose registers. An optional argument @code{intstack} is 5420supported to indicate that the shadow register set contains a valid stack 5421pointer. 5422 5423@item keep_interrupts_masked 5424@cindex @code{keep_interrupts_masked} function attribute, MIPS 5425Keep interrupts masked for the whole function. Without this attribute, 5426GCC tries to reenable interrupts for as much of the function as it can. 5427 5428@item use_debug_exception_return 5429@cindex @code{use_debug_exception_return} function attribute, MIPS 5430Return using the @code{deret} instruction. Interrupt handlers that don't 5431have this attribute return using @code{eret} instead. 5432@end table 5433 5434You can use any combination of these attributes, as shown below: 5435@smallexample 5436void __attribute__ ((interrupt)) v0 (); 5437void __attribute__ ((interrupt, use_shadow_register_set)) v1 (); 5438void __attribute__ ((interrupt, keep_interrupts_masked)) v2 (); 5439void __attribute__ ((interrupt, use_debug_exception_return)) v3 (); 5440void __attribute__ ((interrupt, use_shadow_register_set, 5441 keep_interrupts_masked)) v4 (); 5442void __attribute__ ((interrupt, use_shadow_register_set, 5443 use_debug_exception_return)) v5 (); 5444void __attribute__ ((interrupt, keep_interrupts_masked, 5445 use_debug_exception_return)) v6 (); 5446void __attribute__ ((interrupt, use_shadow_register_set, 5447 keep_interrupts_masked, 5448 use_debug_exception_return)) v7 (); 5449void __attribute__ ((interrupt("eic"))) v8 (); 5450void __attribute__ ((interrupt("vector=hw3"))) v9 (); 5451@end smallexample 5452 5453@item long_call 5454@itemx short_call 5455@itemx near 5456@itemx far 5457@cindex indirect calls, MIPS 5458@cindex @code{long_call} function attribute, MIPS 5459@cindex @code{short_call} function attribute, MIPS 5460@cindex @code{near} function attribute, MIPS 5461@cindex @code{far} function attribute, MIPS 5462These attributes specify how a particular function is called on MIPS@. 5463The attributes override the @option{-mlong-calls} (@pxref{MIPS Options}) 5464command-line switch. The @code{long_call} and @code{far} attributes are 5465synonyms, and cause the compiler to always call 5466the function by first loading its address into a register, and then using 5467the contents of that register. The @code{short_call} and @code{near} 5468attributes are synonyms, and have the opposite 5469effect; they specify that non-PIC calls should be made using the more 5470efficient @code{jal} instruction. 5471 5472@item mips16 5473@itemx nomips16 5474@cindex @code{mips16} function attribute, MIPS 5475@cindex @code{nomips16} function attribute, MIPS 5476 5477On MIPS targets, you can use the @code{mips16} and @code{nomips16} 5478function attributes to locally select or turn off MIPS16 code generation. 5479A function with the @code{mips16} attribute is emitted as MIPS16 code, 5480while MIPS16 code generation is disabled for functions with the 5481@code{nomips16} attribute. These attributes override the 5482@option{-mips16} and @option{-mno-mips16} options on the command line 5483(@pxref{MIPS Options}). 5484 5485When compiling files containing mixed MIPS16 and non-MIPS16 code, the 5486preprocessor symbol @code{__mips16} reflects the setting on the command line, 5487not that within individual functions. Mixed MIPS16 and non-MIPS16 code 5488may interact badly with some GCC extensions such as @code{__builtin_apply} 5489(@pxref{Constructing Calls}). 5490 5491@item micromips, MIPS 5492@itemx nomicromips, MIPS 5493@cindex @code{micromips} function attribute 5494@cindex @code{nomicromips} function attribute 5495 5496On MIPS targets, you can use the @code{micromips} and @code{nomicromips} 5497function attributes to locally select or turn off microMIPS code generation. 5498A function with the @code{micromips} attribute is emitted as microMIPS code, 5499while microMIPS code generation is disabled for functions with the 5500@code{nomicromips} attribute. These attributes override the 5501@option{-mmicromips} and @option{-mno-micromips} options on the command line 5502(@pxref{MIPS Options}). 5503 5504When compiling files containing mixed microMIPS and non-microMIPS code, the 5505preprocessor symbol @code{__mips_micromips} reflects the setting on the 5506command line, 5507not that within individual functions. Mixed microMIPS and non-microMIPS code 5508may interact badly with some GCC extensions such as @code{__builtin_apply} 5509(@pxref{Constructing Calls}). 5510 5511@item nocompression 5512@cindex @code{nocompression} function attribute, MIPS 5513On MIPS targets, you can use the @code{nocompression} function attribute 5514to locally turn off MIPS16 and microMIPS code generation. This attribute 5515overrides the @option{-mips16} and @option{-mmicromips} options on the 5516command line (@pxref{MIPS Options}). 5517@end table 5518 5519@node MSP430 Function Attributes 5520@subsection MSP430 Function Attributes 5521 5522These function attributes are supported by the MSP430 back end: 5523 5524@table @code 5525@item critical 5526@cindex @code{critical} function attribute, MSP430 5527Critical functions disable interrupts upon entry and restore the 5528previous interrupt state upon exit. Critical functions cannot also 5529have the @code{naked}, @code{reentrant} or @code{interrupt} attributes. 5530 5531The MSP430 hardware ensures that interrupts are disabled on entry to 5532@code{interrupt} functions, and restores the previous interrupt state 5533on exit. The @code{critical} attribute is therefore redundant on 5534@code{interrupt} functions. 5535 5536@item interrupt 5537@cindex @code{interrupt} function attribute, MSP430 5538Use this attribute to indicate 5539that the specified function is an interrupt handler. The compiler generates 5540function entry and exit sequences suitable for use in an interrupt handler 5541when this attribute is present. 5542 5543You can provide an argument to the interrupt 5544attribute which specifies a name or number. If the argument is a 5545number it indicates the slot in the interrupt vector table (0 - 31) to 5546which this handler should be assigned. If the argument is a name it 5547is treated as a symbolic name for the vector slot. These names should 5548match up with appropriate entries in the linker script. By default 5549the names @code{watchdog} for vector 26, @code{nmi} for vector 30 and 5550@code{reset} for vector 31 are recognized. 5551 5552@item naked 5553@cindex @code{naked} function attribute, MSP430 5554This attribute allows the compiler to construct the 5555requisite function declaration, while allowing the body of the 5556function to be assembly code. The specified function will not have 5557prologue/epilogue sequences generated by the compiler. Only basic 5558@code{asm} statements can safely be included in naked functions 5559(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 5560basic @code{asm} and C code may appear to work, they cannot be 5561depended upon to work reliably and are not supported. 5562 5563@item reentrant 5564@cindex @code{reentrant} function attribute, MSP430 5565Reentrant functions disable interrupts upon entry and enable them 5566upon exit. Reentrant functions cannot also have the @code{naked} 5567or @code{critical} attributes. They can have the @code{interrupt} 5568attribute. 5569 5570@item wakeup 5571@cindex @code{wakeup} function attribute, MSP430 5572This attribute only applies to interrupt functions. It is silently 5573ignored if applied to a non-interrupt function. A wakeup interrupt 5574function will rouse the processor from any low-power state that it 5575might be in when the function exits. 5576 5577@item lower 5578@itemx upper 5579@itemx either 5580@cindex @code{lower} function attribute, MSP430 5581@cindex @code{upper} function attribute, MSP430 5582@cindex @code{either} function attribute, MSP430 5583On the MSP430 target these attributes can be used to specify whether 5584the function or variable should be placed into low memory, high 5585memory, or the placement should be left to the linker to decide. The 5586attributes are only significant if compiling for the MSP430X 5587architecture in the large memory model. 5588 5589The attributes work in conjunction with a linker script that has been 5590augmented to specify where to place sections with a @code{.lower} and 5591a @code{.upper} prefix. So, for example, as well as placing the 5592@code{.data} section, the script also specifies the placement of a 5593@code{.lower.data} and a @code{.upper.data} section. The intention 5594is that @code{lower} sections are placed into a small but easier to 5595access memory region and the upper sections are placed into a larger, but 5596slower to access, region. 5597 5598The @code{either} attribute is special. It tells the linker to place 5599the object into the corresponding @code{lower} section if there is 5600room for it. If there is insufficient room then the object is placed 5601into the corresponding @code{upper} section instead. Note that the 5602placement algorithm is not very sophisticated. It does not attempt to 5603find an optimal packing of the @code{lower} sections. It just makes 5604one pass over the objects and does the best that it can. Using the 5605@option{-ffunction-sections} and @option{-fdata-sections} command-line 5606options can help the packing, however, since they produce smaller, 5607easier to pack regions. 5608@end table 5609 5610@node NDS32 Function Attributes 5611@subsection NDS32 Function Attributes 5612 5613These function attributes are supported by the NDS32 back end: 5614 5615@table @code 5616@item exception 5617@cindex @code{exception} function attribute 5618@cindex exception handler functions, NDS32 5619Use this attribute on the NDS32 target to indicate that the specified function 5620is an exception handler. The compiler will generate corresponding sections 5621for use in an exception handler. 5622 5623@item interrupt 5624@cindex @code{interrupt} function attribute, NDS32 5625On NDS32 target, this attribute indicates that the specified function 5626is an interrupt handler. The compiler generates corresponding sections 5627for use in an interrupt handler. You can use the following attributes 5628to modify the behavior: 5629@table @code 5630@item nested 5631@cindex @code{nested} function attribute, NDS32 5632This interrupt service routine is interruptible. 5633@item not_nested 5634@cindex @code{not_nested} function attribute, NDS32 5635This interrupt service routine is not interruptible. 5636@item nested_ready 5637@cindex @code{nested_ready} function attribute, NDS32 5638This interrupt service routine is interruptible after @code{PSW.GIE} 5639(global interrupt enable) is set. This allows interrupt service routine to 5640finish some short critical code before enabling interrupts. 5641@item save_all 5642@cindex @code{save_all} function attribute, NDS32 5643The system will help save all registers into stack before entering 5644interrupt handler. 5645@item partial_save 5646@cindex @code{partial_save} function attribute, NDS32 5647The system will help save caller registers into stack before entering 5648interrupt handler. 5649@end table 5650 5651@item naked 5652@cindex @code{naked} function attribute, NDS32 5653This attribute allows the compiler to construct the 5654requisite function declaration, while allowing the body of the 5655function to be assembly code. The specified function will not have 5656prologue/epilogue sequences generated by the compiler. Only basic 5657@code{asm} statements can safely be included in naked functions 5658(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 5659basic @code{asm} and C code may appear to work, they cannot be 5660depended upon to work reliably and are not supported. 5661 5662@item reset 5663@cindex @code{reset} function attribute, NDS32 5664@cindex reset handler functions 5665Use this attribute on the NDS32 target to indicate that the specified function 5666is a reset handler. The compiler will generate corresponding sections 5667for use in a reset handler. You can use the following attributes 5668to provide extra exception handling: 5669@table @code 5670@item nmi 5671@cindex @code{nmi} function attribute, NDS32 5672Provide a user-defined function to handle NMI exception. 5673@item warm 5674@cindex @code{warm} function attribute, NDS32 5675Provide a user-defined function to handle warm reset exception. 5676@end table 5677@end table 5678 5679@node Nios II Function Attributes 5680@subsection Nios II Function Attributes 5681 5682These function attributes are supported by the Nios II back end: 5683 5684@table @code 5685@item target (@var{options}) 5686@cindex @code{target} function attribute 5687As discussed in @ref{Common Function Attributes}, this attribute 5688allows specification of target-specific compilation options. 5689 5690When compiling for Nios II, the following options are allowed: 5691 5692@table @samp 5693@item custom-@var{insn}=@var{N} 5694@itemx no-custom-@var{insn} 5695@cindex @code{target("custom-@var{insn}=@var{N}")} function attribute, Nios II 5696@cindex @code{target("no-custom-@var{insn}")} function attribute, Nios II 5697Each @samp{custom-@var{insn}=@var{N}} attribute locally enables use of a 5698custom instruction with encoding @var{N} when generating code that uses 5699@var{insn}. Similarly, @samp{no-custom-@var{insn}} locally inhibits use of 5700the custom instruction @var{insn}. 5701These target attributes correspond to the 5702@option{-mcustom-@var{insn}=@var{N}} and @option{-mno-custom-@var{insn}} 5703command-line options, and support the same set of @var{insn} keywords. 5704@xref{Nios II Options}, for more information. 5705 5706@item custom-fpu-cfg=@var{name} 5707@cindex @code{target("custom-fpu-cfg=@var{name}")} function attribute, Nios II 5708This attribute corresponds to the @option{-mcustom-fpu-cfg=@var{name}} 5709command-line option, to select a predefined set of custom instructions 5710named @var{name}. 5711@xref{Nios II Options}, for more information. 5712@end table 5713@end table 5714 5715@node Nvidia PTX Function Attributes 5716@subsection Nvidia PTX Function Attributes 5717 5718These function attributes are supported by the Nvidia PTX back end: 5719 5720@table @code 5721@item kernel 5722@cindex @code{kernel} attribute, Nvidia PTX 5723This attribute indicates that the corresponding function should be compiled 5724as a kernel function, which can be invoked from the host via the CUDA RT 5725library. 5726By default functions are only callable only from other PTX functions. 5727 5728Kernel functions must have @code{void} return type. 5729@end table 5730 5731@node PowerPC Function Attributes 5732@subsection PowerPC Function Attributes 5733 5734These function attributes are supported by the PowerPC back end: 5735 5736@table @code 5737@item longcall 5738@itemx shortcall 5739@cindex indirect calls, PowerPC 5740@cindex @code{longcall} function attribute, PowerPC 5741@cindex @code{shortcall} function attribute, PowerPC 5742The @code{longcall} attribute 5743indicates that the function might be far away from the call site and 5744require a different (more expensive) calling sequence. The 5745@code{shortcall} attribute indicates that the function is always close 5746enough for the shorter calling sequence to be used. These attributes 5747override both the @option{-mlongcall} switch and 5748the @code{#pragma longcall} setting. 5749 5750@xref{RS/6000 and PowerPC Options}, for more information on whether long 5751calls are necessary. 5752 5753@item target (@var{options}) 5754@cindex @code{target} function attribute 5755As discussed in @ref{Common Function Attributes}, this attribute 5756allows specification of target-specific compilation options. 5757 5758On the PowerPC, the following options are allowed: 5759 5760@table @samp 5761@item altivec 5762@itemx no-altivec 5763@cindex @code{target("altivec")} function attribute, PowerPC 5764Generate code that uses (does not use) AltiVec instructions. In 576532-bit code, you cannot enable AltiVec instructions unless 5766@option{-mabi=altivec} is used on the command line. 5767 5768@item cmpb 5769@itemx no-cmpb 5770@cindex @code{target("cmpb")} function attribute, PowerPC 5771Generate code that uses (does not use) the compare bytes instruction 5772implemented on the POWER6 processor and other processors that support 5773the PowerPC V2.05 architecture. 5774 5775@item dlmzb 5776@itemx no-dlmzb 5777@cindex @code{target("dlmzb")} function attribute, PowerPC 5778Generate code that uses (does not use) the string-search @samp{dlmzb} 5779instruction on the IBM 405, 440, 464 and 476 processors. This instruction is 5780generated by default when targeting those processors. 5781 5782@item fprnd 5783@itemx no-fprnd 5784@cindex @code{target("fprnd")} function attribute, PowerPC 5785Generate code that uses (does not use) the FP round to integer 5786instructions implemented on the POWER5+ processor and other processors 5787that support the PowerPC V2.03 architecture. 5788 5789@item hard-dfp 5790@itemx no-hard-dfp 5791@cindex @code{target("hard-dfp")} function attribute, PowerPC 5792Generate code that uses (does not use) the decimal floating-point 5793instructions implemented on some POWER processors. 5794 5795@item isel 5796@itemx no-isel 5797@cindex @code{target("isel")} function attribute, PowerPC 5798Generate code that uses (does not use) ISEL instruction. 5799 5800@item mfcrf 5801@itemx no-mfcrf 5802@cindex @code{target("mfcrf")} function attribute, PowerPC 5803Generate code that uses (does not use) the move from condition 5804register field instruction implemented on the POWER4 processor and 5805other processors that support the PowerPC V2.01 architecture. 5806 5807@item mulhw 5808@itemx no-mulhw 5809@cindex @code{target("mulhw")} function attribute, PowerPC 5810Generate code that uses (does not use) the half-word multiply and 5811multiply-accumulate instructions on the IBM 405, 440, 464 and 476 processors. 5812These instructions are generated by default when targeting those 5813processors. 5814 5815@item multiple 5816@itemx no-multiple 5817@cindex @code{target("multiple")} function attribute, PowerPC 5818Generate code that uses (does not use) the load multiple word 5819instructions and the store multiple word instructions. 5820 5821@item update 5822@itemx no-update 5823@cindex @code{target("update")} function attribute, PowerPC 5824Generate code that uses (does not use) the load or store instructions 5825that update the base register to the address of the calculated memory 5826location. 5827 5828@item popcntb 5829@itemx no-popcntb 5830@cindex @code{target("popcntb")} function attribute, PowerPC 5831Generate code that uses (does not use) the popcount and double-precision 5832FP reciprocal estimate instruction implemented on the POWER5 5833processor and other processors that support the PowerPC V2.02 5834architecture. 5835 5836@item popcntd 5837@itemx no-popcntd 5838@cindex @code{target("popcntd")} function attribute, PowerPC 5839Generate code that uses (does not use) the popcount instruction 5840implemented on the POWER7 processor and other processors that support 5841the PowerPC V2.06 architecture. 5842 5843@item powerpc-gfxopt 5844@itemx no-powerpc-gfxopt 5845@cindex @code{target("powerpc-gfxopt")} function attribute, PowerPC 5846Generate code that uses (does not use) the optional PowerPC 5847architecture instructions in the Graphics group, including 5848floating-point select. 5849 5850@item powerpc-gpopt 5851@itemx no-powerpc-gpopt 5852@cindex @code{target("powerpc-gpopt")} function attribute, PowerPC 5853Generate code that uses (does not use) the optional PowerPC 5854architecture instructions in the General Purpose group, including 5855floating-point square root. 5856 5857@item recip-precision 5858@itemx no-recip-precision 5859@cindex @code{target("recip-precision")} function attribute, PowerPC 5860Assume (do not assume) that the reciprocal estimate instructions 5861provide higher-precision estimates than is mandated by the PowerPC 5862ABI. 5863 5864@item string 5865@itemx no-string 5866@cindex @code{target("string")} function attribute, PowerPC 5867Generate code that uses (does not use) the load string instructions 5868and the store string word instructions to save multiple registers and 5869do small block moves. 5870 5871@item vsx 5872@itemx no-vsx 5873@cindex @code{target("vsx")} function attribute, PowerPC 5874Generate code that uses (does not use) vector/scalar (VSX) 5875instructions, and also enable the use of built-in functions that allow 5876more direct access to the VSX instruction set. In 32-bit code, you 5877cannot enable VSX or AltiVec instructions unless 5878@option{-mabi=altivec} is used on the command line. 5879 5880@item friz 5881@itemx no-friz 5882@cindex @code{target("friz")} function attribute, PowerPC 5883Generate (do not generate) the @code{friz} instruction when the 5884@option{-funsafe-math-optimizations} option is used to optimize 5885rounding a floating-point value to 64-bit integer and back to floating 5886point. The @code{friz} instruction does not return the same value if 5887the floating-point number is too large to fit in an integer. 5888 5889@item avoid-indexed-addresses 5890@itemx no-avoid-indexed-addresses 5891@cindex @code{target("avoid-indexed-addresses")} function attribute, PowerPC 5892Generate code that tries to avoid (not avoid) the use of indexed load 5893or store instructions. 5894 5895@item paired 5896@itemx no-paired 5897@cindex @code{target("paired")} function attribute, PowerPC 5898Generate code that uses (does not use) the generation of PAIRED simd 5899instructions. 5900 5901@item longcall 5902@itemx no-longcall 5903@cindex @code{target("longcall")} function attribute, PowerPC 5904Generate code that assumes (does not assume) that all calls are far 5905away so that a longer more expensive calling sequence is required. 5906 5907@item cpu=@var{CPU} 5908@cindex @code{target("cpu=@var{CPU}")} function attribute, PowerPC 5909Specify the architecture to generate code for when compiling the 5910function. If you select the @code{target("cpu=power7")} attribute when 5911generating 32-bit code, VSX and AltiVec instructions are not generated 5912unless you use the @option{-mabi=altivec} option on the command line. 5913 5914@item tune=@var{TUNE} 5915@cindex @code{target("tune=@var{TUNE}")} function attribute, PowerPC 5916Specify the architecture to tune for when compiling the function. If 5917you do not specify the @code{target("tune=@var{TUNE}")} attribute and 5918you do specify the @code{target("cpu=@var{CPU}")} attribute, 5919compilation tunes for the @var{CPU} architecture, and not the 5920default tuning specified on the command line. 5921@end table 5922 5923On the PowerPC, the inliner does not inline a 5924function that has different target options than the caller, unless the 5925callee has a subset of the target options of the caller. 5926@end table 5927 5928@node RISC-V Function Attributes 5929@subsection RISC-V Function Attributes 5930 5931These function attributes are supported by the RISC-V back end: 5932 5933@table @code 5934@item naked 5935@cindex @code{naked} function attribute, RISC-V 5936This attribute allows the compiler to construct the 5937requisite function declaration, while allowing the body of the 5938function to be assembly code. The specified function will not have 5939prologue/epilogue sequences generated by the compiler. Only basic 5940@code{asm} statements can safely be included in naked functions 5941(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 5942basic @code{asm} and C code may appear to work, they cannot be 5943depended upon to work reliably and are not supported. 5944 5945@item interrupt 5946@cindex @code{interrupt} function attribute, RISC-V 5947Use this attribute to indicate that the specified function is an interrupt 5948handler. The compiler generates function entry and exit sequences suitable 5949for use in an interrupt handler when this attribute is present. 5950 5951You can specify the kind of interrupt to be handled by adding an optional 5952parameter to the interrupt attribute like this: 5953 5954@smallexample 5955void f (void) __attribute__ ((interrupt ("user"))); 5956@end smallexample 5957 5958Permissible values for this parameter are @code{user}, @code{supervisor}, 5959and @code{machine}. If there is no parameter, then it defaults to 5960@code{machine}. 5961@end table 5962 5963@node RL78 Function Attributes 5964@subsection RL78 Function Attributes 5965 5966These function attributes are supported by the RL78 back end: 5967 5968@table @code 5969@item interrupt 5970@itemx brk_interrupt 5971@cindex @code{interrupt} function attribute, RL78 5972@cindex @code{brk_interrupt} function attribute, RL78 5973These attributes indicate 5974that the specified function is an interrupt handler. The compiler generates 5975function entry and exit sequences suitable for use in an interrupt handler 5976when this attribute is present. 5977 5978Use @code{brk_interrupt} instead of @code{interrupt} for 5979handlers intended to be used with the @code{BRK} opcode (i.e.@: those 5980that must end with @code{RETB} instead of @code{RETI}). 5981 5982@item naked 5983@cindex @code{naked} function attribute, RL78 5984This attribute allows the compiler to construct the 5985requisite function declaration, while allowing the body of the 5986function to be assembly code. The specified function will not have 5987prologue/epilogue sequences generated by the compiler. Only basic 5988@code{asm} statements can safely be included in naked functions 5989(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 5990basic @code{asm} and C code may appear to work, they cannot be 5991depended upon to work reliably and are not supported. 5992@end table 5993 5994@node RX Function Attributes 5995@subsection RX Function Attributes 5996 5997These function attributes are supported by the RX back end: 5998 5999@table @code 6000@item fast_interrupt 6001@cindex @code{fast_interrupt} function attribute, RX 6002Use this attribute on the RX port to indicate that the specified 6003function is a fast interrupt handler. This is just like the 6004@code{interrupt} attribute, except that @code{freit} is used to return 6005instead of @code{reit}. 6006 6007@item interrupt 6008@cindex @code{interrupt} function attribute, RX 6009Use this attribute to indicate 6010that the specified function is an interrupt handler. The compiler generates 6011function entry and exit sequences suitable for use in an interrupt handler 6012when this attribute is present. 6013 6014On RX and RL78 targets, you may specify one or more vector numbers as arguments 6015to the attribute, as well as naming an alternate table name. 6016Parameters are handled sequentially, so one handler can be assigned to 6017multiple entries in multiple tables. One may also pass the magic 6018string @code{"$default"} which causes the function to be used for any 6019unfilled slots in the current table. 6020 6021This example shows a simple assignment of a function to one vector in 6022the default table (note that preprocessor macros may be used for 6023chip-specific symbolic vector names): 6024@smallexample 6025void __attribute__ ((interrupt (5))) txd1_handler (); 6026@end smallexample 6027 6028This example assigns a function to two slots in the default table 6029(using preprocessor macros defined elsewhere) and makes it the default 6030for the @code{dct} table: 6031@smallexample 6032void __attribute__ ((interrupt (RXD1_VECT,RXD2_VECT,"dct","$default"))) 6033 txd1_handler (); 6034@end smallexample 6035 6036@item naked 6037@cindex @code{naked} function attribute, RX 6038This attribute allows the compiler to construct the 6039requisite function declaration, while allowing the body of the 6040function to be assembly code. The specified function will not have 6041prologue/epilogue sequences generated by the compiler. Only basic 6042@code{asm} statements can safely be included in naked functions 6043(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 6044basic @code{asm} and C code may appear to work, they cannot be 6045depended upon to work reliably and are not supported. 6046 6047@item vector 6048@cindex @code{vector} function attribute, RX 6049This RX attribute is similar to the @code{interrupt} attribute, including its 6050parameters, but does not make the function an interrupt-handler type 6051function (i.e.@: it retains the normal C function calling ABI). See the 6052@code{interrupt} attribute for a description of its arguments. 6053@end table 6054 6055@node S/390 Function Attributes 6056@subsection S/390 Function Attributes 6057 6058These function attributes are supported on the S/390: 6059 6060@table @code 6061@item hotpatch (@var{halfwords-before-function-label},@var{halfwords-after-function-label}) 6062@cindex @code{hotpatch} function attribute, S/390 6063 6064On S/390 System z targets, you can use this function attribute to 6065make GCC generate a ``hot-patching'' function prologue. If the 6066@option{-mhotpatch=} command-line option is used at the same time, 6067the @code{hotpatch} attribute takes precedence. The first of the 6068two arguments specifies the number of halfwords to be added before 6069the function label. A second argument can be used to specify the 6070number of halfwords to be added after the function label. For 6071both arguments the maximum allowed value is 1000000. 6072 6073If both arguments are zero, hotpatching is disabled. 6074 6075@item target (@var{options}) 6076@cindex @code{target} function attribute 6077As discussed in @ref{Common Function Attributes}, this attribute 6078allows specification of target-specific compilation options. 6079 6080On S/390, the following options are supported: 6081 6082@table @samp 6083@item arch= 6084@item tune= 6085@item stack-guard= 6086@item stack-size= 6087@item branch-cost= 6088@item warn-framesize= 6089@item backchain 6090@itemx no-backchain 6091@item hard-dfp 6092@itemx no-hard-dfp 6093@item hard-float 6094@itemx soft-float 6095@item htm 6096@itemx no-htm 6097@item vx 6098@itemx no-vx 6099@item packed-stack 6100@itemx no-packed-stack 6101@item small-exec 6102@itemx no-small-exec 6103@item mvcle 6104@itemx no-mvcle 6105@item warn-dynamicstack 6106@itemx no-warn-dynamicstack 6107@end table 6108 6109The options work exactly like the S/390 specific command line 6110options (without the prefix @option{-m}) except that they do not 6111change any feature macros. For example, 6112 6113@smallexample 6114@code{target("no-vx")} 6115@end smallexample 6116 6117does not undefine the @code{__VEC__} macro. 6118@end table 6119 6120@node SH Function Attributes 6121@subsection SH Function Attributes 6122 6123These function attributes are supported on the SH family of processors: 6124 6125@table @code 6126@item function_vector 6127@cindex @code{function_vector} function attribute, SH 6128@cindex calling functions through the function vector on SH2A 6129On SH2A targets, this attribute declares a function to be called using the 6130TBR relative addressing mode. The argument to this attribute is the entry 6131number of the same function in a vector table containing all the TBR 6132relative addressable functions. For correct operation the TBR must be setup 6133accordingly to point to the start of the vector table before any functions with 6134this attribute are invoked. Usually a good place to do the initialization is 6135the startup routine. The TBR relative vector table can have at max 256 function 6136entries. The jumps to these functions are generated using a SH2A specific, 6137non delayed branch instruction JSR/N @@(disp8,TBR). You must use GAS and GLD 6138from GNU binutils version 2.7 or later for this attribute to work correctly. 6139 6140In an application, for a function being called once, this attribute 6141saves at least 8 bytes of code; and if other successive calls are being 6142made to the same function, it saves 2 bytes of code per each of these 6143calls. 6144 6145@item interrupt_handler 6146@cindex @code{interrupt_handler} function attribute, SH 6147Use this attribute to 6148indicate that the specified function is an interrupt handler. The compiler 6149generates function entry and exit sequences suitable for use in an 6150interrupt handler when this attribute is present. 6151 6152@item nosave_low_regs 6153@cindex @code{nosave_low_regs} function attribute, SH 6154Use this attribute on SH targets to indicate that an @code{interrupt_handler} 6155function should not save and restore registers R0..R7. This can be used on SH3* 6156and SH4* targets that have a second R0..R7 register bank for non-reentrant 6157interrupt handlers. 6158 6159@item renesas 6160@cindex @code{renesas} function attribute, SH 6161On SH targets this attribute specifies that the function or struct follows the 6162Renesas ABI. 6163 6164@item resbank 6165@cindex @code{resbank} function attribute, SH 6166On the SH2A target, this attribute enables the high-speed register 6167saving and restoration using a register bank for @code{interrupt_handler} 6168routines. Saving to the bank is performed automatically after the CPU 6169accepts an interrupt that uses a register bank. 6170 6171The nineteen 32-bit registers comprising general register R0 to R14, 6172control register GBR, and system registers MACH, MACL, and PR and the 6173vector table address offset are saved into a register bank. Register 6174banks are stacked in first-in last-out (FILO) sequence. Restoration 6175from the bank is executed by issuing a RESBANK instruction. 6176 6177@item sp_switch 6178@cindex @code{sp_switch} function attribute, SH 6179Use this attribute on the SH to indicate an @code{interrupt_handler} 6180function should switch to an alternate stack. It expects a string 6181argument that names a global variable holding the address of the 6182alternate stack. 6183 6184@smallexample 6185void *alt_stack; 6186void f () __attribute__ ((interrupt_handler, 6187 sp_switch ("alt_stack"))); 6188@end smallexample 6189 6190@item trap_exit 6191@cindex @code{trap_exit} function attribute, SH 6192Use this attribute on the SH for an @code{interrupt_handler} to return using 6193@code{trapa} instead of @code{rte}. This attribute expects an integer 6194argument specifying the trap number to be used. 6195 6196@item trapa_handler 6197@cindex @code{trapa_handler} function attribute, SH 6198On SH targets this function attribute is similar to @code{interrupt_handler} 6199but it does not save and restore all registers. 6200@end table 6201 6202@node Symbian OS Function Attributes 6203@subsection Symbian OS Function Attributes 6204 6205@xref{Microsoft Windows Function Attributes}, for discussion of the 6206@code{dllexport} and @code{dllimport} attributes. 6207 6208@node V850 Function Attributes 6209@subsection V850 Function Attributes 6210 6211The V850 back end supports these function attributes: 6212 6213@table @code 6214@item interrupt 6215@itemx interrupt_handler 6216@cindex @code{interrupt} function attribute, V850 6217@cindex @code{interrupt_handler} function attribute, V850 6218Use these attributes to indicate 6219that the specified function is an interrupt handler. The compiler generates 6220function entry and exit sequences suitable for use in an interrupt handler 6221when either attribute is present. 6222@end table 6223 6224@node Visium Function Attributes 6225@subsection Visium Function Attributes 6226 6227These function attributes are supported by the Visium back end: 6228 6229@table @code 6230@item interrupt 6231@cindex @code{interrupt} function attribute, Visium 6232Use this attribute to indicate 6233that the specified function is an interrupt handler. The compiler generates 6234function entry and exit sequences suitable for use in an interrupt handler 6235when this attribute is present. 6236@end table 6237 6238@node x86 Function Attributes 6239@subsection x86 Function Attributes 6240 6241These function attributes are supported by the x86 back end: 6242 6243@table @code 6244@item cdecl 6245@cindex @code{cdecl} function attribute, x86-32 6246@cindex functions that pop the argument stack on x86-32 6247@opindex mrtd 6248On the x86-32 targets, the @code{cdecl} attribute causes the compiler to 6249assume that the calling function pops off the stack space used to 6250pass arguments. This is 6251useful to override the effects of the @option{-mrtd} switch. 6252 6253@item fastcall 6254@cindex @code{fastcall} function attribute, x86-32 6255@cindex functions that pop the argument stack on x86-32 6256On x86-32 targets, the @code{fastcall} attribute causes the compiler to 6257pass the first argument (if of integral type) in the register ECX and 6258the second argument (if of integral type) in the register EDX@. Subsequent 6259and other typed arguments are passed on the stack. The called function 6260pops the arguments off the stack. If the number of arguments is variable all 6261arguments are pushed on the stack. 6262 6263@item thiscall 6264@cindex @code{thiscall} function attribute, x86-32 6265@cindex functions that pop the argument stack on x86-32 6266On x86-32 targets, the @code{thiscall} attribute causes the compiler to 6267pass the first argument (if of integral type) in the register ECX. 6268Subsequent and other typed arguments are passed on the stack. The called 6269function pops the arguments off the stack. 6270If the number of arguments is variable all arguments are pushed on the 6271stack. 6272The @code{thiscall} attribute is intended for C++ non-static member functions. 6273As a GCC extension, this calling convention can be used for C functions 6274and for static member methods. 6275 6276@item ms_abi 6277@itemx sysv_abi 6278@cindex @code{ms_abi} function attribute, x86 6279@cindex @code{sysv_abi} function attribute, x86 6280 6281On 32-bit and 64-bit x86 targets, you can use an ABI attribute 6282to indicate which calling convention should be used for a function. The 6283@code{ms_abi} attribute tells the compiler to use the Microsoft ABI, 6284while the @code{sysv_abi} attribute tells the compiler to use the System V 6285ELF ABI, which is used on GNU/Linux and other systems. The default is to use 6286the Microsoft ABI when targeting Windows. On all other systems, the default 6287is the System V ELF ABI. 6288 6289Note, the @code{ms_abi} attribute for Microsoft Windows 64-bit targets currently 6290requires the @option{-maccumulate-outgoing-args} option. 6291 6292@item callee_pop_aggregate_return (@var{number}) 6293@cindex @code{callee_pop_aggregate_return} function attribute, x86 6294 6295On x86-32 targets, you can use this attribute to control how 6296aggregates are returned in memory. If the caller is responsible for 6297popping the hidden pointer together with the rest of the arguments, specify 6298@var{number} equal to zero. If callee is responsible for popping the 6299hidden pointer, specify @var{number} equal to one. 6300 6301The default x86-32 ABI assumes that the callee pops the 6302stack for hidden pointer. However, on x86-32 Microsoft Windows targets, 6303the compiler assumes that the 6304caller pops the stack for hidden pointer. 6305 6306@item ms_hook_prologue 6307@cindex @code{ms_hook_prologue} function attribute, x86 6308 6309On 32-bit and 64-bit x86 targets, you can use 6310this function attribute to make GCC generate the ``hot-patching'' function 6311prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2 6312and newer. 6313 6314@item naked 6315@cindex @code{naked} function attribute, x86 6316This attribute allows the compiler to construct the 6317requisite function declaration, while allowing the body of the 6318function to be assembly code. The specified function will not have 6319prologue/epilogue sequences generated by the compiler. Only basic 6320@code{asm} statements can safely be included in naked functions 6321(@pxref{Basic Asm}). While using extended @code{asm} or a mixture of 6322basic @code{asm} and C code may appear to work, they cannot be 6323depended upon to work reliably and are not supported. 6324 6325@item regparm (@var{number}) 6326@cindex @code{regparm} function attribute, x86 6327@cindex functions that are passed arguments in registers on x86-32 6328On x86-32 targets, the @code{regparm} attribute causes the compiler to 6329pass arguments number one to @var{number} if they are of integral type 6330in registers EAX, EDX, and ECX instead of on the stack. Functions that 6331take a variable number of arguments continue to be passed all of their 6332arguments on the stack. 6333 6334Beware that on some ELF systems this attribute is unsuitable for 6335global functions in shared libraries with lazy binding (which is the 6336default). Lazy binding sends the first call via resolving code in 6337the loader, which might assume EAX, EDX and ECX can be clobbered, as 6338per the standard calling conventions. Solaris 8 is affected by this. 6339Systems with the GNU C Library version 2.1 or higher 6340and FreeBSD are believed to be 6341safe since the loaders there save EAX, EDX and ECX. (Lazy binding can be 6342disabled with the linker or the loader if desired, to avoid the 6343problem.) 6344 6345@item sseregparm 6346@cindex @code{sseregparm} function attribute, x86 6347On x86-32 targets with SSE support, the @code{sseregparm} attribute 6348causes the compiler to pass up to 3 floating-point arguments in 6349SSE registers instead of on the stack. Functions that take a 6350variable number of arguments continue to pass all of their 6351floating-point arguments on the stack. 6352 6353@item force_align_arg_pointer 6354@cindex @code{force_align_arg_pointer} function attribute, x86 6355On x86 targets, the @code{force_align_arg_pointer} attribute may be 6356applied to individual function definitions, generating an alternate 6357prologue and epilogue that realigns the run-time stack if necessary. 6358This supports mixing legacy codes that run with a 4-byte aligned stack 6359with modern codes that keep a 16-byte stack for SSE compatibility. 6360 6361@item stdcall 6362@cindex @code{stdcall} function attribute, x86-32 6363@cindex functions that pop the argument stack on x86-32 6364On x86-32 targets, the @code{stdcall} attribute causes the compiler to 6365assume that the called function pops off the stack space used to 6366pass arguments, unless it takes a variable number of arguments. 6367 6368@item no_caller_saved_registers 6369@cindex @code{no_caller_saved_registers} function attribute, x86 6370Use this attribute to indicate that the specified function has no 6371caller-saved registers. That is, all registers are callee-saved. For 6372example, this attribute can be used for a function called from an 6373interrupt handler. The compiler generates proper function entry and 6374exit sequences to save and restore any modified registers, except for 6375the EFLAGS register. Since GCC doesn't preserve SSE, MMX nor x87 6376states, the GCC option @option{-mgeneral-regs-only} should be used to 6377compile functions with @code{no_caller_saved_registers} attribute. 6378 6379@item interrupt 6380@cindex @code{interrupt} function attribute, x86 6381Use this attribute to indicate that the specified function is an 6382interrupt handler or an exception handler (depending on parameters passed 6383to the function, explained further). The compiler generates function 6384entry and exit sequences suitable for use in an interrupt handler when 6385this attribute is present. The @code{IRET} instruction, instead of the 6386@code{RET} instruction, is used to return from interrupt handlers. All 6387registers, except for the EFLAGS register which is restored by the 6388@code{IRET} instruction, are preserved by the compiler. Since GCC 6389doesn't preserve SSE, MMX nor x87 states, the GCC option 6390@option{-mgeneral-regs-only} should be used to compile interrupt and 6391exception handlers. 6392 6393Any interruptible-without-stack-switch code must be compiled with 6394@option{-mno-red-zone} since interrupt handlers can and will, because 6395of the hardware design, touch the red zone. 6396 6397An interrupt handler must be declared with a mandatory pointer 6398argument: 6399 6400@smallexample 6401struct interrupt_frame; 6402 6403__attribute__ ((interrupt)) 6404void 6405f (struct interrupt_frame *frame) 6406@{ 6407@} 6408@end smallexample 6409 6410@noindent 6411and you must define @code{struct interrupt_frame} as described in the 6412processor's manual. 6413 6414Exception handlers differ from interrupt handlers because the system 6415pushes an error code on the stack. An exception handler declaration is 6416similar to that for an interrupt handler, but with a different mandatory 6417function signature. The compiler arranges to pop the error code off the 6418stack before the @code{IRET} instruction. 6419 6420@smallexample 6421#ifdef __x86_64__ 6422typedef unsigned long long int uword_t; 6423#else 6424typedef unsigned int uword_t; 6425#endif 6426 6427struct interrupt_frame; 6428 6429__attribute__ ((interrupt)) 6430void 6431f (struct interrupt_frame *frame, uword_t error_code) 6432@{ 6433 ... 6434@} 6435@end smallexample 6436 6437Exception handlers should only be used for exceptions that push an error 6438code; you should use an interrupt handler in other cases. The system 6439will crash if the wrong kind of handler is used. 6440 6441@item target (@var{options}) 6442@cindex @code{target} function attribute 6443As discussed in @ref{Common Function Attributes}, this attribute 6444allows specification of target-specific compilation options. 6445 6446On the x86, the following options are allowed: 6447@table @samp 6448@item 3dnow 6449@itemx no-3dnow 6450@cindex @code{target("3dnow")} function attribute, x86 6451Enable/disable the generation of the 3DNow!@: instructions. 6452 6453@item 3dnowa 6454@itemx no-3dnowa 6455@cindex @code{target("3dnowa")} function attribute, x86 6456Enable/disable the generation of the enhanced 3DNow!@: instructions. 6457 6458@item abm 6459@itemx no-abm 6460@cindex @code{target("abm")} function attribute, x86 6461Enable/disable the generation of the advanced bit instructions. 6462 6463@item adx 6464@itemx no-adx 6465@cindex @code{target("adx")} function attribute, x86 6466Enable/disable the generation of the ADX instructions. 6467 6468@item aes 6469@itemx no-aes 6470@cindex @code{target("aes")} function attribute, x86 6471Enable/disable the generation of the AES instructions. 6472 6473@item avx 6474@itemx no-avx 6475@cindex @code{target("avx")} function attribute, x86 6476Enable/disable the generation of the AVX instructions. 6477 6478@item avx2 6479@itemx no-avx2 6480@cindex @code{target("avx2")} function attribute, x86 6481Enable/disable the generation of the AVX2 instructions. 6482 6483@item avx5124fmaps 6484@itemx no-avx5124fmaps 6485@cindex @code{target("avx5124fmaps")} function attribute, x86 6486Enable/disable the generation of the AVX5124FMAPS instructions. 6487 6488@item avx5124vnniw 6489@itemx no-avx5124vnniw 6490@cindex @code{target("avx5124vnniw")} function attribute, x86 6491Enable/disable the generation of the AVX5124VNNIW instructions. 6492 6493@item avx512bitalg 6494@itemx no-avx512bitalg 6495@cindex @code{target("avx512bitalg")} function attribute, x86 6496Enable/disable the generation of the AVX512BITALG instructions. 6497 6498@item avx512bw 6499@itemx no-avx512bw 6500@cindex @code{target("avx512bw")} function attribute, x86 6501Enable/disable the generation of the AVX512BW instructions. 6502 6503@item avx512cd 6504@itemx no-avx512cd 6505@cindex @code{target("avx512cd")} function attribute, x86 6506Enable/disable the generation of the AVX512CD instructions. 6507 6508@item avx512dq 6509@itemx no-avx512dq 6510@cindex @code{target("avx512dq")} function attribute, x86 6511Enable/disable the generation of the AVX512DQ instructions. 6512 6513@item avx512er 6514@itemx no-avx512er 6515@cindex @code{target("avx512er")} function attribute, x86 6516Enable/disable the generation of the AVX512ER instructions. 6517 6518@item avx512f 6519@itemx no-avx512f 6520@cindex @code{target("avx512f")} function attribute, x86 6521Enable/disable the generation of the AVX512F instructions. 6522 6523@item avx512ifma 6524@itemx no-avx512ifma 6525@cindex @code{target("avx512ifma")} function attribute, x86 6526Enable/disable the generation of the AVX512IFMA instructions. 6527 6528@item avx512pf 6529@itemx no-avx512pf 6530@cindex @code{target("avx512pf")} function attribute, x86 6531Enable/disable the generation of the AVX512PF instructions. 6532 6533@item avx512vbmi 6534@itemx no-avx512vbmi 6535@cindex @code{target("avx512vbmi")} function attribute, x86 6536Enable/disable the generation of the AVX512VBMI instructions. 6537 6538@item avx512vbmi2 6539@itemx no-avx512vbmi2 6540@cindex @code{target("avx512vbmi2")} function attribute, x86 6541Enable/disable the generation of the AVX512VBMI2 instructions. 6542 6543@item avx512vl 6544@itemx no-avx512vl 6545@cindex @code{target("avx512vl")} function attribute, x86 6546Enable/disable the generation of the AVX512VL instructions. 6547 6548@item avx512vnni 6549@itemx no-avx512vnni 6550@cindex @code{target("avx512vnni")} function attribute, x86 6551Enable/disable the generation of the AVX512VNNI instructions. 6552 6553@item avx512vpopcntdq 6554@itemx no-avx512vpopcntdq 6555@cindex @code{target("avx512vpopcntdq")} function attribute, x86 6556Enable/disable the generation of the AVX512VPOPCNTDQ instructions. 6557 6558@item bmi 6559@itemx no-bmi 6560@cindex @code{target("bmi")} function attribute, x86 6561Enable/disable the generation of the BMI instructions. 6562 6563@item bmi2 6564@itemx no-bmi2 6565@cindex @code{target("bmi2")} function attribute, x86 6566Enable/disable the generation of the BMI2 instructions. 6567 6568@item cldemote 6569@itemx no-cldemote 6570@cindex @code{target("cldemote")} function attribute, x86 6571Enable/disable the generation of the CLDEMOTE instructions. 6572 6573@item clflushopt 6574@itemx no-clflushopt 6575@cindex @code{target("clflushopt")} function attribute, x86 6576Enable/disable the generation of the CLFLUSHOPT instructions. 6577 6578@item clwb 6579@itemx no-clwb 6580@cindex @code{target("clwb")} function attribute, x86 6581Enable/disable the generation of the CLWB instructions. 6582 6583@item clzero 6584@itemx no-clzero 6585@cindex @code{target("clzero")} function attribute, x86 6586Enable/disable the generation of the CLZERO instructions. 6587 6588@item crc32 6589@itemx no-crc32 6590@cindex @code{target("crc32")} function attribute, x86 6591Enable/disable the generation of the CRC32 instructions. 6592 6593@item cx16 6594@itemx no-cx16 6595@cindex @code{target("cx16")} function attribute, x86 6596Enable/disable the generation of the CMPXCHG16B instructions. 6597 6598@item default 6599@cindex @code{target("default")} function attribute, x86 6600@xref{Function Multiversioning}, where it is used to specify the 6601default function version. 6602 6603@item f16c 6604@itemx no-f16c 6605@cindex @code{target("f16c")} function attribute, x86 6606Enable/disable the generation of the F16C instructions. 6607 6608@item fma 6609@itemx no-fma 6610@cindex @code{target("fma")} function attribute, x86 6611Enable/disable the generation of the FMA instructions. 6612 6613@item fma4 6614@itemx no-fma4 6615@cindex @code{target("fma4")} function attribute, x86 6616Enable/disable the generation of the FMA4 instructions. 6617 6618@item fsgsbase 6619@itemx no-fsgsbase 6620@cindex @code{target("fsgsbase")} function attribute, x86 6621Enable/disable the generation of the FSGSBASE instructions. 6622 6623@item fxsr 6624@itemx no-fxsr 6625@cindex @code{target("fxsr")} function attribute, x86 6626Enable/disable the generation of the FXSR instructions. 6627 6628@item gfni 6629@itemx no-gfni 6630@cindex @code{target("gfni")} function attribute, x86 6631Enable/disable the generation of the GFNI instructions. 6632 6633@item hle 6634@itemx no-hle 6635@cindex @code{target("hle")} function attribute, x86 6636Enable/disable the generation of the HLE instruction prefixes. 6637 6638@item lwp 6639@itemx no-lwp 6640@cindex @code{target("lwp")} function attribute, x86 6641Enable/disable the generation of the LWP instructions. 6642 6643@item lzcnt 6644@itemx no-lzcnt 6645@cindex @code{target("lzcnt")} function attribute, x86 6646Enable/disable the generation of the LZCNT instructions. 6647 6648@item mmx 6649@itemx no-mmx 6650@cindex @code{target("mmx")} function attribute, x86 6651Enable/disable the generation of the MMX instructions. 6652 6653@item movbe 6654@itemx no-movbe 6655@cindex @code{target("movbe")} function attribute, x86 6656Enable/disable the generation of the MOVBE instructions. 6657 6658@item movdir64b 6659@itemx no-movdir64b 6660@cindex @code{target("movdir64b")} function attribute, x86 6661Enable/disable the generation of the MOVDIR64B instructions. 6662 6663@item movdiri 6664@itemx no-movdiri 6665@cindex @code{target("movdiri")} function attribute, x86 6666Enable/disable the generation of the MOVDIRI instructions. 6667 6668@item mwaitx 6669@itemx no-mwaitx 6670@cindex @code{target("mwaitx")} function attribute, x86 6671Enable/disable the generation of the MWAITX instructions. 6672 6673@item pclmul 6674@itemx no-pclmul 6675@cindex @code{target("pclmul")} function attribute, x86 6676Enable/disable the generation of the PCLMUL instructions. 6677 6678@item pconfig 6679@itemx no-pconfig 6680@cindex @code{target("pconfig")} function attribute, x86 6681Enable/disable the generation of the PCONFIG instructions. 6682 6683@item pku 6684@itemx no-pku 6685@cindex @code{target("pku")} function attribute, x86 6686Enable/disable the generation of the PKU instructions. 6687 6688@item popcnt 6689@itemx no-popcnt 6690@cindex @code{target("popcnt")} function attribute, x86 6691Enable/disable the generation of the POPCNT instruction. 6692 6693@item prefetchwt1 6694@itemx no-prefetchwt1 6695@cindex @code{target("prefetchwt1")} function attribute, x86 6696Enable/disable the generation of the PREFETCHWT1 instructions. 6697 6698@item prfchw 6699@itemx no-prfchw 6700@cindex @code{target("prfchw")} function attribute, x86 6701Enable/disable the generation of the PREFETCHW instruction. 6702 6703@item ptwrite 6704@itemx no-ptwrite 6705@cindex @code{target("ptwrite")} function attribute, x86 6706Enable/disable the generation of the PTWRITE instructions. 6707 6708@item rdpid 6709@itemx no-rdpid 6710@cindex @code{target("rdpid")} function attribute, x86 6711Enable/disable the generation of the RDPID instructions. 6712 6713@item rdrnd 6714@itemx no-rdrnd 6715@cindex @code{target("rdrnd")} function attribute, x86 6716Enable/disable the generation of the RDRND instructions. 6717 6718@item rdseed 6719@itemx no-rdseed 6720@cindex @code{target("rdseed")} function attribute, x86 6721Enable/disable the generation of the RDSEED instructions. 6722 6723@item rtm 6724@itemx no-rtm 6725@cindex @code{target("rtm")} function attribute, x86 6726Enable/disable the generation of the RTM instructions. 6727 6728@item sahf 6729@itemx no-sahf 6730@cindex @code{target("sahf")} function attribute, x86 6731Enable/disable the generation of the SAHF instructions. 6732 6733@item sgx 6734@itemx no-sgx 6735@cindex @code{target("sgx")} function attribute, x86 6736Enable/disable the generation of the SGX instructions. 6737 6738@item sha 6739@itemx no-sha 6740@cindex @code{target("sha")} function attribute, x86 6741Enable/disable the generation of the SHA instructions. 6742 6743@item shstk 6744@itemx no-shstk 6745@cindex @code{target("shstk")} function attribute, x86 6746Enable/disable the shadow stack built-in functions from CET. 6747 6748@item sse 6749@itemx no-sse 6750@cindex @code{target("sse")} function attribute, x86 6751Enable/disable the generation of the SSE instructions. 6752 6753@item sse2 6754@itemx no-sse2 6755@cindex @code{target("sse2")} function attribute, x86 6756Enable/disable the generation of the SSE2 instructions. 6757 6758@item sse3 6759@itemx no-sse3 6760@cindex @code{target("sse3")} function attribute, x86 6761Enable/disable the generation of the SSE3 instructions. 6762 6763@item sse4 6764@itemx no-sse4 6765@cindex @code{target("sse4")} function attribute, x86 6766Enable/disable the generation of the SSE4 instructions (both SSE4.1 6767and SSE4.2). 6768 6769@item sse4.1 6770@itemx no-sse4.1 6771@cindex @code{target("sse4.1")} function attribute, x86 6772Enable/disable the generation of the sse4.1 instructions. 6773 6774@item sse4.2 6775@itemx no-sse4.2 6776@cindex @code{target("sse4.2")} function attribute, x86 6777Enable/disable the generation of the sse4.2 instructions. 6778 6779@item sse4a 6780@itemx no-sse4a 6781@cindex @code{target("sse4a")} function attribute, x86 6782Enable/disable the generation of the SSE4A instructions. 6783 6784@item ssse3 6785@itemx no-ssse3 6786@cindex @code{target("ssse3")} function attribute, x86 6787Enable/disable the generation of the SSSE3 instructions. 6788 6789@item tbm 6790@itemx no-tbm 6791@cindex @code{target("tbm")} function attribute, x86 6792Enable/disable the generation of the TBM instructions. 6793 6794@item vaes 6795@itemx no-vaes 6796@cindex @code{target("vaes")} function attribute, x86 6797Enable/disable the generation of the VAES instructions. 6798 6799@item vpclmulqdq 6800@itemx no-vpclmulqdq 6801@cindex @code{target("vpclmulqdq")} function attribute, x86 6802Enable/disable the generation of the VPCLMULQDQ instructions. 6803 6804@item waitpkg 6805@itemx no-waitpkg 6806@cindex @code{target("waitpkg")} function attribute, x86 6807Enable/disable the generation of the WAITPKG instructions. 6808 6809@item wbnoinvd 6810@itemx no-wbnoinvd 6811@cindex @code{target("wbnoinvd")} function attribute, x86 6812Enable/disable the generation of the WBNOINVD instructions. 6813 6814@item xop 6815@itemx no-xop 6816@cindex @code{target("xop")} function attribute, x86 6817Enable/disable the generation of the XOP instructions. 6818 6819@item xsave 6820@itemx no-xsave 6821@cindex @code{target("xsave")} function attribute, x86 6822Enable/disable the generation of the XSAVE instructions. 6823 6824@item xsavec 6825@itemx no-xsavec 6826@cindex @code{target("xsavec")} function attribute, x86 6827Enable/disable the generation of the XSAVEC instructions. 6828 6829@item xsaveopt 6830@itemx no-xsaveopt 6831@cindex @code{target("xsaveopt")} function attribute, x86 6832Enable/disable the generation of the XSAVEOPT instructions. 6833 6834@item xsaves 6835@itemx no-xsaves 6836@cindex @code{target("xsaves")} function attribute, x86 6837Enable/disable the generation of the XSAVES instructions. 6838 6839@item amx-tile 6840@itemx no-amx-tile 6841@cindex @code{target("amx-tile")} function attribute, x86 6842Enable/disable the generation of the AMX-TILE instructions. 6843 6844@item amx-int8 6845@itemx no-amx-int8 6846@cindex @code{target("amx-int8")} function attribute, x86 6847Enable/disable the generation of the AMX-INT8 instructions. 6848 6849@item amx-bf16 6850@itemx no-amx-bf16 6851@cindex @code{target("amx-bf16")} function attribute, x86 6852Enable/disable the generation of the AMX-BF16 instructions. 6853 6854@item uintr 6855@itemx no-uintr 6856@cindex @code{target("uintr")} function attribute, x86 6857Enable/disable the generation of the UINTR instructions. 6858 6859@item hreset 6860@itemx no-hreset 6861@cindex @code{target("hreset")} function attribute, x86 6862Enable/disable the generation of the HRESET instruction. 6863 6864@item kl 6865@itemx no-kl 6866@cindex @code{target("kl")} function attribute, x86 6867Enable/disable the generation of the KEYLOCKER instructions. 6868 6869@item widekl 6870@itemx no-widekl 6871@cindex @code{target("widekl")} function attribute, x86 6872Enable/disable the generation of the WIDEKL instructions. 6873 6874@item avxvnni 6875@itemx no-avxvnni 6876@cindex @code{target("avxvnni")} function attribute, x86 6877Enable/disable the generation of the AVXVNNI instructions. 6878 6879@item cld 6880@itemx no-cld 6881@cindex @code{target("cld")} function attribute, x86 6882Enable/disable the generation of the CLD before string moves. 6883 6884@item fancy-math-387 6885@itemx no-fancy-math-387 6886@cindex @code{target("fancy-math-387")} function attribute, x86 6887Enable/disable the generation of the @code{sin}, @code{cos}, and 6888@code{sqrt} instructions on the 387 floating-point unit. 6889 6890@item ieee-fp 6891@itemx no-ieee-fp 6892@cindex @code{target("ieee-fp")} function attribute, x86 6893Enable/disable the generation of floating point that depends on IEEE arithmetic. 6894 6895@item inline-all-stringops 6896@itemx no-inline-all-stringops 6897@cindex @code{target("inline-all-stringops")} function attribute, x86 6898Enable/disable inlining of string operations. 6899 6900@item inline-stringops-dynamically 6901@itemx no-inline-stringops-dynamically 6902@cindex @code{target("inline-stringops-dynamically")} function attribute, x86 6903Enable/disable the generation of the inline code to do small string 6904operations and calling the library routines for large operations. 6905 6906@item align-stringops 6907@itemx no-align-stringops 6908@cindex @code{target("align-stringops")} function attribute, x86 6909Do/do not align destination of inlined string operations. 6910 6911@item recip 6912@itemx no-recip 6913@cindex @code{target("recip")} function attribute, x86 6914Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS 6915instructions followed an additional Newton-Raphson step instead of 6916doing a floating-point division. 6917 6918@item general-regs-only 6919@cindex @code{target("general-regs-only")} function attribute, x86 6920Generate code which uses only the general registers. 6921 6922@item arch=@var{ARCH} 6923@cindex @code{target("arch=@var{ARCH}")} function attribute, x86 6924Specify the architecture to generate code for in compiling the function. 6925 6926@item tune=@var{TUNE} 6927@cindex @code{target("tune=@var{TUNE}")} function attribute, x86 6928Specify the architecture to tune for in compiling the function. 6929 6930@item fpmath=@var{FPMATH} 6931@cindex @code{target("fpmath=@var{FPMATH}")} function attribute, x86 6932Specify which floating-point unit to use. You must specify the 6933@code{target("fpmath=sse,387")} option as 6934@code{target("fpmath=sse+387")} because the comma would separate 6935different options. 6936 6937@item prefer-vector-width=@var{OPT} 6938@cindex @code{prefer-vector-width} function attribute, x86 6939On x86 targets, the @code{prefer-vector-width} attribute informs the 6940compiler to use @var{OPT}-bit vector width in instructions 6941instead of the default on the selected platform. 6942 6943Valid @var{OPT} values are: 6944 6945@table @samp 6946@item none 6947No extra limitations applied to GCC other than defined by the selected platform. 6948 6949@item 128 6950Prefer 128-bit vector width for instructions. 6951 6952@item 256 6953Prefer 256-bit vector width for instructions. 6954 6955@item 512 6956Prefer 512-bit vector width for instructions. 6957@end table 6958 6959On the x86, the inliner does not inline a 6960function that has different target options than the caller, unless the 6961callee has a subset of the target options of the caller. For example 6962a function declared with @code{target("sse3")} can inline a function 6963with @code{target("sse2")}, since @code{-msse3} implies @code{-msse2}. 6964@end table 6965 6966@item indirect_branch("@var{choice}") 6967@cindex @code{indirect_branch} function attribute, x86 6968On x86 targets, the @code{indirect_branch} attribute causes the compiler 6969to convert indirect call and jump with @var{choice}. @samp{keep} 6970keeps indirect call and jump unmodified. @samp{thunk} converts indirect 6971call and jump to call and return thunk. @samp{thunk-inline} converts 6972indirect call and jump to inlined call and return thunk. 6973@samp{thunk-extern} converts indirect call and jump to external call 6974and return thunk provided in a separate object file. 6975 6976@item function_return("@var{choice}") 6977@cindex @code{function_return} function attribute, x86 6978On x86 targets, the @code{function_return} attribute causes the compiler 6979to convert function return with @var{choice}. @samp{keep} keeps function 6980return unmodified. @samp{thunk} converts function return to call and 6981return thunk. @samp{thunk-inline} converts function return to inlined 6982call and return thunk. @samp{thunk-extern} converts function return to 6983external call and return thunk provided in a separate object file. 6984 6985@item nocf_check 6986@cindex @code{nocf_check} function attribute 6987The @code{nocf_check} attribute on a function is used to inform the 6988compiler that the function's prologue should not be instrumented when 6989compiled with the @option{-fcf-protection=branch} option. The 6990compiler assumes that the function's address is a valid target for a 6991control-flow transfer. 6992 6993The @code{nocf_check} attribute on a type of pointer to function is 6994used to inform the compiler that a call through the pointer should 6995not be instrumented when compiled with the 6996@option{-fcf-protection=branch} option. The compiler assumes 6997that the function's address from the pointer is a valid target for 6998a control-flow transfer. A direct function call through a function 6999name is assumed to be a safe call thus direct calls are not 7000instrumented by the compiler. 7001 7002The @code{nocf_check} attribute is applied to an object's type. 7003In case of assignment of a function address or a function pointer to 7004another pointer, the attribute is not carried over from the right-hand 7005object's type; the type of left-hand object stays unchanged. The 7006compiler checks for @code{nocf_check} attribute mismatch and reports 7007a warning in case of mismatch. 7008 7009@smallexample 7010@{ 7011int foo (void) __attribute__(nocf_check); 7012void (*foo1)(void) __attribute__(nocf_check); 7013void (*foo2)(void); 7014 7015/* foo's address is assumed to be valid. */ 7016int 7017foo (void) 7018 7019 /* This call site is not checked for control-flow 7020 validity. */ 7021 (*foo1)(); 7022 7023 /* A warning is issued about attribute mismatch. */ 7024 foo1 = foo2; 7025 7026 /* This call site is still not checked. */ 7027 (*foo1)(); 7028 7029 /* This call site is checked. */ 7030 (*foo2)(); 7031 7032 /* A warning is issued about attribute mismatch. */ 7033 foo2 = foo1; 7034 7035 /* This call site is still checked. */ 7036 (*foo2)(); 7037 7038 return 0; 7039@} 7040@end smallexample 7041 7042@item cf_check 7043@cindex @code{cf_check} function attribute, x86 7044 7045The @code{cf_check} attribute on a function is used to inform the 7046compiler that ENDBR instruction should be placed at the function 7047entry when @option{-fcf-protection=branch} is enabled. 7048 7049@item indirect_return 7050@cindex @code{indirect_return} function attribute, x86 7051 7052The @code{indirect_return} attribute can be applied to a function, 7053as well as variable or type of function pointer to inform the 7054compiler that the function may return via indirect branch. 7055 7056@item fentry_name("@var{name}") 7057@cindex @code{fentry_name} function attribute, x86 7058On x86 targets, the @code{fentry_name} attribute sets the function to 7059call on function entry when function instrumentation is enabled 7060with @option{-pg -mfentry}. When @var{name} is nop then a 5 byte 7061nop sequence is generated. 7062 7063@item fentry_section("@var{name}") 7064@cindex @code{fentry_section} function attribute, x86 7065On x86 targets, the @code{fentry_section} attribute sets the name 7066of the section to record function entry instrumentation calls in when 7067enabled with @option{-pg -mrecord-mcount} 7068 7069@end table 7070 7071@node Xstormy16 Function Attributes 7072@subsection Xstormy16 Function Attributes 7073 7074These function attributes are supported by the Xstormy16 back end: 7075 7076@table @code 7077@item interrupt 7078@cindex @code{interrupt} function attribute, Xstormy16 7079Use this attribute to indicate 7080that the specified function is an interrupt handler. The compiler generates 7081function entry and exit sequences suitable for use in an interrupt handler 7082when this attribute is present. 7083@end table 7084 7085@node Variable Attributes 7086@section Specifying Attributes of Variables 7087@cindex attribute of variables 7088@cindex variable attributes 7089 7090The keyword @code{__attribute__} allows you to specify special properties 7091of variables, function parameters, or structure, union, and, in C++, class 7092members. This @code{__attribute__} keyword is followed by an attribute 7093specification enclosed in double parentheses. Some attributes are currently 7094defined generically for variables. Other attributes are defined for 7095variables on particular target systems. Other attributes are available 7096for functions (@pxref{Function Attributes}), labels (@pxref{Label Attributes}), 7097enumerators (@pxref{Enumerator Attributes}), statements 7098(@pxref{Statement Attributes}), and for types (@pxref{Type Attributes}). 7099Other front ends might define more attributes 7100(@pxref{C++ Extensions,,Extensions to the C++ Language}). 7101 7102@xref{Attribute Syntax}, for details of the exact syntax for using 7103attributes. 7104 7105@menu 7106* Common Variable Attributes:: 7107* ARC Variable Attributes:: 7108* AVR Variable Attributes:: 7109* Blackfin Variable Attributes:: 7110* H8/300 Variable Attributes:: 7111* IA-64 Variable Attributes:: 7112* M32R/D Variable Attributes:: 7113* MeP Variable Attributes:: 7114* Microsoft Windows Variable Attributes:: 7115* MSP430 Variable Attributes:: 7116* Nvidia PTX Variable Attributes:: 7117* PowerPC Variable Attributes:: 7118* RL78 Variable Attributes:: 7119* V850 Variable Attributes:: 7120* x86 Variable Attributes:: 7121* Xstormy16 Variable Attributes:: 7122@end menu 7123 7124@node Common Variable Attributes 7125@subsection Common Variable Attributes 7126 7127The following attributes are supported on most targets. 7128 7129@table @code 7130 7131@item alias ("@var{target}") 7132@cindex @code{alias} variable attribute 7133The @code{alias} variable attribute causes the declaration to be emitted 7134as an alias for another symbol known as an @dfn{alias target}. Except 7135for top-level qualifiers the alias target must have the same type as 7136the alias. For instance, the following 7137 7138@smallexample 7139int var_target; 7140extern int __attribute__ ((alias ("var_target"))) var_alias; 7141@end smallexample 7142 7143@noindent 7144defines @code{var_alias} to be an alias for the @code{var_target} variable. 7145 7146It is an error if the alias target is not defined in the same translation 7147unit as the alias. 7148 7149Note that in the absence of the attribute GCC assumes that distinct 7150declarations with external linkage denote distinct objects. Using both 7151the alias and the alias target to access the same object is undefined 7152in a translation unit without a declaration of the alias with the attribute. 7153 7154This attribute requires assembler and object file support, and may not be 7155available on all targets. 7156 7157@cindex @code{aligned} variable attribute 7158@item aligned 7159@itemx aligned (@var{alignment}) 7160The @code{aligned} attribute specifies a minimum alignment for the variable 7161or structure field, measured in bytes. When specified, @var{alignment} must 7162be an integer constant power of 2. Specifying no @var{alignment} argument 7163implies the maximum alignment for the target, which is often, but by no 7164means always, 8 or 16 bytes. 7165 7166For example, the declaration: 7167 7168@smallexample 7169int x __attribute__ ((aligned (16))) = 0; 7170@end smallexample 7171 7172@noindent 7173causes the compiler to allocate the global variable @code{x} on a 717416-byte boundary. On a 68040, this could be used in conjunction with 7175an @code{asm} expression to access the @code{move16} instruction which 7176requires 16-byte aligned operands. 7177 7178You can also specify the alignment of structure fields. For example, to 7179create a double-word aligned @code{int} pair, you could write: 7180 7181@smallexample 7182struct foo @{ int x[2] __attribute__ ((aligned (8))); @}; 7183@end smallexample 7184 7185@noindent 7186This is an alternative to creating a union with a @code{double} member, 7187which forces the union to be double-word aligned. 7188 7189As in the preceding examples, you can explicitly specify the alignment 7190(in bytes) that you wish the compiler to use for a given variable or 7191structure field. Alternatively, you can leave out the alignment factor 7192and just ask the compiler to align a variable or field to the 7193default alignment for the target architecture you are compiling for. 7194The default alignment is sufficient for all scalar types, but may not be 7195enough for all vector types on a target that supports vector operations. 7196The default alignment is fixed for a particular target ABI. 7197 7198GCC also provides a target specific macro @code{__BIGGEST_ALIGNMENT__}, 7199which is the largest alignment ever used for any data type on the 7200target machine you are compiling for. For example, you could write: 7201 7202@smallexample 7203short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__))); 7204@end smallexample 7205 7206The compiler automatically sets the alignment for the declared 7207variable or field to @code{__BIGGEST_ALIGNMENT__}. Doing this can 7208often make copy operations more efficient, because the compiler can 7209use whatever instructions copy the biggest chunks of memory when 7210performing copies to or from the variables or fields that you have 7211aligned this way. Note that the value of @code{__BIGGEST_ALIGNMENT__} 7212may change depending on command-line options. 7213 7214When used on a struct, or struct member, the @code{aligned} attribute can 7215only increase the alignment; in order to decrease it, the @code{packed} 7216attribute must be specified as well. When used as part of a typedef, the 7217@code{aligned} attribute can both increase and decrease alignment, and 7218specifying the @code{packed} attribute generates a warning. 7219 7220Note that the effectiveness of @code{aligned} attributes for static 7221variables may be limited by inherent limitations in the system linker 7222and/or object file format. On some systems, the linker is 7223only able to arrange for variables to be aligned up to a certain maximum 7224alignment. (For some linkers, the maximum supported alignment may 7225be very very small.) If your linker is only able to align variables 7226up to a maximum of 8-byte alignment, then specifying @code{aligned(16)} 7227in an @code{__attribute__} still only provides you with 8-byte 7228alignment. See your linker documentation for further information. 7229 7230Stack variables are not affected by linker restrictions; GCC can properly 7231align them on any target. 7232 7233The @code{aligned} attribute can also be used for functions 7234(@pxref{Common Function Attributes}.) 7235 7236@cindex @code{warn_if_not_aligned} variable attribute 7237@item warn_if_not_aligned (@var{alignment}) 7238This attribute specifies a threshold for the structure field, measured 7239in bytes. If the structure field is aligned below the threshold, a 7240warning will be issued. For example, the declaration: 7241 7242@smallexample 7243struct foo 7244@{ 7245 int i1; 7246 int i2; 7247 unsigned long long x __attribute__ ((warn_if_not_aligned (16))); 7248@}; 7249@end smallexample 7250 7251@noindent 7252causes the compiler to issue an warning on @code{struct foo}, like 7253@samp{warning: alignment 8 of 'struct foo' is less than 16}. 7254The compiler also issues a warning, like @samp{warning: 'x' offset 72558 in 'struct foo' isn't aligned to 16}, when the structure field has 7256the misaligned offset: 7257 7258@smallexample 7259struct __attribute__ ((aligned (16))) foo 7260@{ 7261 int i1; 7262 int i2; 7263 unsigned long long x __attribute__ ((warn_if_not_aligned (16))); 7264@}; 7265@end smallexample 7266 7267This warning can be disabled by @option{-Wno-if-not-aligned}. 7268The @code{warn_if_not_aligned} attribute can also be used for types 7269(@pxref{Common Type Attributes}.) 7270 7271@item alloc_size (@var{position}) 7272@itemx alloc_size (@var{position-1}, @var{position-2}) 7273@cindex @code{alloc_size} variable attribute 7274The @code{alloc_size} variable attribute may be applied to the declaration 7275of a pointer to a function that returns a pointer and takes at least one 7276argument of an integer type. It indicates that the returned pointer points 7277to an object whose size is given by the function argument at @var{position-1}, 7278or by the product of the arguments at @var{position-1} and @var{position-2}. 7279Meaningful sizes are positive values less than @code{PTRDIFF_MAX}. Other 7280sizes are disagnosed when detected. GCC uses this information to improve 7281the results of @code{__builtin_object_size}. 7282 7283For instance, the following declarations 7284 7285@smallexample 7286typedef __attribute__ ((alloc_size (1, 2))) void* 7287 (*calloc_ptr) (size_t, size_t); 7288typedef __attribute__ ((alloc_size (1))) void* 7289 (*malloc_ptr) (size_t); 7290@end smallexample 7291 7292@noindent 7293specify that @code{calloc_ptr} is a pointer of a function that, like 7294the standard C function @code{calloc}, returns an object whose size 7295is given by the product of arguments 1 and 2, and similarly, that 7296@code{malloc_ptr}, like the standard C function @code{malloc}, 7297returns an object whose size is given by argument 1 to the function. 7298 7299@item cleanup (@var{cleanup_function}) 7300@cindex @code{cleanup} variable attribute 7301The @code{cleanup} attribute runs a function when the variable goes 7302out of scope. This attribute can only be applied to auto function 7303scope variables; it may not be applied to parameters or variables 7304with static storage duration. The function must take one parameter, 7305a pointer to a type compatible with the variable. The return value 7306of the function (if any) is ignored. 7307 7308If @option{-fexceptions} is enabled, then @var{cleanup_function} 7309is run during the stack unwinding that happens during the 7310processing of the exception. Note that the @code{cleanup} attribute 7311does not allow the exception to be caught, only to perform an action. 7312It is undefined what happens if @var{cleanup_function} does not 7313return normally. 7314 7315@item common 7316@itemx nocommon 7317@cindex @code{common} variable attribute 7318@cindex @code{nocommon} variable attribute 7319@opindex fcommon 7320@opindex fno-common 7321The @code{common} attribute requests GCC to place a variable in 7322``common'' storage. The @code{nocommon} attribute requests the 7323opposite---to allocate space for it directly. 7324 7325These attributes override the default chosen by the 7326@option{-fno-common} and @option{-fcommon} flags respectively. 7327 7328@item copy 7329@itemx copy (@var{variable}) 7330@cindex @code{copy} variable attribute 7331The @code{copy} attribute applies the set of attributes with which 7332@var{variable} has been declared to the declaration of the variable 7333to which the attribute is applied. The attribute is designed for 7334libraries that define aliases that are expected to specify the same 7335set of attributes as the aliased symbols. The @code{copy} attribute 7336can be used with variables, functions or types. However, the kind 7337of symbol to which the attribute is applied (either varible or 7338function) must match the kind of symbol to which the argument refers. 7339The @code{copy} attribute copies only syntactic and semantic attributes 7340but not attributes that affect a symbol's linkage or visibility such as 7341@code{alias}, @code{visibility}, or @code{weak}. The @code{deprecated} 7342attribute is also not copied. @xref{Common Function Attributes}. 7343@xref{Common Type Attributes}. 7344 7345@item deprecated 7346@itemx deprecated (@var{msg}) 7347@cindex @code{deprecated} variable attribute 7348The @code{deprecated} attribute results in a warning if the variable 7349is used anywhere in the source file. This is useful when identifying 7350variables that are expected to be removed in a future version of a 7351program. The warning also includes the location of the declaration 7352of the deprecated variable, to enable users to easily find further 7353information about why the variable is deprecated, or what they should 7354do instead. Note that the warning only occurs for uses: 7355 7356@smallexample 7357extern int old_var __attribute__ ((deprecated)); 7358extern int old_var; 7359int new_fn () @{ return old_var; @} 7360@end smallexample 7361 7362@noindent 7363results in a warning on line 3 but not line 2. The optional @var{msg} 7364argument, which must be a string, is printed in the warning if 7365present. 7366 7367The @code{deprecated} attribute can also be used for functions and 7368types (@pxref{Common Function Attributes}, 7369@pxref{Common Type Attributes}). 7370 7371The message attached to the attribute is affected by the setting of 7372the @option{-fmessage-length} option. 7373 7374@item mode (@var{mode}) 7375@cindex @code{mode} variable attribute 7376This attribute specifies the data type for the declaration---whichever 7377type corresponds to the mode @var{mode}. This in effect lets you 7378request an integer or floating-point type according to its width. 7379 7380@xref{Machine Modes,,, gccint, GNU Compiler Collection (GCC) Internals}, 7381for a list of the possible keywords for @var{mode}. 7382You may also specify a mode of @code{byte} or @code{__byte__} to 7383indicate the mode corresponding to a one-byte integer, @code{word} or 7384@code{__word__} for the mode of a one-word integer, and @code{pointer} 7385or @code{__pointer__} for the mode used to represent pointers. 7386 7387@item nonstring 7388@cindex @code{nonstring} variable attribute 7389The @code{nonstring} variable attribute specifies that an object or member 7390declaration with type array of @code{char}, @code{signed char}, or 7391@code{unsigned char}, or pointer to such a type is intended to store 7392character arrays that do not necessarily contain a terminating @code{NUL}. 7393This is useful in detecting uses of such arrays or pointers with functions 7394that expect @code{NUL}-terminated strings, and to avoid warnings when such 7395an array or pointer is used as an argument to a bounded string manipulation 7396function such as @code{strncpy}. For example, without the attribute, GCC 7397will issue a warning for the @code{strncpy} call below because it may 7398truncate the copy without appending the terminating @code{NUL} character. 7399Using the attribute makes it possible to suppress the warning. However, 7400when the array is declared with the attribute the call to @code{strlen} is 7401diagnosed because when the array doesn't contain a @code{NUL}-terminated 7402string the call is undefined. To copy, compare, of search non-string 7403character arrays use the @code{memcpy}, @code{memcmp}, @code{memchr}, 7404and other functions that operate on arrays of bytes. In addition, 7405calling @code{strnlen} and @code{strndup} with such arrays is safe 7406provided a suitable bound is specified, and not diagnosed. 7407 7408@smallexample 7409struct Data 7410@{ 7411 char name [32] __attribute__ ((nonstring)); 7412@}; 7413 7414int f (struct Data *pd, const char *s) 7415@{ 7416 strncpy (pd->name, s, sizeof pd->name); 7417 @dots{} 7418 return strlen (pd->name); // unsafe, gets a warning 7419@} 7420@end smallexample 7421 7422@item packed 7423@cindex @code{packed} variable attribute 7424The @code{packed} attribute specifies that a structure member should have 7425the smallest possible alignment---one bit for a bit-field and one byte 7426otherwise, unless a larger value is specified with the @code{aligned} 7427attribute. The attribute does not apply to non-member objects. 7428 7429For example in the structure below, the member array @code{x} is packed 7430so that it immediately follows @code{a} with no intervening padding: 7431 7432@smallexample 7433struct foo 7434@{ 7435 char a; 7436 int x[2] __attribute__ ((packed)); 7437@}; 7438@end smallexample 7439 7440@emph{Note:} The 4.1, 4.2 and 4.3 series of GCC ignore the 7441@code{packed} attribute on bit-fields of type @code{char}. This has 7442been fixed in GCC 4.4 but the change can lead to differences in the 7443structure layout. See the documentation of 7444@option{-Wpacked-bitfield-compat} for more information. 7445 7446@item section ("@var{section-name}") 7447@cindex @code{section} variable attribute 7448Normally, the compiler places the objects it generates in sections like 7449@code{data} and @code{bss}. Sometimes, however, you need additional sections, 7450or you need certain particular variables to appear in special sections, 7451for example to map to special hardware. The @code{section} 7452attribute specifies that a variable (or function) lives in a particular 7453section. For example, this small program uses several specific section names: 7454 7455@smallexample 7456struct duart a __attribute__ ((section ("DUART_A"))) = @{ 0 @}; 7457struct duart b __attribute__ ((section ("DUART_B"))) = @{ 0 @}; 7458char stack[10000] __attribute__ ((section ("STACK"))) = @{ 0 @}; 7459int init_data __attribute__ ((section ("INITDATA"))); 7460 7461main() 7462@{ 7463 /* @r{Initialize stack pointer} */ 7464 init_sp (stack + sizeof (stack)); 7465 7466 /* @r{Initialize initialized data} */ 7467 memcpy (&init_data, &data, &edata - &data); 7468 7469 /* @r{Turn on the serial ports} */ 7470 init_duart (&a); 7471 init_duart (&b); 7472@} 7473@end smallexample 7474 7475@noindent 7476Use the @code{section} attribute with 7477@emph{global} variables and not @emph{local} variables, 7478as shown in the example. 7479 7480You may use the @code{section} attribute with initialized or 7481uninitialized global variables but the linker requires 7482each object be defined once, with the exception that uninitialized 7483variables tentatively go in the @code{common} (or @code{bss}) section 7484and can be multiply ``defined''. Using the @code{section} attribute 7485changes what section the variable goes into and may cause the 7486linker to issue an error if an uninitialized variable has multiple 7487definitions. You can force a variable to be initialized with the 7488@option{-fno-common} flag or the @code{nocommon} attribute. 7489 7490Some file formats do not support arbitrary sections so the @code{section} 7491attribute is not available on all platforms. 7492If you need to map the entire contents of a module to a particular 7493section, consider using the facilities of the linker instead. 7494 7495@item tls_model ("@var{tls_model}") 7496@cindex @code{tls_model} variable attribute 7497The @code{tls_model} attribute sets thread-local storage model 7498(@pxref{Thread-Local}) of a particular @code{__thread} variable, 7499overriding @option{-ftls-model=} command-line switch on a per-variable 7500basis. 7501The @var{tls_model} argument should be one of @code{global-dynamic}, 7502@code{local-dynamic}, @code{initial-exec} or @code{local-exec}. 7503 7504Not all targets support this attribute. 7505 7506@item unused 7507@cindex @code{unused} variable attribute 7508This attribute, attached to a variable, means that the variable is meant 7509to be possibly unused. GCC does not produce a warning for this 7510variable. 7511 7512@item used 7513@cindex @code{used} variable attribute 7514This attribute, attached to a variable with static storage, means that 7515the variable must be emitted even if it appears that the variable is not 7516referenced. 7517 7518When applied to a static data member of a C++ class template, the 7519attribute also means that the member is instantiated if the 7520class itself is instantiated. 7521 7522@item retain 7523@cindex @code{retain} variable attribute 7524For ELF targets that support the GNU or FreeBSD OSABIs, this attribute 7525will save the variable from linker garbage collection. To support 7526this behavior, variables that have not been placed in specific sections 7527(e.g. by the @code{section} attribute, or the @code{-fdata-sections} option), 7528will be placed in new, unique sections. 7529 7530This additional functionality requires Binutils version 2.36 or later. 7531 7532@item vector_size (@var{bytes}) 7533@cindex @code{vector_size} variable attribute 7534This attribute specifies the vector size for the type of the declared 7535variable, measured in bytes. The type to which it applies is known as 7536the @dfn{base type}. The @var{bytes} argument must be a positive 7537power-of-two multiple of the base type size. For example, the declaration: 7538 7539@smallexample 7540int foo __attribute__ ((vector_size (16))); 7541@end smallexample 7542 7543@noindent 7544causes the compiler to set the mode for @code{foo}, to be 16 bytes, 7545divided into @code{int} sized units. Assuming a 32-bit @code{int}, 7546@code{foo}'s type is a vector of four units of four bytes each, and 7547the corresponding mode of @code{foo} is @code{V4SI}. 7548@xref{Vector Extensions}, for details of manipulating vector variables. 7549 7550This attribute is only applicable to integral and floating scalars, 7551although arrays, pointers, and function return values are allowed in 7552conjunction with this construct. 7553 7554Aggregates with this attribute are invalid, even if they are of the same 7555size as a corresponding scalar. For example, the declaration: 7556 7557@smallexample 7558struct S @{ int a; @}; 7559struct S __attribute__ ((vector_size (16))) foo; 7560@end smallexample 7561 7562@noindent 7563is invalid even if the size of the structure is the same as the size of 7564the @code{int}. 7565 7566@item visibility ("@var{visibility_type}") 7567@cindex @code{visibility} variable attribute 7568This attribute affects the linkage of the declaration to which it is attached. 7569The @code{visibility} attribute is described in 7570@ref{Common Function Attributes}. 7571 7572@item weak 7573@cindex @code{weak} variable attribute 7574The @code{weak} attribute is described in 7575@ref{Common Function Attributes}. 7576 7577@item noinit 7578@cindex @code{noinit} variable attribute 7579Any data with the @code{noinit} attribute will not be initialized by 7580the C runtime startup code, or the program loader. Not initializing 7581data in this way can reduce program startup times. 7582 7583This attribute is specific to ELF targets and relies on the linker 7584script to place sections with the @code{.noinit} prefix in the right 7585location. 7586 7587@item persistent 7588@cindex @code{persistent} variable attribute 7589Any data with the @code{persistent} attribute will not be initialized by 7590the C runtime startup code, but will be initialized by the program 7591loader. This enables the value of the variable to @samp{persist} 7592between processor resets. 7593 7594This attribute is specific to ELF targets and relies on the linker 7595script to place the sections with the @code{.persistent} prefix in the 7596right location. Specifically, some type of non-volatile, writeable 7597memory is required. 7598 7599@item objc_nullability (@var{nullability kind}) @r{(Objective-C and Objective-C++ only)} 7600@cindex @code{objc_nullability} variable attribute 7601This attribute applies to pointer variables only. It allows marking the 7602pointer with one of four possible values describing the conditions under 7603which the pointer might have a @code{nil} value. In most cases, the 7604attribute is intended to be an internal representation for property and 7605method nullability (specified by language keywords); it is not recommended 7606to use it directly. 7607 7608When @var{nullability kind} is @code{"unspecified"} or @code{0}, nothing is 7609known about the conditions in which the pointer might be @code{nil}. Making 7610this state specific serves to avoid false positives in diagnostics. 7611 7612When @var{nullability kind} is @code{"nonnull"} or @code{1}, the pointer has 7613no meaning if it is @code{nil} and thus the compiler is free to emit 7614diagnostics if it can be determined that the value will be @code{nil}. 7615 7616When @var{nullability kind} is @code{"nullable"} or @code{2}, the pointer might 7617be @code{nil} and carry meaning as such. 7618 7619When @var{nullability kind} is @code{"resettable"} or @code{3} (used only in 7620the context of property attribute lists) this describes the case in which a 7621property setter may take the value @code{nil} (which perhaps causes the 7622property to be reset in some manner to a default) but for which the property 7623getter will never validly return @code{nil}. 7624 7625@end table 7626 7627@node ARC Variable Attributes 7628@subsection ARC Variable Attributes 7629 7630@table @code 7631@item aux 7632@cindex @code{aux} variable attribute, ARC 7633The @code{aux} attribute is used to directly access the ARC's 7634auxiliary register space from C. The auxilirary register number is 7635given via attribute argument. 7636 7637@end table 7638 7639@node AVR Variable Attributes 7640@subsection AVR Variable Attributes 7641 7642@table @code 7643@item progmem 7644@cindex @code{progmem} variable attribute, AVR 7645The @code{progmem} attribute is used on the AVR to place read-only 7646data in the non-volatile program memory (flash). The @code{progmem} 7647attribute accomplishes this by putting respective variables into a 7648section whose name starts with @code{.progmem}. 7649 7650This attribute works similar to the @code{section} attribute 7651but adds additional checking. 7652 7653@table @asis 7654@item @bullet{}@tie{} Ordinary AVR cores with 32 general purpose registers: 7655@code{progmem} affects the location 7656of the data but not how this data is accessed. 7657In order to read data located with the @code{progmem} attribute 7658(inline) assembler must be used. 7659@smallexample 7660/* Use custom macros from @w{@uref{http://nongnu.org/avr-libc/user-manual/,AVR-LibC}} */ 7661#include <avr/pgmspace.h> 7662 7663/* Locate var in flash memory */ 7664const int var[2] PROGMEM = @{ 1, 2 @}; 7665 7666int read_var (int i) 7667@{ 7668 /* Access var[] by accessor macro from avr/pgmspace.h */ 7669 return (int) pgm_read_word (& var[i]); 7670@} 7671@end smallexample 7672 7673AVR is a Harvard architecture processor and data and read-only data 7674normally resides in the data memory (RAM). 7675 7676See also the @ref{AVR Named Address Spaces} section for 7677an alternate way to locate and access data in flash memory. 7678 7679@item @bullet{}@tie{} AVR cores with flash memory visible in the RAM address range: 7680On such devices, there is no need for attribute @code{progmem} or 7681@ref{AVR Named Address Spaces,,@code{__flash}} qualifier at all. 7682Just use standard C / C++. The compiler will generate @code{LD*} 7683instructions. As flash memory is visible in the RAM address range, 7684and the default linker script does @emph{not} locate @code{.rodata} in 7685RAM, no special features are needed in order not to waste RAM for 7686read-only data or to read from flash. You might even get slightly better 7687performance by 7688avoiding @code{progmem} and @code{__flash}. This applies to devices from 7689families @code{avrtiny} and @code{avrxmega3}, see @ref{AVR Options} for 7690an overview. 7691 7692@item @bullet{}@tie{}Reduced AVR Tiny cores like ATtiny40: 7693The compiler adds @code{0x4000} 7694to the addresses of objects and declarations in @code{progmem} and locates 7695the objects in flash memory, namely in section @code{.progmem.data}. 7696The offset is needed because the flash memory is visible in the RAM 7697address space starting at address @code{0x4000}. 7698 7699Data in @code{progmem} can be accessed by means of ordinary C@tie{}code, 7700no special functions or macros are needed. 7701 7702@smallexample 7703/* var is located in flash memory */ 7704extern const int var[2] __attribute__((progmem)); 7705 7706int read_var (int i) 7707@{ 7708 return var[i]; 7709@} 7710@end smallexample 7711 7712Please notice that on these devices, there is no need for @code{progmem} 7713at all. 7714 7715@end table 7716 7717@item io 7718@itemx io (@var{addr}) 7719@cindex @code{io} variable attribute, AVR 7720Variables with the @code{io} attribute are used to address 7721memory-mapped peripherals in the io address range. 7722If an address is specified, the variable 7723is assigned that address, and the value is interpreted as an 7724address in the data address space. 7725Example: 7726 7727@smallexample 7728volatile int porta __attribute__((io (0x22))); 7729@end smallexample 7730 7731The address specified in the address in the data address range. 7732 7733Otherwise, the variable it is not assigned an address, but the 7734compiler will still use in/out instructions where applicable, 7735assuming some other module assigns an address in the io address range. 7736Example: 7737 7738@smallexample 7739extern volatile int porta __attribute__((io)); 7740@end smallexample 7741 7742@item io_low 7743@itemx io_low (@var{addr}) 7744@cindex @code{io_low} variable attribute, AVR 7745This is like the @code{io} attribute, but additionally it informs the 7746compiler that the object lies in the lower half of the I/O area, 7747allowing the use of @code{cbi}, @code{sbi}, @code{sbic} and @code{sbis} 7748instructions. 7749 7750@item address 7751@itemx address (@var{addr}) 7752@cindex @code{address} variable attribute, AVR 7753Variables with the @code{address} attribute are used to address 7754memory-mapped peripherals that may lie outside the io address range. 7755 7756@smallexample 7757volatile int porta __attribute__((address (0x600))); 7758@end smallexample 7759 7760@item absdata 7761@cindex @code{absdata} variable attribute, AVR 7762Variables in static storage and with the @code{absdata} attribute can 7763be accessed by the @code{LDS} and @code{STS} instructions which take 7764absolute addresses. 7765 7766@itemize @bullet 7767@item 7768This attribute is only supported for the reduced AVR Tiny core 7769like ATtiny40. 7770 7771@item 7772You must make sure that respective data is located in the 7773address range @code{0x40}@dots{}@code{0xbf} accessible by 7774@code{LDS} and @code{STS}. One way to achieve this as an 7775appropriate linker description file. 7776 7777@item 7778If the location does not fit the address range of @code{LDS} 7779and @code{STS}, there is currently (Binutils 2.26) just an unspecific 7780warning like 7781@quotation 7782@code{module.c:(.text+0x1c): warning: internal error: out of range error} 7783@end quotation 7784 7785@end itemize 7786 7787See also the @option{-mabsdata} @ref{AVR Options,command-line option}. 7788 7789@end table 7790 7791@node Blackfin Variable Attributes 7792@subsection Blackfin Variable Attributes 7793 7794Three attributes are currently defined for the Blackfin. 7795 7796@table @code 7797@item l1_data 7798@itemx l1_data_A 7799@itemx l1_data_B 7800@cindex @code{l1_data} variable attribute, Blackfin 7801@cindex @code{l1_data_A} variable attribute, Blackfin 7802@cindex @code{l1_data_B} variable attribute, Blackfin 7803Use these attributes on the Blackfin to place the variable into L1 Data SRAM. 7804Variables with @code{l1_data} attribute are put into the specific section 7805named @code{.l1.data}. Those with @code{l1_data_A} attribute are put into 7806the specific section named @code{.l1.data.A}. Those with @code{l1_data_B} 7807attribute are put into the specific section named @code{.l1.data.B}. 7808 7809@item l2 7810@cindex @code{l2} variable attribute, Blackfin 7811Use this attribute on the Blackfin to place the variable into L2 SRAM. 7812Variables with @code{l2} attribute are put into the specific section 7813named @code{.l2.data}. 7814@end table 7815 7816@node H8/300 Variable Attributes 7817@subsection H8/300 Variable Attributes 7818 7819These variable attributes are available for H8/300 targets: 7820 7821@table @code 7822@item eightbit_data 7823@cindex @code{eightbit_data} variable attribute, H8/300 7824@cindex eight-bit data on the H8/300, H8/300H, and H8S 7825Use this attribute on the H8/300, H8/300H, and H8S to indicate that the specified 7826variable should be placed into the eight-bit data section. 7827The compiler generates more efficient code for certain operations 7828on data in the eight-bit data area. Note the eight-bit data area is limited to 7829256 bytes of data. 7830 7831You must use GAS and GLD from GNU binutils version 2.7 or later for 7832this attribute to work correctly. 7833 7834@item tiny_data 7835@cindex @code{tiny_data} variable attribute, H8/300 7836@cindex tiny data section on the H8/300H and H8S 7837Use this attribute on the H8/300H and H8S to indicate that the specified 7838variable should be placed into the tiny data section. 7839The compiler generates more efficient code for loads and stores 7840on data in the tiny data section. Note the tiny data area is limited to 7841slightly under 32KB of data. 7842 7843@end table 7844 7845@node IA-64 Variable Attributes 7846@subsection IA-64 Variable Attributes 7847 7848The IA-64 back end supports the following variable attribute: 7849 7850@table @code 7851@item model (@var{model-name}) 7852@cindex @code{model} variable attribute, IA-64 7853 7854On IA-64, use this attribute to set the addressability of an object. 7855At present, the only supported identifier for @var{model-name} is 7856@code{small}, indicating addressability via ``small'' (22-bit) 7857addresses (so that their addresses can be loaded with the @code{addl} 7858instruction). Caveat: such addressing is by definition not position 7859independent and hence this attribute must not be used for objects 7860defined by shared libraries. 7861 7862@end table 7863 7864@node M32R/D Variable Attributes 7865@subsection M32R/D Variable Attributes 7866 7867One attribute is currently defined for the M32R/D@. 7868 7869@table @code 7870@item model (@var{model-name}) 7871@cindex @code{model-name} variable attribute, M32R/D 7872@cindex variable addressability on the M32R/D 7873Use this attribute on the M32R/D to set the addressability of an object. 7874The identifier @var{model-name} is one of @code{small}, @code{medium}, 7875or @code{large}, representing each of the code models. 7876 7877Small model objects live in the lower 16MB of memory (so that their 7878addresses can be loaded with the @code{ld24} instruction). 7879 7880Medium and large model objects may live anywhere in the 32-bit address space 7881(the compiler generates @code{seth/add3} instructions to load their 7882addresses). 7883@end table 7884 7885@node MeP Variable Attributes 7886@subsection MeP Variable Attributes 7887 7888The MeP target has a number of addressing modes and busses. The 7889@code{near} space spans the standard memory space's first 16 megabytes 7890(24 bits). The @code{far} space spans the entire 32-bit memory space. 7891The @code{based} space is a 128-byte region in the memory space that 7892is addressed relative to the @code{$tp} register. The @code{tiny} 7893space is a 65536-byte region relative to the @code{$gp} register. In 7894addition to these memory regions, the MeP target has a separate 16-bit 7895control bus which is specified with @code{cb} attributes. 7896 7897@table @code 7898 7899@item based 7900@cindex @code{based} variable attribute, MeP 7901Any variable with the @code{based} attribute is assigned to the 7902@code{.based} section, and is accessed with relative to the 7903@code{$tp} register. 7904 7905@item tiny 7906@cindex @code{tiny} variable attribute, MeP 7907Likewise, the @code{tiny} attribute assigned variables to the 7908@code{.tiny} section, relative to the @code{$gp} register. 7909 7910@item near 7911@cindex @code{near} variable attribute, MeP 7912Variables with the @code{near} attribute are assumed to have addresses 7913that fit in a 24-bit addressing mode. This is the default for large 7914variables (@code{-mtiny=4} is the default) but this attribute can 7915override @code{-mtiny=} for small variables, or override @code{-ml}. 7916 7917@item far 7918@cindex @code{far} variable attribute, MeP 7919Variables with the @code{far} attribute are addressed using a full 792032-bit address. Since this covers the entire memory space, this 7921allows modules to make no assumptions about where variables might be 7922stored. 7923 7924@item io 7925@cindex @code{io} variable attribute, MeP 7926@itemx io (@var{addr}) 7927Variables with the @code{io} attribute are used to address 7928memory-mapped peripherals. If an address is specified, the variable 7929is assigned that address, else it is not assigned an address (it is 7930assumed some other module assigns an address). Example: 7931 7932@smallexample 7933int timer_count __attribute__((io(0x123))); 7934@end smallexample 7935 7936@item cb 7937@itemx cb (@var{addr}) 7938@cindex @code{cb} variable attribute, MeP 7939Variables with the @code{cb} attribute are used to access the control 7940bus, using special instructions. @code{addr} indicates the control bus 7941address. Example: 7942 7943@smallexample 7944int cpu_clock __attribute__((cb(0x123))); 7945@end smallexample 7946 7947@end table 7948 7949@node Microsoft Windows Variable Attributes 7950@subsection Microsoft Windows Variable Attributes 7951 7952You can use these attributes on Microsoft Windows targets. 7953@ref{x86 Variable Attributes} for additional Windows compatibility 7954attributes available on all x86 targets. 7955 7956@table @code 7957@item dllimport 7958@itemx dllexport 7959@cindex @code{dllimport} variable attribute 7960@cindex @code{dllexport} variable attribute 7961The @code{dllimport} and @code{dllexport} attributes are described in 7962@ref{Microsoft Windows Function Attributes}. 7963 7964@item selectany 7965@cindex @code{selectany} variable attribute 7966The @code{selectany} attribute causes an initialized global variable to 7967have link-once semantics. When multiple definitions of the variable are 7968encountered by the linker, the first is selected and the remainder are 7969discarded. Following usage by the Microsoft compiler, the linker is told 7970@emph{not} to warn about size or content differences of the multiple 7971definitions. 7972 7973Although the primary usage of this attribute is for POD types, the 7974attribute can also be applied to global C++ objects that are initialized 7975by a constructor. In this case, the static initialization and destruction 7976code for the object is emitted in each translation defining the object, 7977but the calls to the constructor and destructor are protected by a 7978link-once guard variable. 7979 7980The @code{selectany} attribute is only available on Microsoft Windows 7981targets. You can use @code{__declspec (selectany)} as a synonym for 7982@code{__attribute__ ((selectany))} for compatibility with other 7983compilers. 7984 7985@item shared 7986@cindex @code{shared} variable attribute 7987On Microsoft Windows, in addition to putting variable definitions in a named 7988section, the section can also be shared among all running copies of an 7989executable or DLL@. For example, this small program defines shared data 7990by putting it in a named section @code{shared} and marking the section 7991shareable: 7992 7993@smallexample 7994int foo __attribute__((section ("shared"), shared)) = 0; 7995 7996int 7997main() 7998@{ 7999 /* @r{Read and write foo. All running 8000 copies see the same value.} */ 8001 return 0; 8002@} 8003@end smallexample 8004 8005@noindent 8006You may only use the @code{shared} attribute along with @code{section} 8007attribute with a fully-initialized global definition because of the way 8008linkers work. See @code{section} attribute for more information. 8009 8010The @code{shared} attribute is only available on Microsoft Windows@. 8011 8012@end table 8013 8014@node MSP430 Variable Attributes 8015@subsection MSP430 Variable Attributes 8016 8017@table @code 8018@item upper 8019@itemx either 8020@cindex @code{upper} variable attribute, MSP430 8021@cindex @code{either} variable attribute, MSP430 8022These attributes are the same as the MSP430 function attributes of the 8023same name (@pxref{MSP430 Function Attributes}). 8024 8025@item lower 8026@cindex @code{lower} variable attribute, MSP430 8027This option behaves mostly the same as the MSP430 function attribute of the 8028same name (@pxref{MSP430 Function Attributes}), but it has some additional 8029functionality. 8030 8031If @option{-mdata-region=}@{@code{upper,either,none}@} has been passed, or 8032the @code{section} attribute is applied to a variable, the compiler will 8033generate 430X instructions to handle it. This is because the compiler has 8034to assume that the variable could get placed in the upper memory region 8035(above address 0xFFFF). Marking the variable with the @code{lower} attribute 8036informs the compiler that the variable will be placed in lower memory so it 8037is safe to use 430 instructions to handle it. 8038 8039In the case of the @code{section} attribute, the section name given 8040will be used, and the @code{.lower} prefix will not be added. 8041 8042@end table 8043 8044@node Nvidia PTX Variable Attributes 8045@subsection Nvidia PTX Variable Attributes 8046 8047These variable attributes are supported by the Nvidia PTX back end: 8048 8049@table @code 8050@item shared 8051@cindex @code{shared} attribute, Nvidia PTX 8052Use this attribute to place a variable in the @code{.shared} memory space. 8053This memory space is private to each cooperative thread array; only threads 8054within one thread block refer to the same instance of the variable. 8055The runtime does not initialize variables in this memory space. 8056@end table 8057 8058@node PowerPC Variable Attributes 8059@subsection PowerPC Variable Attributes 8060 8061Three attributes currently are defined for PowerPC configurations: 8062@code{altivec}, @code{ms_struct} and @code{gcc_struct}. 8063 8064@cindex @code{ms_struct} variable attribute, PowerPC 8065@cindex @code{gcc_struct} variable attribute, PowerPC 8066For full documentation of the struct attributes please see the 8067documentation in @ref{x86 Variable Attributes}. 8068 8069@cindex @code{altivec} variable attribute, PowerPC 8070For documentation of @code{altivec} attribute please see the 8071documentation in @ref{PowerPC Type Attributes}. 8072 8073@node RL78 Variable Attributes 8074@subsection RL78 Variable Attributes 8075 8076@cindex @code{saddr} variable attribute, RL78 8077The RL78 back end supports the @code{saddr} variable attribute. This 8078specifies placement of the corresponding variable in the SADDR area, 8079which can be accessed more efficiently than the default memory region. 8080 8081@node V850 Variable Attributes 8082@subsection V850 Variable Attributes 8083 8084These variable attributes are supported by the V850 back end: 8085 8086@table @code 8087 8088@item sda 8089@cindex @code{sda} variable attribute, V850 8090Use this attribute to explicitly place a variable in the small data area, 8091which can hold up to 64 kilobytes. 8092 8093@item tda 8094@cindex @code{tda} variable attribute, V850 8095Use this attribute to explicitly place a variable in the tiny data area, 8096which can hold up to 256 bytes in total. 8097 8098@item zda 8099@cindex @code{zda} variable attribute, V850 8100Use this attribute to explicitly place a variable in the first 32 kilobytes 8101of memory. 8102@end table 8103 8104@node x86 Variable Attributes 8105@subsection x86 Variable Attributes 8106 8107Two attributes are currently defined for x86 configurations: 8108@code{ms_struct} and @code{gcc_struct}. 8109 8110@table @code 8111@item ms_struct 8112@itemx gcc_struct 8113@cindex @code{ms_struct} variable attribute, x86 8114@cindex @code{gcc_struct} variable attribute, x86 8115 8116If @code{packed} is used on a structure, or if bit-fields are used, 8117it may be that the Microsoft ABI lays out the structure differently 8118than the way GCC normally does. Particularly when moving packed 8119data between functions compiled with GCC and the native Microsoft compiler 8120(either via function call or as data in a file), it may be necessary to access 8121either format. 8122 8123The @code{ms_struct} and @code{gcc_struct} attributes correspond 8124to the @option{-mms-bitfields} and @option{-mno-ms-bitfields} 8125command-line options, respectively; 8126see @ref{x86 Options}, for details of how structure layout is affected. 8127@xref{x86 Type Attributes}, for information about the corresponding 8128attributes on types. 8129 8130@end table 8131 8132@node Xstormy16 Variable Attributes 8133@subsection Xstormy16 Variable Attributes 8134 8135One attribute is currently defined for xstormy16 configurations: 8136@code{below100}. 8137 8138@table @code 8139@item below100 8140@cindex @code{below100} variable attribute, Xstormy16 8141 8142If a variable has the @code{below100} attribute (@code{BELOW100} is 8143allowed also), GCC places the variable in the first 0x100 bytes of 8144memory and use special opcodes to access it. Such variables are 8145placed in either the @code{.bss_below100} section or the 8146@code{.data_below100} section. 8147 8148@end table 8149 8150@node Type Attributes 8151@section Specifying Attributes of Types 8152@cindex attribute of types 8153@cindex type attributes 8154 8155The keyword @code{__attribute__} allows you to specify various special 8156properties of types. Some type attributes apply only to structure and 8157union types, and in C++, also class types, while others can apply to 8158any type defined via a @code{typedef} declaration. Unless otherwise 8159specified, the same restrictions and effects apply to attributes regardless 8160of whether a type is a trivial structure or a C++ class with user-defined 8161constructors, destructors, or a copy assignment. 8162 8163Other attributes are defined for functions (@pxref{Function Attributes}), 8164labels (@pxref{Label Attributes}), enumerators (@pxref{Enumerator 8165Attributes}), statements (@pxref{Statement Attributes}), and for variables 8166(@pxref{Variable Attributes}). 8167 8168The @code{__attribute__} keyword is followed by an attribute specification 8169enclosed in double parentheses. 8170 8171You may specify type attributes in an enum, struct or union type 8172declaration or definition by placing them immediately after the 8173@code{struct}, @code{union} or @code{enum} keyword. You can also place 8174them just past the closing curly brace of the definition, but this is less 8175preferred because logically the type should be fully defined at 8176the closing brace. 8177 8178You can also include type attributes in a @code{typedef} declaration. 8179@xref{Attribute Syntax}, for details of the exact syntax for using 8180attributes. 8181 8182@menu 8183* Common Type Attributes:: 8184* ARC Type Attributes:: 8185* ARM Type Attributes:: 8186* MeP Type Attributes:: 8187* PowerPC Type Attributes:: 8188* x86 Type Attributes:: 8189@end menu 8190 8191@node Common Type Attributes 8192@subsection Common Type Attributes 8193 8194The following type attributes are supported on most targets. 8195 8196@table @code 8197@cindex @code{aligned} type attribute 8198@item aligned 8199@itemx aligned (@var{alignment}) 8200The @code{aligned} attribute specifies a minimum alignment (in bytes) for 8201variables of the specified type. When specified, @var{alignment} must be 8202a power of 2. Specifying no @var{alignment} argument implies the maximum 8203alignment for the target, which is often, but by no means always, 8 or 16 8204bytes. For example, the declarations: 8205 8206@smallexample 8207struct __attribute__ ((aligned (8))) S @{ short f[3]; @}; 8208typedef int more_aligned_int __attribute__ ((aligned (8))); 8209@end smallexample 8210 8211@noindent 8212force the compiler to ensure (as far as it can) that each variable whose 8213type is @code{struct S} or @code{more_aligned_int} is allocated and 8214aligned @emph{at least} on a 8-byte boundary. On a SPARC, having all 8215variables of type @code{struct S} aligned to 8-byte boundaries allows 8216the compiler to use the @code{ldd} and @code{std} (doubleword load and 8217store) instructions when copying one variable of type @code{struct S} to 8218another, thus improving run-time efficiency. 8219 8220Note that the alignment of any given @code{struct} or @code{union} type 8221is required by the ISO C standard to be at least a perfect multiple of 8222the lowest common multiple of the alignments of all of the members of 8223the @code{struct} or @code{union} in question. This means that you @emph{can} 8224effectively adjust the alignment of a @code{struct} or @code{union} 8225type by attaching an @code{aligned} attribute to any one of the members 8226of such a type, but the notation illustrated in the example above is a 8227more obvious, intuitive, and readable way to request the compiler to 8228adjust the alignment of an entire @code{struct} or @code{union} type. 8229 8230As in the preceding example, you can explicitly specify the alignment 8231(in bytes) that you wish the compiler to use for a given @code{struct} 8232or @code{union} type. Alternatively, you can leave out the alignment factor 8233and just ask the compiler to align a type to the maximum 8234useful alignment for the target machine you are compiling for. For 8235example, you could write: 8236 8237@smallexample 8238struct __attribute__ ((aligned)) S @{ short f[3]; @}; 8239@end smallexample 8240 8241Whenever you leave out the alignment factor in an @code{aligned} 8242attribute specification, the compiler automatically sets the alignment 8243for the type to the largest alignment that is ever used for any data 8244type on the target machine you are compiling for. Doing this can often 8245make copy operations more efficient, because the compiler can use 8246whatever instructions copy the biggest chunks of memory when performing 8247copies to or from the variables that have types that you have aligned 8248this way. 8249 8250In the example above, if the size of each @code{short} is 2 bytes, then 8251the size of the entire @code{struct S} type is 6 bytes. The smallest 8252power of two that is greater than or equal to that is 8, so the 8253compiler sets the alignment for the entire @code{struct S} type to 8 8254bytes. 8255 8256Note that although you can ask the compiler to select a time-efficient 8257alignment for a given type and then declare only individual stand-alone 8258objects of that type, the compiler's ability to select a time-efficient 8259alignment is primarily useful only when you plan to create arrays of 8260variables having the relevant (efficiently aligned) type. If you 8261declare or use arrays of variables of an efficiently-aligned type, then 8262it is likely that your program also does pointer arithmetic (or 8263subscripting, which amounts to the same thing) on pointers to the 8264relevant type, and the code that the compiler generates for these 8265pointer arithmetic operations is often more efficient for 8266efficiently-aligned types than for other types. 8267 8268Note that the effectiveness of @code{aligned} attributes may be limited 8269by inherent limitations in your linker. On many systems, the linker is 8270only able to arrange for variables to be aligned up to a certain maximum 8271alignment. (For some linkers, the maximum supported alignment may 8272be very very small.) If your linker is only able to align variables 8273up to a maximum of 8-byte alignment, then specifying @code{aligned (16)} 8274in an @code{__attribute__} still only provides you with 8-byte 8275alignment. See your linker documentation for further information. 8276 8277When used on a struct, or struct member, the @code{aligned} attribute can 8278only increase the alignment; in order to decrease it, the @code{packed} 8279attribute must be specified as well. When used as part of a typedef, the 8280@code{aligned} attribute can both increase and decrease alignment, and 8281specifying the @code{packed} attribute generates a warning. 8282 8283@cindex @code{warn_if_not_aligned} type attribute 8284@item warn_if_not_aligned (@var{alignment}) 8285This attribute specifies a threshold for the structure field, measured 8286in bytes. If the structure field is aligned below the threshold, a 8287warning will be issued. For example, the declaration: 8288 8289@smallexample 8290typedef unsigned long long __u64 8291 __attribute__((aligned (4), warn_if_not_aligned (8))); 8292 8293struct foo 8294@{ 8295 int i1; 8296 int i2; 8297 __u64 x; 8298@}; 8299@end smallexample 8300 8301@noindent 8302causes the compiler to issue an warning on @code{struct foo}, like 8303@samp{warning: alignment 4 of 'struct foo' is less than 8}. 8304It is used to define @code{struct foo} in such a way that 8305@code{struct foo} has the same layout and the structure field @code{x} 8306has the same alignment when @code{__u64} is aligned at either 4 or 83078 bytes. Align @code{struct foo} to 8 bytes: 8308 8309@smallexample 8310struct __attribute__ ((aligned (8))) foo 8311@{ 8312 int i1; 8313 int i2; 8314 __u64 x; 8315@}; 8316@end smallexample 8317 8318@noindent 8319silences the warning. The compiler also issues a warning, like 8320@samp{warning: 'x' offset 12 in 'struct foo' isn't aligned to 8}, 8321when the structure field has the misaligned offset: 8322 8323@smallexample 8324struct __attribute__ ((aligned (8))) foo 8325@{ 8326 int i1; 8327 int i2; 8328 int i3; 8329 __u64 x; 8330@}; 8331@end smallexample 8332 8333This warning can be disabled by @option{-Wno-if-not-aligned}. 8334 8335@item alloc_size (@var{position}) 8336@itemx alloc_size (@var{position-1}, @var{position-2}) 8337@cindex @code{alloc_size} type attribute 8338The @code{alloc_size} type attribute may be applied to the definition 8339of a type of a function that returns a pointer and takes at least one 8340argument of an integer type. It indicates that the returned pointer 8341points to an object whose size is given by the function argument at 8342@var{position-1}, or by the product of the arguments at @var{position-1} 8343and @var{position-2}. Meaningful sizes are positive values less than 8344@code{PTRDIFF_MAX}. Other sizes are disagnosed when detected. GCC uses 8345this information to improve the results of @code{__builtin_object_size}. 8346 8347For instance, the following declarations 8348 8349@smallexample 8350typedef __attribute__ ((alloc_size (1, 2))) void* 8351 calloc_type (size_t, size_t); 8352typedef __attribute__ ((alloc_size (1))) void* 8353 malloc_type (size_t); 8354@end smallexample 8355 8356@noindent 8357specify that @code{calloc_type} is a type of a function that, like 8358the standard C function @code{calloc}, returns an object whose size 8359is given by the product of arguments 1 and 2, and that 8360@code{malloc_type}, like the standard C function @code{malloc}, 8361returns an object whose size is given by argument 1 to the function. 8362 8363@item copy 8364@itemx copy (@var{expression}) 8365@cindex @code{copy} type attribute 8366The @code{copy} attribute applies the set of attributes with which 8367the type of the @var{expression} has been declared to the declaration 8368of the type to which the attribute is applied. The attribute is 8369designed for libraries that define aliases that are expected to 8370specify the same set of attributes as the aliased symbols. 8371The @code{copy} attribute can be used with types, variables, or 8372functions. However, the kind of symbol to which the attribute is 8373applied (either varible or function) must match the kind of symbol 8374to which the argument refers. 8375The @code{copy} attribute copies only syntactic and semantic attributes 8376but not attributes that affect a symbol's linkage or visibility such as 8377@code{alias}, @code{visibility}, or @code{weak}. The @code{deprecated} 8378attribute is also not copied. @xref{Common Function Attributes}. 8379@xref{Common Variable Attributes}. 8380 8381For example, suppose @code{struct A} below is defined in some third 8382party library header to have the alignment requirement @code{N} and 8383to force a warning whenever a variable of the type is not so aligned 8384due to attribute @code{packed}. Specifying the @code{copy} attribute 8385on the definition on the unrelated @code{struct B} has the effect of 8386copying all relevant attributes from the type referenced by the pointer 8387expression to @code{struct B}. 8388 8389@smallexample 8390struct __attribute__ ((aligned (N), warn_if_not_aligned (N))) 8391A @{ /* @r{@dots{}} */ @}; 8392struct __attribute__ ((copy ( (struct A *)0)) B @{ /* @r{@dots{}} */ @}; 8393@end smallexample 8394 8395@item deprecated 8396@itemx deprecated (@var{msg}) 8397@cindex @code{deprecated} type attribute 8398The @code{deprecated} attribute results in a warning if the type 8399is used anywhere in the source file. This is useful when identifying 8400types that are expected to be removed in a future version of a program. 8401If possible, the warning also includes the location of the declaration 8402of the deprecated type, to enable users to easily find further 8403information about why the type is deprecated, or what they should do 8404instead. Note that the warnings only occur for uses and then only 8405if the type is being applied to an identifier that itself is not being 8406declared as deprecated. 8407 8408@smallexample 8409typedef int T1 __attribute__ ((deprecated)); 8410T1 x; 8411typedef T1 T2; 8412T2 y; 8413typedef T1 T3 __attribute__ ((deprecated)); 8414T3 z __attribute__ ((deprecated)); 8415@end smallexample 8416 8417@noindent 8418results in a warning on line 2 and 3 but not lines 4, 5, or 6. No 8419warning is issued for line 4 because T2 is not explicitly 8420deprecated. Line 5 has no warning because T3 is explicitly 8421deprecated. Similarly for line 6. The optional @var{msg} 8422argument, which must be a string, is printed in the warning if 8423present. Control characters in the string will be replaced with 8424escape sequences, and if the @option{-fmessage-length} option is set 8425to 0 (its default value) then any newline characters will be ignored. 8426 8427The @code{deprecated} attribute can also be used for functions and 8428variables (@pxref{Function Attributes}, @pxref{Variable Attributes}.) 8429 8430The message attached to the attribute is affected by the setting of 8431the @option{-fmessage-length} option. 8432 8433@item designated_init 8434@cindex @code{designated_init} type attribute 8435This attribute may only be applied to structure types. It indicates 8436that any initialization of an object of this type must use designated 8437initializers rather than positional initializers. The intent of this 8438attribute is to allow the programmer to indicate that a structure's 8439layout may change, and that therefore relying on positional 8440initialization will result in future breakage. 8441 8442GCC emits warnings based on this attribute by default; use 8443@option{-Wno-designated-init} to suppress them. 8444 8445@item may_alias 8446@cindex @code{may_alias} type attribute 8447Accesses through pointers to types with this attribute are not subject 8448to type-based alias analysis, but are instead assumed to be able to alias 8449any other type of objects. 8450In the context of section 6.5 paragraph 7 of the C99 standard, 8451an lvalue expression 8452dereferencing such a pointer is treated like having a character type. 8453See @option{-fstrict-aliasing} for more information on aliasing issues. 8454This extension exists to support some vector APIs, in which pointers to 8455one vector type are permitted to alias pointers to a different vector type. 8456 8457Note that an object of a type with this attribute does not have any 8458special semantics. 8459 8460Example of use: 8461 8462@smallexample 8463typedef short __attribute__ ((__may_alias__)) short_a; 8464 8465int 8466main (void) 8467@{ 8468 int a = 0x12345678; 8469 short_a *b = (short_a *) &a; 8470 8471 b[1] = 0; 8472 8473 if (a == 0x12345678) 8474 abort(); 8475 8476 exit(0); 8477@} 8478@end smallexample 8479 8480@noindent 8481If you replaced @code{short_a} with @code{short} in the variable 8482declaration, the above program would abort when compiled with 8483@option{-fstrict-aliasing}, which is on by default at @option{-O2} or 8484above. 8485 8486@item mode (@var{mode}) 8487@cindex @code{mode} type attribute 8488This attribute specifies the data type for the declaration---whichever 8489type corresponds to the mode @var{mode}. This in effect lets you 8490request an integer or floating-point type according to its width. 8491 8492@xref{Machine Modes,,, gccint, GNU Compiler Collection (GCC) Internals}, 8493for a list of the possible keywords for @var{mode}. 8494You may also specify a mode of @code{byte} or @code{__byte__} to 8495indicate the mode corresponding to a one-byte integer, @code{word} or 8496@code{__word__} for the mode of a one-word integer, and @code{pointer} 8497or @code{__pointer__} for the mode used to represent pointers. 8498 8499@item packed 8500@cindex @code{packed} type attribute 8501This attribute, attached to a @code{struct}, @code{union}, or C++ @code{class} 8502type definition, specifies that each of its members (other than zero-width 8503bit-fields) is placed to minimize the memory required. This is equivalent 8504to specifying the @code{packed} attribute on each of the members. 8505 8506@opindex fshort-enums 8507When attached to an @code{enum} definition, the @code{packed} attribute 8508indicates that the smallest integral type should be used. 8509Specifying the @option{-fshort-enums} flag on the command line 8510is equivalent to specifying the @code{packed} 8511attribute on all @code{enum} definitions. 8512 8513In the following example @code{struct my_packed_struct}'s members are 8514packed closely together, but the internal layout of its @code{s} member 8515is not packed---to do that, @code{struct my_unpacked_struct} needs to 8516be packed too. 8517 8518@smallexample 8519struct my_unpacked_struct 8520 @{ 8521 char c; 8522 int i; 8523 @}; 8524 8525struct __attribute__ ((__packed__)) my_packed_struct 8526 @{ 8527 char c; 8528 int i; 8529 struct my_unpacked_struct s; 8530 @}; 8531@end smallexample 8532 8533You may only specify the @code{packed} attribute on the definition 8534of an @code{enum}, @code{struct}, @code{union}, or @code{class}, 8535not on a @code{typedef} that does not also define the enumerated type, 8536structure, union, or class. 8537 8538@item scalar_storage_order ("@var{endianness}") 8539@cindex @code{scalar_storage_order} type attribute 8540When attached to a @code{union} or a @code{struct}, this attribute sets 8541the storage order, aka endianness, of the scalar fields of the type, as 8542well as the array fields whose component is scalar. The supported 8543endiannesses are @code{big-endian} and @code{little-endian}. The attribute 8544has no effects on fields which are themselves a @code{union}, a @code{struct} 8545or an array whose component is a @code{union} or a @code{struct}, and it is 8546possible for these fields to have a different scalar storage order than the 8547enclosing type. 8548 8549This attribute is supported only for targets that use a uniform default 8550scalar storage order (fortunately, most of them), i.e.@: targets that store 8551the scalars either all in big-endian or all in little-endian. 8552 8553Additional restrictions are enforced for types with the reverse scalar 8554storage order with regard to the scalar storage order of the target: 8555 8556@itemize 8557@item Taking the address of a scalar field of a @code{union} or a 8558@code{struct} with reverse scalar storage order is not permitted and yields 8559an error. 8560@item Taking the address of an array field, whose component is scalar, of 8561a @code{union} or a @code{struct} with reverse scalar storage order is 8562permitted but yields a warning, unless @option{-Wno-scalar-storage-order} 8563is specified. 8564@item Taking the address of a @code{union} or a @code{struct} with reverse 8565scalar storage order is permitted. 8566@end itemize 8567 8568These restrictions exist because the storage order attribute is lost when 8569the address of a scalar or the address of an array with scalar component is 8570taken, so storing indirectly through this address generally does not work. 8571The second case is nevertheless allowed to be able to perform a block copy 8572from or to the array. 8573 8574Moreover, the use of type punning or aliasing to toggle the storage order 8575is not supported; that is to say, a given scalar object cannot be accessed 8576through distinct types that assign a different storage order to it. 8577 8578@item transparent_union 8579@cindex @code{transparent_union} type attribute 8580 8581This attribute, attached to a @code{union} type definition, indicates 8582that any function parameter having that union type causes calls to that 8583function to be treated in a special way. 8584 8585First, the argument corresponding to a transparent union type can be of 8586any type in the union; no cast is required. Also, if the union contains 8587a pointer type, the corresponding argument can be a null pointer 8588constant or a void pointer expression; and if the union contains a void 8589pointer type, the corresponding argument can be any pointer expression. 8590If the union member type is a pointer, qualifiers like @code{const} on 8591the referenced type must be respected, just as with normal pointer 8592conversions. 8593 8594Second, the argument is passed to the function using the calling 8595conventions of the first member of the transparent union, not the calling 8596conventions of the union itself. All members of the union must have the 8597same machine representation; this is necessary for this argument passing 8598to work properly. 8599 8600Transparent unions are designed for library functions that have multiple 8601interfaces for compatibility reasons. For example, suppose the 8602@code{wait} function must accept either a value of type @code{int *} to 8603comply with POSIX, or a value of type @code{union wait *} to comply with 8604the 4.1BSD interface. If @code{wait}'s parameter were @code{void *}, 8605@code{wait} would accept both kinds of arguments, but it would also 8606accept any other pointer type and this would make argument type checking 8607less useful. Instead, @code{<sys/wait.h>} might define the interface 8608as follows: 8609 8610@smallexample 8611typedef union __attribute__ ((__transparent_union__)) 8612 @{ 8613 int *__ip; 8614 union wait *__up; 8615 @} wait_status_ptr_t; 8616 8617pid_t wait (wait_status_ptr_t); 8618@end smallexample 8619 8620@noindent 8621This interface allows either @code{int *} or @code{union wait *} 8622arguments to be passed, using the @code{int *} calling convention. 8623The program can call @code{wait} with arguments of either type: 8624 8625@smallexample 8626int w1 () @{ int w; return wait (&w); @} 8627int w2 () @{ union wait w; return wait (&w); @} 8628@end smallexample 8629 8630@noindent 8631With this interface, @code{wait}'s implementation might look like this: 8632 8633@smallexample 8634pid_t wait (wait_status_ptr_t p) 8635@{ 8636 return waitpid (-1, p.__ip, 0); 8637@} 8638@end smallexample 8639 8640@item unused 8641@cindex @code{unused} type attribute 8642When attached to a type (including a @code{union} or a @code{struct}), 8643this attribute means that variables of that type are meant to appear 8644possibly unused. GCC does not produce a warning for any variables of 8645that type, even if the variable appears to do nothing. This is often 8646the case with lock or thread classes, which are usually defined and then 8647not referenced, but contain constructors and destructors that have 8648nontrivial bookkeeping functions. 8649 8650@item vector_size (@var{bytes}) 8651@cindex @code{vector_size} type attribute 8652This attribute specifies the vector size for the type, measured in bytes. 8653The type to which it applies is known as the @dfn{base type}. The @var{bytes} 8654argument must be a positive power-of-two multiple of the base type size. For 8655example, the following declarations: 8656 8657@smallexample 8658typedef __attribute__ ((vector_size (32))) int int_vec32_t ; 8659typedef __attribute__ ((vector_size (32))) int* int_vec32_ptr_t; 8660typedef __attribute__ ((vector_size (32))) int int_vec32_arr3_t[3]; 8661@end smallexample 8662 8663@noindent 8664define @code{int_vec32_t} to be a 32-byte vector type composed of @code{int} 8665sized units. With @code{int} having a size of 4 bytes, the type defines 8666a vector of eight units, four bytes each. The mode of variables of type 8667@code{int_vec32_t} is @code{V8SI}. @code{int_vec32_ptr_t} is then defined 8668to be a pointer to such a vector type, and @code{int_vec32_arr3_t} to be 8669an array of three such vectors. @xref{Vector Extensions}, for details of 8670manipulating objects of vector types. 8671 8672This attribute is only applicable to integral and floating scalar types. 8673In function declarations the attribute applies to the function return 8674type. 8675 8676For example, the following: 8677@smallexample 8678__attribute__ ((vector_size (16))) float get_flt_vec16 (void); 8679@end smallexample 8680declares @code{get_flt_vec16} to be a function returning a 16-byte vector 8681with the base type @code{float}. 8682 8683@item visibility 8684@cindex @code{visibility} type attribute 8685In C++, attribute visibility (@pxref{Function Attributes}) can also be 8686applied to class, struct, union and enum types. Unlike other type 8687attributes, the attribute must appear between the initial keyword and 8688the name of the type; it cannot appear after the body of the type. 8689 8690Note that the type visibility is applied to vague linkage entities 8691associated with the class (vtable, typeinfo node, etc.). In 8692particular, if a class is thrown as an exception in one shared object 8693and caught in another, the class must have default visibility. 8694Otherwise the two shared objects are unable to use the same 8695typeinfo node and exception handling will break. 8696 8697@item objc_root_class @r{(Objective-C and Objective-C++ only)} 8698@cindex @code{objc_root_class} type attribute 8699This attribute marks a class as being a root class, and thus allows 8700the compiler to elide any warnings about a missing superclass and to 8701make additional checks for mandatory methods as needed. 8702 8703@end table 8704 8705To specify multiple attributes, separate them by commas within the 8706double parentheses: for example, @samp{__attribute__ ((aligned (16), 8707packed))}. 8708 8709@node ARC Type Attributes 8710@subsection ARC Type Attributes 8711 8712@cindex @code{uncached} type attribute, ARC 8713Declaring objects with @code{uncached} allows you to exclude 8714data-cache participation in load and store operations on those objects 8715without involving the additional semantic implications of 8716@code{volatile}. The @code{.di} instruction suffix is used for all 8717loads and stores of data declared @code{uncached}. 8718 8719@node ARM Type Attributes 8720@subsection ARM Type Attributes 8721 8722@cindex @code{notshared} type attribute, ARM 8723On those ARM targets that support @code{dllimport} (such as Symbian 8724OS), you can use the @code{notshared} attribute to indicate that the 8725virtual table and other similar data for a class should not be 8726exported from a DLL@. For example: 8727 8728@smallexample 8729class __declspec(notshared) C @{ 8730public: 8731 __declspec(dllimport) C(); 8732 virtual void f(); 8733@} 8734 8735__declspec(dllexport) 8736C::C() @{@} 8737@end smallexample 8738 8739@noindent 8740In this code, @code{C::C} is exported from the current DLL, but the 8741virtual table for @code{C} is not exported. (You can use 8742@code{__attribute__} instead of @code{__declspec} if you prefer, but 8743most Symbian OS code uses @code{__declspec}.) 8744 8745@node MeP Type Attributes 8746@subsection MeP Type Attributes 8747 8748@cindex @code{based} type attribute, MeP 8749@cindex @code{tiny} type attribute, MeP 8750@cindex @code{near} type attribute, MeP 8751@cindex @code{far} type attribute, MeP 8752Many of the MeP variable attributes may be applied to types as well. 8753Specifically, the @code{based}, @code{tiny}, @code{near}, and 8754@code{far} attributes may be applied to either. The @code{io} and 8755@code{cb} attributes may not be applied to types. 8756 8757@node PowerPC Type Attributes 8758@subsection PowerPC Type Attributes 8759 8760Three attributes currently are defined for PowerPC configurations: 8761@code{altivec}, @code{ms_struct} and @code{gcc_struct}. 8762 8763@cindex @code{ms_struct} type attribute, PowerPC 8764@cindex @code{gcc_struct} type attribute, PowerPC 8765For full documentation of the @code{ms_struct} and @code{gcc_struct} 8766attributes please see the documentation in @ref{x86 Type Attributes}. 8767 8768@cindex @code{altivec} type attribute, PowerPC 8769The @code{altivec} attribute allows one to declare AltiVec vector data 8770types supported by the AltiVec Programming Interface Manual. The 8771attribute requires an argument to specify one of three vector types: 8772@code{vector__}, @code{pixel__} (always followed by unsigned short), 8773and @code{bool__} (always followed by unsigned). 8774 8775@smallexample 8776__attribute__((altivec(vector__))) 8777__attribute__((altivec(pixel__))) unsigned short 8778__attribute__((altivec(bool__))) unsigned 8779@end smallexample 8780 8781These attributes mainly are intended to support the @code{__vector}, 8782@code{__pixel}, and @code{__bool} AltiVec keywords. 8783 8784@node x86 Type Attributes 8785@subsection x86 Type Attributes 8786 8787Two attributes are currently defined for x86 configurations: 8788@code{ms_struct} and @code{gcc_struct}. 8789 8790@table @code 8791 8792@item ms_struct 8793@itemx gcc_struct 8794@cindex @code{ms_struct} type attribute, x86 8795@cindex @code{gcc_struct} type attribute, x86 8796 8797If @code{packed} is used on a structure, or if bit-fields are used 8798it may be that the Microsoft ABI packs them differently 8799than GCC normally packs them. Particularly when moving packed 8800data between functions compiled with GCC and the native Microsoft compiler 8801(either via function call or as data in a file), it may be necessary to access 8802either format. 8803 8804The @code{ms_struct} and @code{gcc_struct} attributes correspond 8805to the @option{-mms-bitfields} and @option{-mno-ms-bitfields} 8806command-line options, respectively; 8807see @ref{x86 Options}, for details of how structure layout is affected. 8808@xref{x86 Variable Attributes}, for information about the corresponding 8809attributes on variables. 8810 8811@end table 8812 8813@node Label Attributes 8814@section Label Attributes 8815@cindex Label Attributes 8816 8817GCC allows attributes to be set on C labels. @xref{Attribute Syntax}, for 8818details of the exact syntax for using attributes. Other attributes are 8819available for functions (@pxref{Function Attributes}), variables 8820(@pxref{Variable Attributes}), enumerators (@pxref{Enumerator Attributes}), 8821statements (@pxref{Statement Attributes}), and for types 8822(@pxref{Type Attributes}). A label attribute followed 8823by a declaration appertains to the label and not the declaration. 8824 8825This example uses the @code{cold} label attribute to indicate the 8826@code{ErrorHandling} branch is unlikely to be taken and that the 8827@code{ErrorHandling} label is unused: 8828 8829@smallexample 8830 8831 asm goto ("some asm" : : : : NoError); 8832 8833/* This branch (the fall-through from the asm) is less commonly used */ 8834ErrorHandling: 8835 __attribute__((cold, unused)); /* Semi-colon is required here */ 8836 printf("error\n"); 8837 return 0; 8838 8839NoError: 8840 printf("no error\n"); 8841 return 1; 8842@end smallexample 8843 8844@table @code 8845@item unused 8846@cindex @code{unused} label attribute 8847This feature is intended for program-generated code that may contain 8848unused labels, but which is compiled with @option{-Wall}. It is 8849not normally appropriate to use in it human-written code, though it 8850could be useful in cases where the code that jumps to the label is 8851contained within an @code{#ifdef} conditional. 8852 8853@item hot 8854@cindex @code{hot} label attribute 8855The @code{hot} attribute on a label is used to inform the compiler that 8856the path following the label is more likely than paths that are not so 8857annotated. This attribute is used in cases where @code{__builtin_expect} 8858cannot be used, for instance with computed goto or @code{asm goto}. 8859 8860@item cold 8861@cindex @code{cold} label attribute 8862The @code{cold} attribute on labels is used to inform the compiler that 8863the path following the label is unlikely to be executed. This attribute 8864is used in cases where @code{__builtin_expect} cannot be used, for instance 8865with computed goto or @code{asm goto}. 8866 8867@end table 8868 8869@node Enumerator Attributes 8870@section Enumerator Attributes 8871@cindex Enumerator Attributes 8872 8873GCC allows attributes to be set on enumerators. @xref{Attribute Syntax}, for 8874details of the exact syntax for using attributes. Other attributes are 8875available for functions (@pxref{Function Attributes}), variables 8876(@pxref{Variable Attributes}), labels (@pxref{Label Attributes}), statements 8877(@pxref{Statement Attributes}), and for types (@pxref{Type Attributes}). 8878 8879This example uses the @code{deprecated} enumerator attribute to indicate the 8880@code{oldval} enumerator is deprecated: 8881 8882@smallexample 8883enum E @{ 8884 oldval __attribute__((deprecated)), 8885 newval 8886@}; 8887 8888int 8889fn (void) 8890@{ 8891 return oldval; 8892@} 8893@end smallexample 8894 8895@table @code 8896@item deprecated 8897@cindex @code{deprecated} enumerator attribute 8898The @code{deprecated} attribute results in a warning if the enumerator 8899is used anywhere in the source file. This is useful when identifying 8900enumerators that are expected to be removed in a future version of a 8901program. The warning also includes the location of the declaration 8902of the deprecated enumerator, to enable users to easily find further 8903information about why the enumerator is deprecated, or what they should 8904do instead. Note that the warnings only occurs for uses. 8905 8906@end table 8907 8908@node Statement Attributes 8909@section Statement Attributes 8910@cindex Statement Attributes 8911 8912GCC allows attributes to be set on null statements. @xref{Attribute Syntax}, 8913for details of the exact syntax for using attributes. Other attributes are 8914available for functions (@pxref{Function Attributes}), variables 8915(@pxref{Variable Attributes}), labels (@pxref{Label Attributes}), enumerators 8916(@pxref{Enumerator Attributes}), and for types (@pxref{Type Attributes}). 8917 8918This example uses the @code{fallthrough} statement attribute to indicate that 8919the @option{-Wimplicit-fallthrough} warning should not be emitted: 8920 8921@smallexample 8922switch (cond) 8923 @{ 8924 case 1: 8925 bar (1); 8926 __attribute__((fallthrough)); 8927 case 2: 8928 @dots{} 8929 @} 8930@end smallexample 8931 8932@table @code 8933@item fallthrough 8934@cindex @code{fallthrough} statement attribute 8935The @code{fallthrough} attribute with a null statement serves as a 8936fallthrough statement. It hints to the compiler that a statement 8937that falls through to another case label, or user-defined label 8938in a switch statement is intentional and thus the 8939@option{-Wimplicit-fallthrough} warning must not trigger. The 8940fallthrough attribute may appear at most once in each attribute 8941list, and may not be mixed with other attributes. It can only 8942be used in a switch statement (the compiler will issue an error 8943otherwise), after a preceding statement and before a logically 8944succeeding case label, or user-defined label. 8945 8946@end table 8947 8948@node Attribute Syntax 8949@section Attribute Syntax 8950@cindex attribute syntax 8951 8952This section describes the syntax with which @code{__attribute__} may be 8953used, and the constructs to which attribute specifiers bind, for the C 8954language. Some details may vary for C++ and Objective-C@. Because of 8955infelicities in the grammar for attributes, some forms described here 8956may not be successfully parsed in all cases. 8957 8958There are some problems with the semantics of attributes in C++. For 8959example, there are no manglings for attributes, although they may affect 8960code generation, so problems may arise when attributed types are used in 8961conjunction with templates or overloading. Similarly, @code{typeid} 8962does not distinguish between types with different attributes. Support 8963for attributes in C++ may be restricted in future to attributes on 8964declarations only, but not on nested declarators. 8965 8966@xref{Function Attributes}, for details of the semantics of attributes 8967applying to functions. @xref{Variable Attributes}, for details of the 8968semantics of attributes applying to variables. @xref{Type Attributes}, 8969for details of the semantics of attributes applying to structure, union 8970and enumerated types. 8971@xref{Label Attributes}, for details of the semantics of attributes 8972applying to labels. 8973@xref{Enumerator Attributes}, for details of the semantics of attributes 8974applying to enumerators. 8975@xref{Statement Attributes}, for details of the semantics of attributes 8976applying to statements. 8977 8978An @dfn{attribute specifier} is of the form 8979@code{__attribute__ ((@var{attribute-list}))}. An @dfn{attribute list} 8980is a possibly empty comma-separated sequence of @dfn{attributes}, where 8981each attribute is one of the following: 8982 8983@itemize @bullet 8984@item 8985Empty. Empty attributes are ignored. 8986 8987@item 8988An attribute name 8989(which may be an identifier such as @code{unused}, or a reserved 8990word such as @code{const}). 8991 8992@item 8993An attribute name followed by a parenthesized list of 8994parameters for the attribute. 8995These parameters take one of the following forms: 8996 8997@itemize @bullet 8998@item 8999An identifier. For example, @code{mode} attributes use this form. 9000 9001@item 9002An identifier followed by a comma and a non-empty comma-separated list 9003of expressions. For example, @code{format} attributes use this form. 9004 9005@item 9006A possibly empty comma-separated list of expressions. For example, 9007@code{format_arg} attributes use this form with the list being a single 9008integer constant expression, and @code{alias} attributes use this form 9009with the list being a single string constant. 9010@end itemize 9011@end itemize 9012 9013An @dfn{attribute specifier list} is a sequence of one or more attribute 9014specifiers, not separated by any other tokens. 9015 9016You may optionally specify attribute names with @samp{__} 9017preceding and following the name. 9018This allows you to use them in header files without 9019being concerned about a possible macro of the same name. For example, 9020you may use the attribute name @code{__noreturn__} instead of @code{noreturn}. 9021 9022 9023@subsubheading Label Attributes 9024 9025In GNU C, an attribute specifier list may appear after the colon following a 9026label, other than a @code{case} or @code{default} label. GNU C++ only permits 9027attributes on labels if the attribute specifier is immediately 9028followed by a semicolon (i.e., the label applies to an empty 9029statement). If the semicolon is missing, C++ label attributes are 9030ambiguous, as it is permissible for a declaration, which could begin 9031with an attribute list, to be labelled in C++. Declarations cannot be 9032labelled in C90 or C99, so the ambiguity does not arise there. 9033 9034@subsubheading Enumerator Attributes 9035 9036In GNU C, an attribute specifier list may appear as part of an enumerator. 9037The attribute goes after the enumeration constant, before @code{=}, if 9038present. The optional attribute in the enumerator appertains to the 9039enumeration constant. It is not possible to place the attribute after 9040the constant expression, if present. 9041 9042@subsubheading Statement Attributes 9043In GNU C, an attribute specifier list may appear as part of a null 9044statement. The attribute goes before the semicolon. 9045 9046@subsubheading Type Attributes 9047 9048An attribute specifier list may appear as part of a @code{struct}, 9049@code{union} or @code{enum} specifier. It may go either immediately 9050after the @code{struct}, @code{union} or @code{enum} keyword, or after 9051the closing brace. The former syntax is preferred. 9052Where attribute specifiers follow the closing brace, they are considered 9053to relate to the structure, union or enumerated type defined, not to any 9054enclosing declaration the type specifier appears in, and the type 9055defined is not complete until after the attribute specifiers. 9056@c Otherwise, there would be the following problems: a shift/reduce 9057@c conflict between attributes binding the struct/union/enum and 9058@c binding to the list of specifiers/qualifiers; and "aligned" 9059@c attributes could use sizeof for the structure, but the size could be 9060@c changed later by "packed" attributes. 9061 9062 9063@subsubheading All other attributes 9064 9065Otherwise, an attribute specifier appears as part of a declaration, 9066counting declarations of unnamed parameters and type names, and relates 9067to that declaration (which may be nested in another declaration, for 9068example in the case of a parameter declaration), or to a particular declarator 9069within a declaration. Where an 9070attribute specifier is applied to a parameter declared as a function or 9071an array, it should apply to the function or array rather than the 9072pointer to which the parameter is implicitly converted, but this is not 9073yet correctly implemented. 9074 9075Any list of specifiers and qualifiers at the start of a declaration may 9076contain attribute specifiers, whether or not such a list may in that 9077context contain storage class specifiers. (Some attributes, however, 9078are essentially in the nature of storage class specifiers, and only make 9079sense where storage class specifiers may be used; for example, 9080@code{section}.) There is one necessary limitation to this syntax: the 9081first old-style parameter declaration in a function definition cannot 9082begin with an attribute specifier, because such an attribute applies to 9083the function instead by syntax described below (which, however, is not 9084yet implemented in this case). In some other cases, attribute 9085specifiers are permitted by this grammar but not yet supported by the 9086compiler. All attribute specifiers in this place relate to the 9087declaration as a whole. In the obsolescent usage where a type of 9088@code{int} is implied by the absence of type specifiers, such a list of 9089specifiers and qualifiers may be an attribute specifier list with no 9090other specifiers or qualifiers. 9091 9092At present, the first parameter in a function prototype must have some 9093type specifier that is not an attribute specifier; this resolves an 9094ambiguity in the interpretation of @code{void f(int 9095(__attribute__((foo)) x))}, but is subject to change. At present, if 9096the parentheses of a function declarator contain only attributes then 9097those attributes are ignored, rather than yielding an error or warning 9098or implying a single parameter of type int, but this is subject to 9099change. 9100 9101An attribute specifier list may appear immediately before a declarator 9102(other than the first) in a comma-separated list of declarators in a 9103declaration of more than one identifier using a single list of 9104specifiers and qualifiers. Such attribute specifiers apply 9105only to the identifier before whose declarator they appear. For 9106example, in 9107 9108@smallexample 9109__attribute__((noreturn)) void d0 (void), 9110 __attribute__((format(printf, 1, 2))) d1 (const char *, ...), 9111 d2 (void); 9112@end smallexample 9113 9114@noindent 9115the @code{noreturn} attribute applies to all the functions 9116declared; the @code{format} attribute only applies to @code{d1}. 9117 9118An attribute specifier list may appear immediately before the comma, 9119@code{=} or semicolon terminating the declaration of an identifier other 9120than a function definition. Such attribute specifiers apply 9121to the declared object or function. Where an 9122assembler name for an object or function is specified (@pxref{Asm 9123Labels}), the attribute must follow the @code{asm} 9124specification. 9125 9126An attribute specifier list may, in future, be permitted to appear after 9127the declarator in a function definition (before any old-style parameter 9128declarations or the function body). 9129 9130Attribute specifiers may be mixed with type qualifiers appearing inside 9131the @code{[]} of a parameter array declarator, in the C99 construct by 9132which such qualifiers are applied to the pointer to which the array is 9133implicitly converted. Such attribute specifiers apply to the pointer, 9134not to the array, but at present this is not implemented and they are 9135ignored. 9136 9137An attribute specifier list may appear at the start of a nested 9138declarator. At present, there are some limitations in this usage: the 9139attributes correctly apply to the declarator, but for most individual 9140attributes the semantics this implies are not implemented. 9141When attribute specifiers follow the @code{*} of a pointer 9142declarator, they may be mixed with any type qualifiers present. 9143The following describes the formal semantics of this syntax. It makes the 9144most sense if you are familiar with the formal specification of 9145declarators in the ISO C standard. 9146 9147Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration @code{T 9148D1}, where @code{T} contains declaration specifiers that specify a type 9149@var{Type} (such as @code{int}) and @code{D1} is a declarator that 9150contains an identifier @var{ident}. The type specified for @var{ident} 9151for derived declarators whose type does not include an attribute 9152specifier is as in the ISO C standard. 9153 9154If @code{D1} has the form @code{( @var{attribute-specifier-list} D )}, 9155and the declaration @code{T D} specifies the type 9156``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 9157@code{T D1} specifies the type ``@var{derived-declarator-type-list} 9158@var{attribute-specifier-list} @var{Type}'' for @var{ident}. 9159 9160If @code{D1} has the form @code{* 9161@var{type-qualifier-and-attribute-specifier-list} D}, and the 9162declaration @code{T D} specifies the type 9163``@var{derived-declarator-type-list} @var{Type}'' for @var{ident}, then 9164@code{T D1} specifies the type ``@var{derived-declarator-type-list} 9165@var{type-qualifier-and-attribute-specifier-list} pointer to @var{Type}'' for 9166@var{ident}. 9167 9168For example, 9169 9170@smallexample 9171void (__attribute__((noreturn)) ****f) (void); 9172@end smallexample 9173 9174@noindent 9175specifies the type ``pointer to pointer to pointer to pointer to 9176non-returning function returning @code{void}''. As another example, 9177 9178@smallexample 9179char *__attribute__((aligned(8))) *f; 9180@end smallexample 9181 9182@noindent 9183specifies the type ``pointer to 8-byte-aligned pointer to @code{char}''. 9184Note again that this does not work with most attributes; for example, 9185the usage of @samp{aligned} and @samp{noreturn} attributes given above 9186is not yet supported. 9187 9188For compatibility with existing code written for compiler versions that 9189did not implement attributes on nested declarators, some laxity is 9190allowed in the placing of attributes. If an attribute that only applies 9191to types is applied to a declaration, it is treated as applying to 9192the type of that declaration. If an attribute that only applies to 9193declarations is applied to the type of a declaration, it is treated 9194as applying to that declaration; and, for compatibility with code 9195placing the attributes immediately before the identifier declared, such 9196an attribute applied to a function return type is treated as 9197applying to the function type, and such an attribute applied to an array 9198element type is treated as applying to the array type. If an 9199attribute that only applies to function types is applied to a 9200pointer-to-function type, it is treated as applying to the pointer 9201target type; if such an attribute is applied to a function return type 9202that is not a pointer-to-function type, it is treated as applying 9203to the function type. 9204 9205@node Function Prototypes 9206@section Prototypes and Old-Style Function Definitions 9207@cindex function prototype declarations 9208@cindex old-style function definitions 9209@cindex promotion of formal parameters 9210 9211GNU C extends ISO C to allow a function prototype to override a later 9212old-style non-prototype definition. Consider the following example: 9213 9214@smallexample 9215/* @r{Use prototypes unless the compiler is old-fashioned.} */ 9216#ifdef __STDC__ 9217#define P(x) x 9218#else 9219#define P(x) () 9220#endif 9221 9222/* @r{Prototype function declaration.} */ 9223int isroot P((uid_t)); 9224 9225/* @r{Old-style function definition.} */ 9226int 9227isroot (x) /* @r{??? lossage here ???} */ 9228 uid_t x; 9229@{ 9230 return x == 0; 9231@} 9232@end smallexample 9233 9234Suppose the type @code{uid_t} happens to be @code{short}. ISO C does 9235not allow this example, because subword arguments in old-style 9236non-prototype definitions are promoted. Therefore in this example the 9237function definition's argument is really an @code{int}, which does not 9238match the prototype argument type of @code{short}. 9239 9240This restriction of ISO C makes it hard to write code that is portable 9241to traditional C compilers, because the programmer does not know 9242whether the @code{uid_t} type is @code{short}, @code{int}, or 9243@code{long}. Therefore, in cases like these GNU C allows a prototype 9244to override a later old-style definition. More precisely, in GNU C, a 9245function prototype argument type overrides the argument type specified 9246by a later old-style definition if the former type is the same as the 9247latter type before promotion. Thus in GNU C the above example is 9248equivalent to the following: 9249 9250@smallexample 9251int isroot (uid_t); 9252 9253int 9254isroot (uid_t x) 9255@{ 9256 return x == 0; 9257@} 9258@end smallexample 9259 9260@noindent 9261GNU C++ does not support old-style function definitions, so this 9262extension is irrelevant. 9263 9264@node C++ Comments 9265@section C++ Style Comments 9266@cindex @code{//} 9267@cindex C++ comments 9268@cindex comments, C++ style 9269 9270In GNU C, you may use C++ style comments, which start with @samp{//} and 9271continue until the end of the line. Many other C implementations allow 9272such comments, and they are included in the 1999 C standard. However, 9273C++ style comments are not recognized if you specify an @option{-std} 9274option specifying a version of ISO C before C99, or @option{-ansi} 9275(equivalent to @option{-std=c90}). 9276 9277@node Dollar Signs 9278@section Dollar Signs in Identifier Names 9279@cindex $ 9280@cindex dollar signs in identifier names 9281@cindex identifier names, dollar signs in 9282 9283In GNU C, you may normally use dollar signs in identifier names. 9284This is because many traditional C implementations allow such identifiers. 9285However, dollar signs in identifiers are not supported on a few target 9286machines, typically because the target assembler does not allow them. 9287 9288@node Character Escapes 9289@section The Character @key{ESC} in Constants 9290 9291You can use the sequence @samp{\e} in a string or character constant to 9292stand for the ASCII character @key{ESC}. 9293 9294@node Alignment 9295@section Determining the Alignment of Functions, Types or Variables 9296@cindex alignment 9297@cindex type alignment 9298@cindex variable alignment 9299 9300The keyword @code{__alignof__} determines the alignment requirement of 9301a function, object, or a type, or the minimum alignment usually required 9302by a type. Its syntax is just like @code{sizeof} and C11 @code{_Alignof}. 9303 9304For example, if the target machine requires a @code{double} value to be 9305aligned on an 8-byte boundary, then @code{__alignof__ (double)} is 8. 9306This is true on many RISC machines. On more traditional machine 9307designs, @code{__alignof__ (double)} is 4 or even 2. 9308 9309Some machines never actually require alignment; they allow references to any 9310data type even at an odd address. For these machines, @code{__alignof__} 9311reports the smallest alignment that GCC gives the data type, usually as 9312mandated by the target ABI. 9313 9314If the operand of @code{__alignof__} is an lvalue rather than a type, 9315its value is the required alignment for its type, taking into account 9316any minimum alignment specified by attribute @code{aligned} 9317(@pxref{Common Variable Attributes}). For example, after this 9318declaration: 9319 9320@smallexample 9321struct foo @{ int x; char y; @} foo1; 9322@end smallexample 9323 9324@noindent 9325the value of @code{__alignof__ (foo1.y)} is 1, even though its actual 9326alignment is probably 2 or 4, the same as @code{__alignof__ (int)}. 9327It is an error to ask for the alignment of an incomplete type other 9328than @code{void}. 9329 9330If the operand of the @code{__alignof__} expression is a function, 9331the expression evaluates to the alignment of the function which may 9332be specified by attribute @code{aligned} (@pxref{Common Function Attributes}). 9333 9334@node Inline 9335@section An Inline Function is As Fast As a Macro 9336@cindex inline functions 9337@cindex integrating function code 9338@cindex open coding 9339@cindex macros, inline alternative 9340 9341By declaring a function inline, you can direct GCC to make 9342calls to that function faster. One way GCC can achieve this is to 9343integrate that function's code into the code for its callers. This 9344makes execution faster by eliminating the function-call overhead; in 9345addition, if any of the actual argument values are constant, their 9346known values may permit simplifications at compile time so that not 9347all of the inline function's code needs to be included. The effect on 9348code size is less predictable; object code may be larger or smaller 9349with function inlining, depending on the particular case. You can 9350also direct GCC to try to integrate all ``simple enough'' functions 9351into their callers with the option @option{-finline-functions}. 9352 9353GCC implements three different semantics of declaring a function 9354inline. One is available with @option{-std=gnu89} or 9355@option{-fgnu89-inline} or when @code{gnu_inline} attribute is present 9356on all inline declarations, another when 9357@option{-std=c99}, 9358@option{-std=gnu99} or an option for a later C version is used 9359(without @option{-fgnu89-inline}), and the third 9360is used when compiling C++. 9361 9362To declare a function inline, use the @code{inline} keyword in its 9363declaration, like this: 9364 9365@smallexample 9366static inline int 9367inc (int *a) 9368@{ 9369 return (*a)++; 9370@} 9371@end smallexample 9372 9373If you are writing a header file to be included in ISO C90 programs, write 9374@code{__inline__} instead of @code{inline}. @xref{Alternate Keywords}. 9375 9376The three types of inlining behave similarly in two important cases: 9377when the @code{inline} keyword is used on a @code{static} function, 9378like the example above, and when a function is first declared without 9379using the @code{inline} keyword and then is defined with 9380@code{inline}, like this: 9381 9382@smallexample 9383extern int inc (int *a); 9384inline int 9385inc (int *a) 9386@{ 9387 return (*a)++; 9388@} 9389@end smallexample 9390 9391In both of these common cases, the program behaves the same as if you 9392had not used the @code{inline} keyword, except for its speed. 9393 9394@cindex inline functions, omission of 9395@opindex fkeep-inline-functions 9396When a function is both inline and @code{static}, if all calls to the 9397function are integrated into the caller, and the function's address is 9398never used, then the function's own assembler code is never referenced. 9399In this case, GCC does not actually output assembler code for the 9400function, unless you specify the option @option{-fkeep-inline-functions}. 9401If there is a nonintegrated call, then the function is compiled to 9402assembler code as usual. The function must also be compiled as usual if 9403the program refers to its address, because that cannot be inlined. 9404 9405@opindex Winline 9406Note that certain usages in a function definition can make it unsuitable 9407for inline substitution. Among these usages are: variadic functions, 9408use of @code{alloca}, use of computed goto (@pxref{Labels as Values}), 9409use of nonlocal goto, use of nested functions, use of @code{setjmp}, use 9410of @code{__builtin_longjmp} and use of @code{__builtin_return} or 9411@code{__builtin_apply_args}. Using @option{-Winline} warns when a 9412function marked @code{inline} could not be substituted, and gives the 9413reason for the failure. 9414 9415@cindex automatic @code{inline} for C++ member fns 9416@cindex @code{inline} automatic for C++ member fns 9417@cindex member fns, automatically @code{inline} 9418@cindex C++ member fns, automatically @code{inline} 9419@opindex fno-default-inline 9420As required by ISO C++, GCC considers member functions defined within 9421the body of a class to be marked inline even if they are 9422not explicitly declared with the @code{inline} keyword. You can 9423override this with @option{-fno-default-inline}; @pxref{C++ Dialect 9424Options,,Options Controlling C++ Dialect}. 9425 9426GCC does not inline any functions when not optimizing unless you specify 9427the @samp{always_inline} attribute for the function, like this: 9428 9429@smallexample 9430/* @r{Prototype.} */ 9431inline void foo (const char) __attribute__((always_inline)); 9432@end smallexample 9433 9434The remainder of this section is specific to GNU C90 inlining. 9435 9436@cindex non-static inline function 9437When an inline function is not @code{static}, then the compiler must assume 9438that there may be calls from other source files; since a global symbol can 9439be defined only once in any program, the function must not be defined in 9440the other source files, so the calls therein cannot be integrated. 9441Therefore, a non-@code{static} inline function is always compiled on its 9442own in the usual fashion. 9443 9444If you specify both @code{inline} and @code{extern} in the function 9445definition, then the definition is used only for inlining. In no case 9446is the function compiled on its own, not even if you refer to its 9447address explicitly. Such an address becomes an external reference, as 9448if you had only declared the function, and had not defined it. 9449 9450This combination of @code{inline} and @code{extern} has almost the 9451effect of a macro. The way to use it is to put a function definition in 9452a header file with these keywords, and put another copy of the 9453definition (lacking @code{inline} and @code{extern}) in a library file. 9454The definition in the header file causes most calls to the function 9455to be inlined. If any uses of the function remain, they refer to 9456the single copy in the library. 9457 9458@node Volatiles 9459@section When is a Volatile Object Accessed? 9460@cindex accessing volatiles 9461@cindex volatile read 9462@cindex volatile write 9463@cindex volatile access 9464 9465C has the concept of volatile objects. These are normally accessed by 9466pointers and used for accessing hardware or inter-thread 9467communication. The standard encourages compilers to refrain from 9468optimizations concerning accesses to volatile objects, but leaves it 9469implementation defined as to what constitutes a volatile access. The 9470minimum requirement is that at a sequence point all previous accesses 9471to volatile objects have stabilized and no subsequent accesses have 9472occurred. Thus an implementation is free to reorder and combine 9473volatile accesses that occur between sequence points, but cannot do 9474so for accesses across a sequence point. The use of volatile does 9475not allow you to violate the restriction on updating objects multiple 9476times between two sequence points. 9477 9478Accesses to non-volatile objects are not ordered with respect to 9479volatile accesses. You cannot use a volatile object as a memory 9480barrier to order a sequence of writes to non-volatile memory. For 9481instance: 9482 9483@smallexample 9484int *ptr = @var{something}; 9485volatile int vobj; 9486*ptr = @var{something}; 9487vobj = 1; 9488@end smallexample 9489 9490@noindent 9491Unless @var{*ptr} and @var{vobj} can be aliased, it is not guaranteed 9492that the write to @var{*ptr} occurs by the time the update 9493of @var{vobj} happens. If you need this guarantee, you must use 9494a stronger memory barrier such as: 9495 9496@smallexample 9497int *ptr = @var{something}; 9498volatile int vobj; 9499*ptr = @var{something}; 9500asm volatile ("" : : : "memory"); 9501vobj = 1; 9502@end smallexample 9503 9504A scalar volatile object is read when it is accessed in a void context: 9505 9506@smallexample 9507volatile int *src = @var{somevalue}; 9508*src; 9509@end smallexample 9510 9511Such expressions are rvalues, and GCC implements this as a 9512read of the volatile object being pointed to. 9513 9514Assignments are also expressions and have an rvalue. However when 9515assigning to a scalar volatile, the volatile object is not reread, 9516regardless of whether the assignment expression's rvalue is used or 9517not. If the assignment's rvalue is used, the value is that assigned 9518to the volatile object. For instance, there is no read of @var{vobj} 9519in all the following cases: 9520 9521@smallexample 9522int obj; 9523volatile int vobj; 9524vobj = @var{something}; 9525obj = vobj = @var{something}; 9526obj ? vobj = @var{onething} : vobj = @var{anotherthing}; 9527obj = (@var{something}, vobj = @var{anotherthing}); 9528@end smallexample 9529 9530If you need to read the volatile object after an assignment has 9531occurred, you must use a separate expression with an intervening 9532sequence point. 9533 9534As bit-fields are not individually addressable, volatile bit-fields may 9535be implicitly read when written to, or when adjacent bit-fields are 9536accessed. Bit-field operations may be optimized such that adjacent 9537bit-fields are only partially accessed, if they straddle a storage unit 9538boundary. For these reasons it is unwise to use volatile bit-fields to 9539access hardware. 9540 9541@node Using Assembly Language with C 9542@section How to Use Inline Assembly Language in C Code 9543@cindex @code{asm} keyword 9544@cindex assembly language in C 9545@cindex inline assembly language 9546@cindex mixing assembly language and C 9547 9548The @code{asm} keyword allows you to embed assembler instructions 9549within C code. GCC provides two forms of inline @code{asm} 9550statements. A @dfn{basic @code{asm}} statement is one with no 9551operands (@pxref{Basic Asm}), while an @dfn{extended @code{asm}} 9552statement (@pxref{Extended Asm}) includes one or more operands. 9553The extended form is preferred for mixing C and assembly language 9554within a function, but to include assembly language at 9555top level you must use basic @code{asm}. 9556 9557You can also use the @code{asm} keyword to override the assembler name 9558for a C symbol, or to place a C variable in a specific register. 9559 9560@menu 9561* Basic Asm:: Inline assembler without operands. 9562* Extended Asm:: Inline assembler with operands. 9563* Constraints:: Constraints for @code{asm} operands 9564* Asm Labels:: Specifying the assembler name to use for a C symbol. 9565* Explicit Register Variables:: Defining variables residing in specified 9566 registers. 9567* Size of an asm:: How GCC calculates the size of an @code{asm} block. 9568@end menu 9569 9570@node Basic Asm 9571@subsection Basic Asm --- Assembler Instructions Without Operands 9572@cindex basic @code{asm} 9573@cindex assembly language in C, basic 9574 9575A basic @code{asm} statement has the following syntax: 9576 9577@example 9578asm @var{asm-qualifiers} ( @var{AssemblerInstructions} ) 9579@end example 9580 9581The @code{asm} keyword is a GNU extension. 9582When writing code that can be compiled with @option{-ansi} and the 9583various @option{-std} options, use @code{__asm__} instead of 9584@code{asm} (@pxref{Alternate Keywords}). 9585 9586@subsubheading Qualifiers 9587@table @code 9588@item volatile 9589The optional @code{volatile} qualifier has no effect. 9590All basic @code{asm} blocks are implicitly volatile. 9591 9592@item inline 9593If you use the @code{inline} qualifier, then for inlining purposes the size 9594of the @code{asm} statement is taken as the smallest size possible (@pxref{Size 9595of an asm}). 9596@end table 9597 9598@subsubheading Parameters 9599@table @var 9600 9601@item AssemblerInstructions 9602This is a literal string that specifies the assembler code. The string can 9603contain any instructions recognized by the assembler, including directives. 9604GCC does not parse the assembler instructions themselves and 9605does not know what they mean or even whether they are valid assembler input. 9606 9607You may place multiple assembler instructions together in a single @code{asm} 9608string, separated by the characters normally used in assembly code for the 9609system. A combination that works in most places is a newline to break the 9610line, plus a tab character (written as @samp{\n\t}). 9611Some assemblers allow semicolons as a line separator. However, 9612note that some assembler dialects use semicolons to start a comment. 9613@end table 9614 9615@subsubheading Remarks 9616Using extended @code{asm} (@pxref{Extended Asm}) typically produces 9617smaller, safer, and more efficient code, and in most cases it is a 9618better solution than basic @code{asm}. However, there are two 9619situations where only basic @code{asm} can be used: 9620 9621@itemize @bullet 9622@item 9623Extended @code{asm} statements have to be inside a C 9624function, so to write inline assembly language at file scope (``top-level''), 9625outside of C functions, you must use basic @code{asm}. 9626You can use this technique to emit assembler directives, 9627define assembly language macros that can be invoked elsewhere in the file, 9628or write entire functions in assembly language. 9629Basic @code{asm} statements outside of functions may not use any 9630qualifiers. 9631 9632@item 9633Functions declared 9634with the @code{naked} attribute also require basic @code{asm} 9635(@pxref{Function Attributes}). 9636@end itemize 9637 9638Safely accessing C data and calling functions from basic @code{asm} is more 9639complex than it may appear. To access C data, it is better to use extended 9640@code{asm}. 9641 9642Do not expect a sequence of @code{asm} statements to remain perfectly 9643consecutive after compilation. If certain instructions need to remain 9644consecutive in the output, put them in a single multi-instruction @code{asm} 9645statement. Note that GCC's optimizers can move @code{asm} statements 9646relative to other code, including across jumps. 9647 9648@code{asm} statements may not perform jumps into other @code{asm} statements. 9649GCC does not know about these jumps, and therefore cannot take 9650account of them when deciding how to optimize. Jumps from @code{asm} to C 9651labels are only supported in extended @code{asm}. 9652 9653Under certain circumstances, GCC may duplicate (or remove duplicates of) your 9654assembly code when optimizing. This can lead to unexpected duplicate 9655symbol errors during compilation if your assembly code defines symbols or 9656labels. 9657 9658@strong{Warning:} The C standards do not specify semantics for @code{asm}, 9659making it a potential source of incompatibilities between compilers. These 9660incompatibilities may not produce compiler warnings/errors. 9661 9662GCC does not parse basic @code{asm}'s @var{AssemblerInstructions}, which 9663means there is no way to communicate to the compiler what is happening 9664inside them. GCC has no visibility of symbols in the @code{asm} and may 9665discard them as unreferenced. It also does not know about side effects of 9666the assembler code, such as modifications to memory or registers. Unlike 9667some compilers, GCC assumes that no changes to general purpose registers 9668occur. This assumption may change in a future release. 9669 9670To avoid complications from future changes to the semantics and the 9671compatibility issues between compilers, consider replacing basic @code{asm} 9672with extended @code{asm}. See 9673@uref{https://gcc.gnu.org/wiki/ConvertBasicAsmToExtended, How to convert 9674from basic asm to extended asm} for information about how to perform this 9675conversion. 9676 9677The compiler copies the assembler instructions in a basic @code{asm} 9678verbatim to the assembly language output file, without 9679processing dialects or any of the @samp{%} operators that are available with 9680extended @code{asm}. This results in minor differences between basic 9681@code{asm} strings and extended @code{asm} templates. For example, to refer to 9682registers you might use @samp{%eax} in basic @code{asm} and 9683@samp{%%eax} in extended @code{asm}. 9684 9685On targets such as x86 that support multiple assembler dialects, 9686all basic @code{asm} blocks use the assembler dialect specified by the 9687@option{-masm} command-line option (@pxref{x86 Options}). 9688Basic @code{asm} provides no 9689mechanism to provide different assembler strings for different dialects. 9690 9691For basic @code{asm} with non-empty assembler string GCC assumes 9692the assembler block does not change any general purpose registers, 9693but it may read or write any globally accessible variable. 9694 9695Here is an example of basic @code{asm} for i386: 9696 9697@example 9698/* Note that this code will not compile with -masm=intel */ 9699#define DebugBreak() asm("int $3") 9700@end example 9701 9702@node Extended Asm 9703@subsection Extended Asm - Assembler Instructions with C Expression Operands 9704@cindex extended @code{asm} 9705@cindex assembly language in C, extended 9706 9707With extended @code{asm} you can read and write C variables from 9708assembler and perform jumps from assembler code to C labels. 9709Extended @code{asm} syntax uses colons (@samp{:}) to delimit 9710the operand parameters after the assembler template: 9711 9712@example 9713asm @var{asm-qualifiers} ( @var{AssemblerTemplate} 9714 : @var{OutputOperands} 9715 @r{[} : @var{InputOperands} 9716 @r{[} : @var{Clobbers} @r{]} @r{]}) 9717 9718asm @var{asm-qualifiers} ( @var{AssemblerTemplate} 9719 : @var{OutputOperands} 9720 : @var{InputOperands} 9721 : @var{Clobbers} 9722 : @var{GotoLabels}) 9723@end example 9724where in the last form, @var{asm-qualifiers} contains @code{goto} (and in the 9725first form, not). 9726 9727The @code{asm} keyword is a GNU extension. 9728When writing code that can be compiled with @option{-ansi} and the 9729various @option{-std} options, use @code{__asm__} instead of 9730@code{asm} (@pxref{Alternate Keywords}). 9731 9732@subsubheading Qualifiers 9733@table @code 9734 9735@item volatile 9736The typical use of extended @code{asm} statements is to manipulate input 9737values to produce output values. However, your @code{asm} statements may 9738also produce side effects. If so, you may need to use the @code{volatile} 9739qualifier to disable certain optimizations. @xref{Volatile}. 9740 9741@item inline 9742If you use the @code{inline} qualifier, then for inlining purposes the size 9743of the @code{asm} statement is taken as the smallest size possible 9744(@pxref{Size of an asm}). 9745 9746@item goto 9747This qualifier informs the compiler that the @code{asm} statement may 9748perform a jump to one of the labels listed in the @var{GotoLabels}. 9749@xref{GotoLabels}. 9750@end table 9751 9752@subsubheading Parameters 9753@table @var 9754@item AssemblerTemplate 9755This is a literal string that is the template for the assembler code. It is a 9756combination of fixed text and tokens that refer to the input, output, 9757and goto parameters. @xref{AssemblerTemplate}. 9758 9759@item OutputOperands 9760A comma-separated list of the C variables modified by the instructions in the 9761@var{AssemblerTemplate}. An empty list is permitted. @xref{OutputOperands}. 9762 9763@item InputOperands 9764A comma-separated list of C expressions read by the instructions in the 9765@var{AssemblerTemplate}. An empty list is permitted. @xref{InputOperands}. 9766 9767@item Clobbers 9768A comma-separated list of registers or other values changed by the 9769@var{AssemblerTemplate}, beyond those listed as outputs. 9770An empty list is permitted. @xref{Clobbers and Scratch Registers}. 9771 9772@item GotoLabels 9773When you are using the @code{goto} form of @code{asm}, this section contains 9774the list of all C labels to which the code in the 9775@var{AssemblerTemplate} may jump. 9776@xref{GotoLabels}. 9777 9778@code{asm} statements may not perform jumps into other @code{asm} statements, 9779only to the listed @var{GotoLabels}. 9780GCC's optimizers do not know about other jumps; therefore they cannot take 9781account of them when deciding how to optimize. 9782@end table 9783 9784The total number of input + output + goto operands is limited to 30. 9785 9786@subsubheading Remarks 9787The @code{asm} statement allows you to include assembly instructions directly 9788within C code. This may help you to maximize performance in time-sensitive 9789code or to access assembly instructions that are not readily available to C 9790programs. 9791 9792Note that extended @code{asm} statements must be inside a function. Only 9793basic @code{asm} may be outside functions (@pxref{Basic Asm}). 9794Functions declared with the @code{naked} attribute also require basic 9795@code{asm} (@pxref{Function Attributes}). 9796 9797While the uses of @code{asm} are many and varied, it may help to think of an 9798@code{asm} statement as a series of low-level instructions that convert input 9799parameters to output parameters. So a simple (if not particularly useful) 9800example for i386 using @code{asm} might look like this: 9801 9802@example 9803int src = 1; 9804int dst; 9805 9806asm ("mov %1, %0\n\t" 9807 "add $1, %0" 9808 : "=r" (dst) 9809 : "r" (src)); 9810 9811printf("%d\n", dst); 9812@end example 9813 9814This code copies @code{src} to @code{dst} and add 1 to @code{dst}. 9815 9816@anchor{Volatile} 9817@subsubsection Volatile 9818@cindex volatile @code{asm} 9819@cindex @code{asm} volatile 9820 9821GCC's optimizers sometimes discard @code{asm} statements if they determine 9822there is no need for the output variables. Also, the optimizers may move 9823code out of loops if they believe that the code will always return the same 9824result (i.e.@: none of its input values change between calls). Using the 9825@code{volatile} qualifier disables these optimizations. @code{asm} statements 9826that have no output operands and @code{asm goto} statements, 9827are implicitly volatile. 9828 9829This i386 code demonstrates a case that does not use (or require) the 9830@code{volatile} qualifier. If it is performing assertion checking, this code 9831uses @code{asm} to perform the validation. Otherwise, @code{dwRes} is 9832unreferenced by any code. As a result, the optimizers can discard the 9833@code{asm} statement, which in turn removes the need for the entire 9834@code{DoCheck} routine. By omitting the @code{volatile} qualifier when it 9835isn't needed you allow the optimizers to produce the most efficient code 9836possible. 9837 9838@example 9839void DoCheck(uint32_t dwSomeValue) 9840@{ 9841 uint32_t dwRes; 9842 9843 // Assumes dwSomeValue is not zero. 9844 asm ("bsfl %1,%0" 9845 : "=r" (dwRes) 9846 : "r" (dwSomeValue) 9847 : "cc"); 9848 9849 assert(dwRes > 3); 9850@} 9851@end example 9852 9853The next example shows a case where the optimizers can recognize that the input 9854(@code{dwSomeValue}) never changes during the execution of the function and can 9855therefore move the @code{asm} outside the loop to produce more efficient code. 9856Again, using the @code{volatile} qualifier disables this type of optimization. 9857 9858@example 9859void do_print(uint32_t dwSomeValue) 9860@{ 9861 uint32_t dwRes; 9862 9863 for (uint32_t x=0; x < 5; x++) 9864 @{ 9865 // Assumes dwSomeValue is not zero. 9866 asm ("bsfl %1,%0" 9867 : "=r" (dwRes) 9868 : "r" (dwSomeValue) 9869 : "cc"); 9870 9871 printf("%u: %u %u\n", x, dwSomeValue, dwRes); 9872 @} 9873@} 9874@end example 9875 9876The following example demonstrates a case where you need to use the 9877@code{volatile} qualifier. 9878It uses the x86 @code{rdtsc} instruction, which reads 9879the computer's time-stamp counter. Without the @code{volatile} qualifier, 9880the optimizers might assume that the @code{asm} block will always return the 9881same value and therefore optimize away the second call. 9882 9883@example 9884uint64_t msr; 9885 9886asm volatile ( "rdtsc\n\t" // Returns the time in EDX:EAX. 9887 "shl $32, %%rdx\n\t" // Shift the upper bits left. 9888 "or %%rdx, %0" // 'Or' in the lower bits. 9889 : "=a" (msr) 9890 : 9891 : "rdx"); 9892 9893printf("msr: %llx\n", msr); 9894 9895// Do other work... 9896 9897// Reprint the timestamp 9898asm volatile ( "rdtsc\n\t" // Returns the time in EDX:EAX. 9899 "shl $32, %%rdx\n\t" // Shift the upper bits left. 9900 "or %%rdx, %0" // 'Or' in the lower bits. 9901 : "=a" (msr) 9902 : 9903 : "rdx"); 9904 9905printf("msr: %llx\n", msr); 9906@end example 9907 9908GCC's optimizers do not treat this code like the non-volatile code in the 9909earlier examples. They do not move it out of loops or omit it on the 9910assumption that the result from a previous call is still valid. 9911 9912Note that the compiler can move even @code{volatile asm} instructions relative 9913to other code, including across jump instructions. For example, on many 9914targets there is a system register that controls the rounding mode of 9915floating-point operations. Setting it with a @code{volatile asm} statement, 9916as in the following PowerPC example, does not work reliably. 9917 9918@example 9919asm volatile("mtfsf 255, %0" : : "f" (fpenv)); 9920sum = x + y; 9921@end example 9922 9923The compiler may move the addition back before the @code{volatile asm} 9924statement. To make it work as expected, add an artificial dependency to 9925the @code{asm} by referencing a variable in the subsequent code, for 9926example: 9927 9928@example 9929asm volatile ("mtfsf 255,%1" : "=X" (sum) : "f" (fpenv)); 9930sum = x + y; 9931@end example 9932 9933Under certain circumstances, GCC may duplicate (or remove duplicates of) your 9934assembly code when optimizing. This can lead to unexpected duplicate symbol 9935errors during compilation if your @code{asm} code defines symbols or labels. 9936Using @samp{%=} 9937(@pxref{AssemblerTemplate}) may help resolve this problem. 9938 9939@anchor{AssemblerTemplate} 9940@subsubsection Assembler Template 9941@cindex @code{asm} assembler template 9942 9943An assembler template is a literal string containing assembler instructions. 9944The compiler replaces tokens in the template that refer 9945to inputs, outputs, and goto labels, 9946and then outputs the resulting string to the assembler. The 9947string can contain any instructions recognized by the assembler, including 9948directives. GCC does not parse the assembler instructions 9949themselves and does not know what they mean or even whether they are valid 9950assembler input. However, it does count the statements 9951(@pxref{Size of an asm}). 9952 9953You may place multiple assembler instructions together in a single @code{asm} 9954string, separated by the characters normally used in assembly code for the 9955system. A combination that works in most places is a newline to break the 9956line, plus a tab character to move to the instruction field (written as 9957@samp{\n\t}). 9958Some assemblers allow semicolons as a line separator. However, note 9959that some assembler dialects use semicolons to start a comment. 9960 9961Do not expect a sequence of @code{asm} statements to remain perfectly 9962consecutive after compilation, even when you are using the @code{volatile} 9963qualifier. If certain instructions need to remain consecutive in the output, 9964put them in a single multi-instruction @code{asm} statement. 9965 9966Accessing data from C programs without using input/output operands (such as 9967by using global symbols directly from the assembler template) may not work as 9968expected. Similarly, calling functions directly from an assembler template 9969requires a detailed understanding of the target assembler and ABI. 9970 9971Since GCC does not parse the assembler template, 9972it has no visibility of any 9973symbols it references. This may result in GCC discarding those symbols as 9974unreferenced unless they are also listed as input, output, or goto operands. 9975 9976@subsubheading Special format strings 9977 9978In addition to the tokens described by the input, output, and goto operands, 9979these tokens have special meanings in the assembler template: 9980 9981@table @samp 9982@item %% 9983Outputs a single @samp{%} into the assembler code. 9984 9985@item %= 9986Outputs a number that is unique to each instance of the @code{asm} 9987statement in the entire compilation. This option is useful when creating local 9988labels and referring to them multiple times in a single template that 9989generates multiple assembler instructions. 9990 9991@item %@{ 9992@itemx %| 9993@itemx %@} 9994Outputs @samp{@{}, @samp{|}, and @samp{@}} characters (respectively) 9995into the assembler code. When unescaped, these characters have special 9996meaning to indicate multiple assembler dialects, as described below. 9997@end table 9998 9999@subsubheading Multiple assembler dialects in @code{asm} templates 10000 10001On targets such as x86, GCC supports multiple assembler dialects. 10002The @option{-masm} option controls which dialect GCC uses as its 10003default for inline assembler. The target-specific documentation for the 10004@option{-masm} option contains the list of supported dialects, as well as the 10005default dialect if the option is not specified. This information may be 10006important to understand, since assembler code that works correctly when 10007compiled using one dialect will likely fail if compiled using another. 10008@xref{x86 Options}. 10009 10010If your code needs to support multiple assembler dialects (for example, if 10011you are writing public headers that need to support a variety of compilation 10012options), use constructs of this form: 10013 10014@example 10015@{ dialect0 | dialect1 | dialect2... @} 10016@end example 10017 10018This construct outputs @code{dialect0} 10019when using dialect #0 to compile the code, 10020@code{dialect1} for dialect #1, etc. If there are fewer alternatives within the 10021braces than the number of dialects the compiler supports, the construct 10022outputs nothing. 10023 10024For example, if an x86 compiler supports two dialects 10025(@samp{att}, @samp{intel}), an 10026assembler template such as this: 10027 10028@example 10029"bt@{l %[Offset],%[Base] | %[Base],%[Offset]@}; jc %l2" 10030@end example 10031 10032@noindent 10033is equivalent to one of 10034 10035@example 10036"btl %[Offset],%[Base] ; jc %l2" @r{/* att dialect */} 10037"bt %[Base],%[Offset]; jc %l2" @r{/* intel dialect */} 10038@end example 10039 10040Using that same compiler, this code: 10041 10042@example 10043"xchg@{l@}\t@{%%@}ebx, %1" 10044@end example 10045 10046@noindent 10047corresponds to either 10048 10049@example 10050"xchgl\t%%ebx, %1" @r{/* att dialect */} 10051"xchg\tebx, %1" @r{/* intel dialect */} 10052@end example 10053 10054There is no support for nesting dialect alternatives. 10055 10056@anchor{OutputOperands} 10057@subsubsection Output Operands 10058@cindex @code{asm} output operands 10059 10060An @code{asm} statement has zero or more output operands indicating the names 10061of C variables modified by the assembler code. 10062 10063In this i386 example, @code{old} (referred to in the template string as 10064@code{%0}) and @code{*Base} (as @code{%1}) are outputs and @code{Offset} 10065(@code{%2}) is an input: 10066 10067@example 10068bool old; 10069 10070__asm__ ("btsl %2,%1\n\t" // Turn on zero-based bit #Offset in Base. 10071 "sbb %0,%0" // Use the CF to calculate old. 10072 : "=r" (old), "+rm" (*Base) 10073 : "Ir" (Offset) 10074 : "cc"); 10075 10076return old; 10077@end example 10078 10079Operands are separated by commas. Each operand has this format: 10080 10081@example 10082@r{[} [@var{asmSymbolicName}] @r{]} @var{constraint} (@var{cvariablename}) 10083@end example 10084 10085@table @var 10086@item asmSymbolicName 10087Specifies a symbolic name for the operand. 10088Reference the name in the assembler template 10089by enclosing it in square brackets 10090(i.e.@: @samp{%[Value]}). The scope of the name is the @code{asm} statement 10091that contains the definition. Any valid C variable name is acceptable, 10092including names already defined in the surrounding code. No two operands 10093within the same @code{asm} statement can use the same symbolic name. 10094 10095When not using an @var{asmSymbolicName}, use the (zero-based) position 10096of the operand 10097in the list of operands in the assembler template. For example if there are 10098three output operands, use @samp{%0} in the template to refer to the first, 10099@samp{%1} for the second, and @samp{%2} for the third. 10100 10101@item constraint 10102A string constant specifying constraints on the placement of the operand; 10103@xref{Constraints}, for details. 10104 10105Output constraints must begin with either @samp{=} (a variable overwriting an 10106existing value) or @samp{+} (when reading and writing). When using 10107@samp{=}, do not assume the location contains the existing value 10108on entry to the @code{asm}, except 10109when the operand is tied to an input; @pxref{InputOperands,,Input Operands}. 10110 10111After the prefix, there must be one or more additional constraints 10112(@pxref{Constraints}) that describe where the value resides. Common 10113constraints include @samp{r} for register and @samp{m} for memory. 10114When you list more than one possible location (for example, @code{"=rm"}), 10115the compiler chooses the most efficient one based on the current context. 10116If you list as many alternates as the @code{asm} statement allows, you permit 10117the optimizers to produce the best possible code. 10118If you must use a specific register, but your Machine Constraints do not 10119provide sufficient control to select the specific register you want, 10120local register variables may provide a solution (@pxref{Local Register 10121Variables}). 10122 10123@item cvariablename 10124Specifies a C lvalue expression to hold the output, typically a variable name. 10125The enclosing parentheses are a required part of the syntax. 10126 10127@end table 10128 10129When the compiler selects the registers to use to 10130represent the output operands, it does not use any of the clobbered registers 10131(@pxref{Clobbers and Scratch Registers}). 10132 10133Output operand expressions must be lvalues. The compiler cannot check whether 10134the operands have data types that are reasonable for the instruction being 10135executed. For output expressions that are not directly addressable (for 10136example a bit-field), the constraint must allow a register. In that case, GCC 10137uses the register as the output of the @code{asm}, and then stores that 10138register into the output. 10139 10140Operands using the @samp{+} constraint modifier count as two operands 10141(that is, both as input and output) towards the total maximum of 30 operands 10142per @code{asm} statement. 10143 10144Use the @samp{&} constraint modifier (@pxref{Modifiers}) on all output 10145operands that must not overlap an input. Otherwise, 10146GCC may allocate the output operand in the same register as an unrelated 10147input operand, on the assumption that the assembler code consumes its 10148inputs before producing outputs. This assumption may be false if the assembler 10149code actually consists of more than one instruction. 10150 10151The same problem can occur if one output parameter (@var{a}) allows a register 10152constraint and another output parameter (@var{b}) allows a memory constraint. 10153The code generated by GCC to access the memory address in @var{b} can contain 10154registers which @emph{might} be shared by @var{a}, and GCC considers those 10155registers to be inputs to the asm. As above, GCC assumes that such input 10156registers are consumed before any outputs are written. This assumption may 10157result in incorrect behavior if the @code{asm} statement writes to @var{a} 10158before using 10159@var{b}. Combining the @samp{&} modifier with the register constraint on @var{a} 10160ensures that modifying @var{a} does not affect the address referenced by 10161@var{b}. Otherwise, the location of @var{b} 10162is undefined if @var{a} is modified before using @var{b}. 10163 10164@code{asm} supports operand modifiers on operands (for example @samp{%k2} 10165instead of simply @samp{%2}). Typically these qualifiers are hardware 10166dependent. The list of supported modifiers for x86 is found at 10167@ref{x86Operandmodifiers,x86 Operand modifiers}. 10168 10169If the C code that follows the @code{asm} makes no use of any of the output 10170operands, use @code{volatile} for the @code{asm} statement to prevent the 10171optimizers from discarding the @code{asm} statement as unneeded 10172(see @ref{Volatile}). 10173 10174This code makes no use of the optional @var{asmSymbolicName}. Therefore it 10175references the first output operand as @code{%0} (were there a second, it 10176would be @code{%1}, etc). The number of the first input operand is one greater 10177than that of the last output operand. In this i386 example, that makes 10178@code{Mask} referenced as @code{%1}: 10179 10180@example 10181uint32_t Mask = 1234; 10182uint32_t Index; 10183 10184 asm ("bsfl %1, %0" 10185 : "=r" (Index) 10186 : "r" (Mask) 10187 : "cc"); 10188@end example 10189 10190That code overwrites the variable @code{Index} (@samp{=}), 10191placing the value in a register (@samp{r}). 10192Using the generic @samp{r} constraint instead of a constraint for a specific 10193register allows the compiler to pick the register to use, which can result 10194in more efficient code. This may not be possible if an assembler instruction 10195requires a specific register. 10196 10197The following i386 example uses the @var{asmSymbolicName} syntax. 10198It produces the 10199same result as the code above, but some may consider it more readable or more 10200maintainable since reordering index numbers is not necessary when adding or 10201removing operands. The names @code{aIndex} and @code{aMask} 10202are only used in this example to emphasize which 10203names get used where. 10204It is acceptable to reuse the names @code{Index} and @code{Mask}. 10205 10206@example 10207uint32_t Mask = 1234; 10208uint32_t Index; 10209 10210 asm ("bsfl %[aMask], %[aIndex]" 10211 : [aIndex] "=r" (Index) 10212 : [aMask] "r" (Mask) 10213 : "cc"); 10214@end example 10215 10216Here are some more examples of output operands. 10217 10218@example 10219uint32_t c = 1; 10220uint32_t d; 10221uint32_t *e = &c; 10222 10223asm ("mov %[e], %[d]" 10224 : [d] "=rm" (d) 10225 : [e] "rm" (*e)); 10226@end example 10227 10228Here, @code{d} may either be in a register or in memory. Since the compiler 10229might already have the current value of the @code{uint32_t} location 10230pointed to by @code{e} 10231in a register, you can enable it to choose the best location 10232for @code{d} by specifying both constraints. 10233 10234@anchor{FlagOutputOperands} 10235@subsubsection Flag Output Operands 10236@cindex @code{asm} flag output operands 10237 10238Some targets have a special register that holds the ``flags'' for the 10239result of an operation or comparison. Normally, the contents of that 10240register are either unmodifed by the asm, or the @code{asm} statement is 10241considered to clobber the contents. 10242 10243On some targets, a special form of output operand exists by which 10244conditions in the flags register may be outputs of the asm. The set of 10245conditions supported are target specific, but the general rule is that 10246the output variable must be a scalar integer, and the value is boolean. 10247When supported, the target defines the preprocessor symbol 10248@code{__GCC_ASM_FLAG_OUTPUTS__}. 10249 10250Because of the special nature of the flag output operands, the constraint 10251may not include alternatives. 10252 10253Most often, the target has only one flags register, and thus is an implied 10254operand of many instructions. In this case, the operand should not be 10255referenced within the assembler template via @code{%0} etc, as there's 10256no corresponding text in the assembly language. 10257 10258@table @asis 10259@item ARM 10260@itemx AArch64 10261The flag output constraints for the ARM family are of the form 10262@samp{=@@cc@var{cond}} where @var{cond} is one of the standard 10263conditions defined in the ARM ARM for @code{ConditionHolds}. 10264 10265@table @code 10266@item eq 10267Z flag set, or equal 10268@item ne 10269Z flag clear or not equal 10270@item cs 10271@itemx hs 10272C flag set or unsigned greater than equal 10273@item cc 10274@itemx lo 10275C flag clear or unsigned less than 10276@item mi 10277N flag set or ``minus'' 10278@item pl 10279N flag clear or ``plus'' 10280@item vs 10281V flag set or signed overflow 10282@item vc 10283V flag clear 10284@item hi 10285unsigned greater than 10286@item ls 10287unsigned less than equal 10288@item ge 10289signed greater than equal 10290@item lt 10291signed less than 10292@item gt 10293signed greater than 10294@item le 10295signed less than equal 10296@end table 10297 10298The flag output constraints are not supported in thumb1 mode. 10299 10300@item x86 family 10301The flag output constraints for the x86 family are of the form 10302@samp{=@@cc@var{cond}} where @var{cond} is one of the standard 10303conditions defined in the ISA manual for @code{j@var{cc}} or 10304@code{set@var{cc}}. 10305 10306@table @code 10307@item a 10308``above'' or unsigned greater than 10309@item ae 10310``above or equal'' or unsigned greater than or equal 10311@item b 10312``below'' or unsigned less than 10313@item be 10314``below or equal'' or unsigned less than or equal 10315@item c 10316carry flag set 10317@item e 10318@itemx z 10319``equal'' or zero flag set 10320@item g 10321signed greater than 10322@item ge 10323signed greater than or equal 10324@item l 10325signed less than 10326@item le 10327signed less than or equal 10328@item o 10329overflow flag set 10330@item p 10331parity flag set 10332@item s 10333sign flag set 10334@item na 10335@itemx nae 10336@itemx nb 10337@itemx nbe 10338@itemx nc 10339@itemx ne 10340@itemx ng 10341@itemx nge 10342@itemx nl 10343@itemx nle 10344@itemx no 10345@itemx np 10346@itemx ns 10347@itemx nz 10348``not'' @var{flag}, or inverted versions of those above 10349@end table 10350 10351@end table 10352 10353@anchor{InputOperands} 10354@subsubsection Input Operands 10355@cindex @code{asm} input operands 10356@cindex @code{asm} expressions 10357 10358Input operands make values from C variables and expressions available to the 10359assembly code. 10360 10361Operands are separated by commas. Each operand has this format: 10362 10363@example 10364@r{[} [@var{asmSymbolicName}] @r{]} @var{constraint} (@var{cexpression}) 10365@end example 10366 10367@table @var 10368@item asmSymbolicName 10369Specifies a symbolic name for the operand. 10370Reference the name in the assembler template 10371by enclosing it in square brackets 10372(i.e.@: @samp{%[Value]}). The scope of the name is the @code{asm} statement 10373that contains the definition. Any valid C variable name is acceptable, 10374including names already defined in the surrounding code. No two operands 10375within the same @code{asm} statement can use the same symbolic name. 10376 10377When not using an @var{asmSymbolicName}, use the (zero-based) position 10378of the operand 10379in the list of operands in the assembler template. For example if there are 10380two output operands and three inputs, 10381use @samp{%2} in the template to refer to the first input operand, 10382@samp{%3} for the second, and @samp{%4} for the third. 10383 10384@item constraint 10385A string constant specifying constraints on the placement of the operand; 10386@xref{Constraints}, for details. 10387 10388Input constraint strings may not begin with either @samp{=} or @samp{+}. 10389When you list more than one possible location (for example, @samp{"irm"}), 10390the compiler chooses the most efficient one based on the current context. 10391If you must use a specific register, but your Machine Constraints do not 10392provide sufficient control to select the specific register you want, 10393local register variables may provide a solution (@pxref{Local Register 10394Variables}). 10395 10396Input constraints can also be digits (for example, @code{"0"}). This indicates 10397that the specified input must be in the same place as the output constraint 10398at the (zero-based) index in the output constraint list. 10399When using @var{asmSymbolicName} syntax for the output operands, 10400you may use these names (enclosed in brackets @samp{[]}) instead of digits. 10401 10402@item cexpression 10403This is the C variable or expression being passed to the @code{asm} statement 10404as input. The enclosing parentheses are a required part of the syntax. 10405 10406@end table 10407 10408When the compiler selects the registers to use to represent the input 10409operands, it does not use any of the clobbered registers 10410(@pxref{Clobbers and Scratch Registers}). 10411 10412If there are no output operands but there are input operands, place two 10413consecutive colons where the output operands would go: 10414 10415@example 10416__asm__ ("some instructions" 10417 : /* No outputs. */ 10418 : "r" (Offset / 8)); 10419@end example 10420 10421@strong{Warning:} Do @emph{not} modify the contents of input-only operands 10422(except for inputs tied to outputs). The compiler assumes that on exit from 10423the @code{asm} statement these operands contain the same values as they 10424had before executing the statement. 10425It is @emph{not} possible to use clobbers 10426to inform the compiler that the values in these inputs are changing. One 10427common work-around is to tie the changing input variable to an output variable 10428that never gets used. Note, however, that if the code that follows the 10429@code{asm} statement makes no use of any of the output operands, the GCC 10430optimizers may discard the @code{asm} statement as unneeded 10431(see @ref{Volatile}). 10432 10433@code{asm} supports operand modifiers on operands (for example @samp{%k2} 10434instead of simply @samp{%2}). Typically these qualifiers are hardware 10435dependent. The list of supported modifiers for x86 is found at 10436@ref{x86Operandmodifiers,x86 Operand modifiers}. 10437 10438In this example using the fictitious @code{combine} instruction, the 10439constraint @code{"0"} for input operand 1 says that it must occupy the same 10440location as output operand 0. Only input operands may use numbers in 10441constraints, and they must each refer to an output operand. Only a number (or 10442the symbolic assembler name) in the constraint can guarantee that one operand 10443is in the same place as another. The mere fact that @code{foo} is the value of 10444both operands is not enough to guarantee that they are in the same place in 10445the generated assembler code. 10446 10447@example 10448asm ("combine %2, %0" 10449 : "=r" (foo) 10450 : "0" (foo), "g" (bar)); 10451@end example 10452 10453Here is an example using symbolic names. 10454 10455@example 10456asm ("cmoveq %1, %2, %[result]" 10457 : [result] "=r"(result) 10458 : "r" (test), "r" (new), "[result]" (old)); 10459@end example 10460 10461@anchor{Clobbers and Scratch Registers} 10462@subsubsection Clobbers and Scratch Registers 10463@cindex @code{asm} clobbers 10464@cindex @code{asm} scratch registers 10465 10466While the compiler is aware of changes to entries listed in the output 10467operands, the inline @code{asm} code may modify more than just the outputs. For 10468example, calculations may require additional registers, or the processor may 10469overwrite a register as a side effect of a particular assembler instruction. 10470In order to inform the compiler of these changes, list them in the clobber 10471list. Clobber list items are either register names or the special clobbers 10472(listed below). Each clobber list item is a string constant 10473enclosed in double quotes and separated by commas. 10474 10475Clobber descriptions may not in any way overlap with an input or output 10476operand. For example, you may not have an operand describing a register class 10477with one member when listing that register in the clobber list. Variables 10478declared to live in specific registers (@pxref{Explicit Register 10479Variables}) and used 10480as @code{asm} input or output operands must have no part mentioned in the 10481clobber description. In particular, there is no way to specify that input 10482operands get modified without also specifying them as output operands. 10483 10484When the compiler selects which registers to use to represent input and output 10485operands, it does not use any of the clobbered registers. As a result, 10486clobbered registers are available for any use in the assembler code. 10487 10488Another restriction is that the clobber list should not contain the 10489stack pointer register. This is because the compiler requires the 10490value of the stack pointer to be the same after an @code{asm} 10491statement as it was on entry to the statement. However, previous 10492versions of GCC did not enforce this rule and allowed the stack 10493pointer to appear in the list, with unclear semantics. This behavior 10494is deprecated and listing the stack pointer may become an error in 10495future versions of GCC@. 10496 10497Here is a realistic example for the VAX showing the use of clobbered 10498registers: 10499 10500@example 10501asm volatile ("movc3 %0, %1, %2" 10502 : /* No outputs. */ 10503 : "g" (from), "g" (to), "g" (count) 10504 : "r0", "r1", "r2", "r3", "r4", "r5", "memory"); 10505@end example 10506 10507Also, there are two special clobber arguments: 10508 10509@table @code 10510@item "cc" 10511The @code{"cc"} clobber indicates that the assembler code modifies the flags 10512register. On some machines, GCC represents the condition codes as a specific 10513hardware register; @code{"cc"} serves to name this register. 10514On other machines, condition code handling is different, 10515and specifying @code{"cc"} has no effect. But 10516it is valid no matter what the target. 10517 10518@item "memory" 10519The @code{"memory"} clobber tells the compiler that the assembly code 10520performs memory 10521reads or writes to items other than those listed in the input and output 10522operands (for example, accessing the memory pointed to by one of the input 10523parameters). To ensure memory contains correct values, GCC may need to flush 10524specific register values to memory before executing the @code{asm}. Further, 10525the compiler does not assume that any values read from memory before an 10526@code{asm} remain unchanged after that @code{asm}; it reloads them as 10527needed. 10528Using the @code{"memory"} clobber effectively forms a read/write 10529memory barrier for the compiler. 10530 10531Note that this clobber does not prevent the @emph{processor} from doing 10532speculative reads past the @code{asm} statement. To prevent that, you need 10533processor-specific fence instructions. 10534 10535@end table 10536 10537Flushing registers to memory has performance implications and may be 10538an issue for time-sensitive code. You can provide better information 10539to GCC to avoid this, as shown in the following examples. At a 10540minimum, aliasing rules allow GCC to know what memory @emph{doesn't} 10541need to be flushed. 10542 10543Here is a fictitious sum of squares instruction, that takes two 10544pointers to floating point values in memory and produces a floating 10545point register output. 10546Notice that @code{x}, and @code{y} both appear twice in the @code{asm} 10547parameters, once to specify memory accessed, and once to specify a 10548base register used by the @code{asm}. You won't normally be wasting a 10549register by doing this as GCC can use the same register for both 10550purposes. However, it would be foolish to use both @code{%1} and 10551@code{%3} for @code{x} in this @code{asm} and expect them to be the 10552same. In fact, @code{%3} may well not be a register. It might be a 10553symbolic memory reference to the object pointed to by @code{x}. 10554 10555@smallexample 10556asm ("sumsq %0, %1, %2" 10557 : "+f" (result) 10558 : "r" (x), "r" (y), "m" (*x), "m" (*y)); 10559@end smallexample 10560 10561Here is a fictitious @code{*z++ = *x++ * *y++} instruction. 10562Notice that the @code{x}, @code{y} and @code{z} pointer registers 10563must be specified as input/output because the @code{asm} modifies 10564them. 10565 10566@smallexample 10567asm ("vecmul %0, %1, %2" 10568 : "+r" (z), "+r" (x), "+r" (y), "=m" (*z) 10569 : "m" (*x), "m" (*y)); 10570@end smallexample 10571 10572An x86 example where the string memory argument is of unknown length. 10573 10574@smallexample 10575asm("repne scasb" 10576 : "=c" (count), "+D" (p) 10577 : "m" (*(const char (*)[]) p), "0" (-1), "a" (0)); 10578@end smallexample 10579 10580If you know the above will only be reading a ten byte array then you 10581could instead use a memory input like: 10582@code{"m" (*(const char (*)[10]) p)}. 10583 10584Here is an example of a PowerPC vector scale implemented in assembly, 10585complete with vector and condition code clobbers, and some initialized 10586offset registers that are unchanged by the @code{asm}. 10587 10588@smallexample 10589void 10590dscal (size_t n, double *x, double alpha) 10591@{ 10592 asm ("/* lots of asm here */" 10593 : "+m" (*(double (*)[n]) x), "+&r" (n), "+b" (x) 10594 : "d" (alpha), "b" (32), "b" (48), "b" (64), 10595 "b" (80), "b" (96), "b" (112) 10596 : "cr0", 10597 "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39", 10598 "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47"); 10599@} 10600@end smallexample 10601 10602Rather than allocating fixed registers via clobbers to provide scratch 10603registers for an @code{asm} statement, an alternative is to define a 10604variable and make it an early-clobber output as with @code{a2} and 10605@code{a3} in the example below. This gives the compiler register 10606allocator more freedom. You can also define a variable and make it an 10607output tied to an input as with @code{a0} and @code{a1}, tied 10608respectively to @code{ap} and @code{lda}. Of course, with tied 10609outputs your @code{asm} can't use the input value after modifying the 10610output register since they are one and the same register. What's 10611more, if you omit the early-clobber on the output, it is possible that 10612GCC might allocate the same register to another of the inputs if GCC 10613could prove they had the same value on entry to the @code{asm}. This 10614is why @code{a1} has an early-clobber. Its tied input, @code{lda} 10615might conceivably be known to have the value 16 and without an 10616early-clobber share the same register as @code{%11}. On the other 10617hand, @code{ap} can't be the same as any of the other inputs, so an 10618early-clobber on @code{a0} is not needed. It is also not desirable in 10619this case. An early-clobber on @code{a0} would cause GCC to allocate 10620a separate register for the @code{"m" (*(const double (*)[]) ap)} 10621input. Note that tying an input to an output is the way to set up an 10622initialized temporary register modified by an @code{asm} statement. 10623An input not tied to an output is assumed by GCC to be unchanged, for 10624example @code{"b" (16)} below sets up @code{%11} to 16, and GCC might 10625use that register in following code if the value 16 happened to be 10626needed. You can even use a normal @code{asm} output for a scratch if 10627all inputs that might share the same register are consumed before the 10628scratch is used. The VSX registers clobbered by the @code{asm} 10629statement could have used this technique except for GCC's limit on the 10630number of @code{asm} parameters. 10631 10632@smallexample 10633static void 10634dgemv_kernel_4x4 (long n, const double *ap, long lda, 10635 const double *x, double *y, double alpha) 10636@{ 10637 double *a0; 10638 double *a1; 10639 double *a2; 10640 double *a3; 10641 10642 __asm__ 10643 ( 10644 /* lots of asm here */ 10645 "#n=%1 ap=%8=%12 lda=%13 x=%7=%10 y=%0=%2 alpha=%9 o16=%11\n" 10646 "#a0=%3 a1=%4 a2=%5 a3=%6" 10647 : 10648 "+m" (*(double (*)[n]) y), 10649 "+&r" (n), // 1 10650 "+b" (y), // 2 10651 "=b" (a0), // 3 10652 "=&b" (a1), // 4 10653 "=&b" (a2), // 5 10654 "=&b" (a3) // 6 10655 : 10656 "m" (*(const double (*)[n]) x), 10657 "m" (*(const double (*)[]) ap), 10658 "d" (alpha), // 9 10659 "r" (x), // 10 10660 "b" (16), // 11 10661 "3" (ap), // 12 10662 "4" (lda) // 13 10663 : 10664 "cr0", 10665 "vs32","vs33","vs34","vs35","vs36","vs37", 10666 "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47" 10667 ); 10668@} 10669@end smallexample 10670 10671@anchor{GotoLabels} 10672@subsubsection Goto Labels 10673@cindex @code{asm} goto labels 10674 10675@code{asm goto} allows assembly code to jump to one or more C labels. The 10676@var{GotoLabels} section in an @code{asm goto} statement contains 10677a comma-separated 10678list of all C labels to which the assembler code may jump. GCC assumes that 10679@code{asm} execution falls through to the next statement (if this is not the 10680case, consider using the @code{__builtin_unreachable} intrinsic after the 10681@code{asm} statement). Optimization of @code{asm goto} may be improved by 10682using the @code{hot} and @code{cold} label attributes (@pxref{Label 10683Attributes}). 10684 10685If the assembler code does modify anything, use the @code{"memory"} clobber 10686to force the 10687optimizers to flush all register values to memory and reload them if 10688necessary after the @code{asm} statement. 10689 10690Also note that an @code{asm goto} statement is always implicitly 10691considered volatile. 10692 10693Be careful when you set output operands inside @code{asm goto} only on 10694some possible control flow paths. If you don't set up the output on 10695given path and never use it on this path, it is okay. Otherwise, you 10696should use @samp{+} constraint modifier meaning that the operand is 10697input and output one. With this modifier you will have the correct 10698values on all possible paths from the @code{asm goto}. 10699 10700To reference a label in the assembler template, prefix it with 10701@samp{%l} (lowercase @samp{L}) followed by its (zero-based) position 10702in @var{GotoLabels} plus the number of input and output operands. 10703Output operand with constraint modifier @samp{+} is counted as two 10704operands because it is considered as one output and one input operand. 10705For example, if the @code{asm} has three inputs, one output operand 10706with constraint modifier @samp{+} and one output operand with 10707constraint modifier @samp{=} and references two labels, refer to the 10708first label as @samp{%l6} and the second as @samp{%l7}). 10709 10710Alternately, you can reference labels using the actual C label name 10711enclosed in brackets. For example, to reference a label named 10712@code{carry}, you can use @samp{%l[carry]}. The label must still be 10713listed in the @var{GotoLabels} section when using this approach. It 10714is better to use the named references for labels as in this case you 10715can avoid counting input and output operands and special treatment of 10716output operands with constraint modifier @samp{+}. 10717 10718Here is an example of @code{asm goto} for i386: 10719 10720@example 10721asm goto ( 10722 "btl %1, %0\n\t" 10723 "jc %l2" 10724 : /* No outputs. */ 10725 : "r" (p1), "r" (p2) 10726 : "cc" 10727 : carry); 10728 10729return 0; 10730 10731carry: 10732return 1; 10733@end example 10734 10735The following example shows an @code{asm goto} that uses a memory clobber. 10736 10737@example 10738int frob(int x) 10739@{ 10740 int y; 10741 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5" 10742 : /* No outputs. */ 10743 : "r"(x), "r"(&y) 10744 : "r5", "memory" 10745 : error); 10746 return y; 10747error: 10748 return -1; 10749@} 10750@end example 10751 10752The following example shows an @code{asm goto} that uses an output. 10753 10754@example 10755int foo(int count) 10756@{ 10757 asm goto ("dec %0; jb %l[stop]" 10758 : "+r" (count) 10759 : 10760 : 10761 : stop); 10762 return count; 10763stop: 10764 return 0; 10765@} 10766@end example 10767 10768The following artificial example shows an @code{asm goto} that sets 10769up an output only on one path inside the @code{asm goto}. Usage of 10770constraint modifier @code{=} instead of @code{+} would be wrong as 10771@code{factor} is used on all paths from the @code{asm goto}. 10772 10773@example 10774int foo(int inp) 10775@{ 10776 int factor = 0; 10777 asm goto ("cmp %1, 10; jb %l[lab]; mov 2, %0" 10778 : "+r" (factor) 10779 : "r" (inp) 10780 : 10781 : lab); 10782lab: 10783 return inp * factor; /* return 2 * inp or 0 if inp < 10 */ 10784@} 10785@end example 10786 10787@anchor{x86Operandmodifiers} 10788@subsubsection x86 Operand Modifiers 10789 10790References to input, output, and goto operands in the assembler template 10791of extended @code{asm} statements can use 10792modifiers to affect the way the operands are formatted in 10793the code output to the assembler. For example, the 10794following code uses the @samp{h} and @samp{b} modifiers for x86: 10795 10796@example 10797uint16_t num; 10798asm volatile ("xchg %h0, %b0" : "+a" (num) ); 10799@end example 10800 10801@noindent 10802These modifiers generate this assembler code: 10803 10804@example 10805xchg %ah, %al 10806@end example 10807 10808The rest of this discussion uses the following code for illustrative purposes. 10809 10810@example 10811int main() 10812@{ 10813 int iInt = 1; 10814 10815top: 10816 10817 asm volatile goto ("some assembler instructions here" 10818 : /* No outputs. */ 10819 : "q" (iInt), "X" (sizeof(unsigned char) + 1), "i" (42) 10820 : /* No clobbers. */ 10821 : top); 10822@} 10823@end example 10824 10825With no modifiers, this is what the output from the operands would be 10826for the @samp{att} and @samp{intel} dialects of assembler: 10827 10828@multitable {Operand} {$.L2} {OFFSET FLAT:.L2} 10829@headitem Operand @tab @samp{att} @tab @samp{intel} 10830@item @code{%0} 10831@tab @code{%eax} 10832@tab @code{eax} 10833@item @code{%1} 10834@tab @code{$2} 10835@tab @code{2} 10836@item @code{%3} 10837@tab @code{$.L3} 10838@tab @code{OFFSET FLAT:.L3} 10839@item @code{%4} 10840@tab @code{$8} 10841@tab @code{8} 10842@item @code{%5} 10843@tab @code{%xmm0} 10844@tab @code{xmm0} 10845@item @code{%7} 10846@tab @code{$0} 10847@tab @code{0} 10848@end multitable 10849 10850The table below shows the list of supported modifiers and their effects. 10851 10852@multitable {Modifier} {Print the opcode suffix for the size of th} {Operand} {@samp{att}} {@samp{intel}} 10853@headitem Modifier @tab Description @tab Operand @tab @samp{att} @tab @samp{intel} 10854@item @code{A} 10855@tab Print an absolute memory reference. 10856@tab @code{%A0} 10857@tab @code{*%rax} 10858@tab @code{rax} 10859@item @code{b} 10860@tab Print the QImode name of the register. 10861@tab @code{%b0} 10862@tab @code{%al} 10863@tab @code{al} 10864@item @code{B} 10865@tab print the opcode suffix of b. 10866@tab @code{%B0} 10867@tab @code{b} 10868@tab 10869@item @code{c} 10870@tab Require a constant operand and print the constant expression with no punctuation. 10871@tab @code{%c1} 10872@tab @code{2} 10873@tab @code{2} 10874@item @code{d} 10875@tab print duplicated register operand for AVX instruction. 10876@tab @code{%d5} 10877@tab @code{%xmm0, %xmm0} 10878@tab @code{xmm0, xmm0} 10879@item @code{E} 10880@tab Print the address in Double Integer (DImode) mode (8 bytes) when the target is 64-bit. 10881Otherwise mode is unspecified (VOIDmode). 10882@tab @code{%E1} 10883@tab @code{%(rax)} 10884@tab @code{[rax]} 10885@item @code{g} 10886@tab Print the V16SFmode name of the register. 10887@tab @code{%g0} 10888@tab @code{%zmm0} 10889@tab @code{zmm0} 10890@item @code{h} 10891@tab Print the QImode name for a ``high'' register. 10892@tab @code{%h0} 10893@tab @code{%ah} 10894@tab @code{ah} 10895@item @code{H} 10896@tab Add 8 bytes to an offsettable memory reference. Useful when accessing the 10897high 8 bytes of SSE values. For a memref in (%rax), it generates 10898@tab @code{%H0} 10899@tab @code{8(%rax)} 10900@tab @code{8[rax]} 10901@item @code{k} 10902@tab Print the SImode name of the register. 10903@tab @code{%k0} 10904@tab @code{%eax} 10905@tab @code{eax} 10906@item @code{l} 10907@tab Print the label name with no punctuation. 10908@tab @code{%l3} 10909@tab @code{.L3} 10910@tab @code{.L3} 10911@item @code{L} 10912@tab print the opcode suffix of l. 10913@tab @code{%L0} 10914@tab @code{l} 10915@tab 10916@item @code{N} 10917@tab print maskz. 10918@tab @code{%N7} 10919@tab @code{@{z@}} 10920@tab @code{@{z@}} 10921@item @code{p} 10922@tab Print raw symbol name (without syntax-specific prefixes). 10923@tab @code{%p2} 10924@tab @code{42} 10925@tab @code{42} 10926@item @code{P} 10927@tab If used for a function, print the PLT suffix and generate PIC code. 10928For example, emit @code{foo@@PLT} instead of 'foo' for the function 10929foo(). If used for a constant, drop all syntax-specific prefixes and 10930issue the bare constant. See @code{p} above. 10931@item @code{q} 10932@tab Print the DImode name of the register. 10933@tab @code{%q0} 10934@tab @code{%rax} 10935@tab @code{rax} 10936@item @code{Q} 10937@tab print the opcode suffix of q. 10938@tab @code{%Q0} 10939@tab @code{q} 10940@tab 10941@item @code{R} 10942@tab print embedded rounding and sae. 10943@tab @code{%R4} 10944@tab @code{@{rn-sae@}, } 10945@tab @code{, @{rn-sae@}} 10946@item @code{r} 10947@tab print only sae. 10948@tab @code{%r4} 10949@tab @code{@{sae@}, } 10950@tab @code{, @{sae@}} 10951@item @code{s} 10952@tab print a shift double count, followed by the assemblers argument 10953delimiterprint the opcode suffix of s. 10954@tab @code{%s1} 10955@tab @code{$2, } 10956@tab @code{2, } 10957@item @code{S} 10958@tab print the opcode suffix of s. 10959@tab @code{%S0} 10960@tab @code{s} 10961@tab 10962@item @code{t} 10963@tab print the V8SFmode name of the register. 10964@tab @code{%t5} 10965@tab @code{%ymm0} 10966@tab @code{ymm0} 10967@item @code{T} 10968@tab print the opcode suffix of t. 10969@tab @code{%T0} 10970@tab @code{t} 10971@tab 10972@item @code{V} 10973@tab print naked full integer register name without %. 10974@tab @code{%V0} 10975@tab @code{eax} 10976@tab @code{eax} 10977@item @code{w} 10978@tab Print the HImode name of the register. 10979@tab @code{%w0} 10980@tab @code{%ax} 10981@tab @code{ax} 10982@item @code{W} 10983@tab print the opcode suffix of w. 10984@tab @code{%W0} 10985@tab @code{w} 10986@tab 10987@item @code{x} 10988@tab print the V4SFmode name of the register. 10989@tab @code{%x5} 10990@tab @code{%xmm0} 10991@tab @code{xmm0} 10992@item @code{y} 10993@tab print "st(0)" instead of "st" as a register. 10994@tab @code{%y6} 10995@tab @code{%st(0)} 10996@tab @code{st(0)} 10997@item @code{z} 10998@tab Print the opcode suffix for the size of the current integer operand (one of @code{b}/@code{w}/@code{l}/@code{q}). 10999@tab @code{%z0} 11000@tab @code{l} 11001@tab 11002@item @code{Z} 11003@tab Like @code{z}, with special suffixes for x87 instructions. 11004@end multitable 11005 11006 11007@anchor{x86floatingpointasmoperands} 11008@subsubsection x86 Floating-Point @code{asm} Operands 11009 11010On x86 targets, there are several rules on the usage of stack-like registers 11011in the operands of an @code{asm}. These rules apply only to the operands 11012that are stack-like registers: 11013 11014@enumerate 11015@item 11016Given a set of input registers that die in an @code{asm}, it is 11017necessary to know which are implicitly popped by the @code{asm}, and 11018which must be explicitly popped by GCC@. 11019 11020An input register that is implicitly popped by the @code{asm} must be 11021explicitly clobbered, unless it is constrained to match an 11022output operand. 11023 11024@item 11025For any input register that is implicitly popped by an @code{asm}, it is 11026necessary to know how to adjust the stack to compensate for the pop. 11027If any non-popped input is closer to the top of the reg-stack than 11028the implicitly popped register, it would not be possible to know what the 11029stack looked like---it's not clear how the rest of the stack ``slides 11030up''. 11031 11032All implicitly popped input registers must be closer to the top of 11033the reg-stack than any input that is not implicitly popped. 11034 11035It is possible that if an input dies in an @code{asm}, the compiler might 11036use the input register for an output reload. Consider this example: 11037 11038@smallexample 11039asm ("foo" : "=t" (a) : "f" (b)); 11040@end smallexample 11041 11042@noindent 11043This code says that input @code{b} is not popped by the @code{asm}, and that 11044the @code{asm} pushes a result onto the reg-stack, i.e., the stack is one 11045deeper after the @code{asm} than it was before. But, it is possible that 11046reload may think that it can use the same register for both the input and 11047the output. 11048 11049To prevent this from happening, 11050if any input operand uses the @samp{f} constraint, all output register 11051constraints must use the @samp{&} early-clobber modifier. 11052 11053The example above is correctly written as: 11054 11055@smallexample 11056asm ("foo" : "=&t" (a) : "f" (b)); 11057@end smallexample 11058 11059@item 11060Some operands need to be in particular places on the stack. All 11061output operands fall in this category---GCC has no other way to 11062know which registers the outputs appear in unless you indicate 11063this in the constraints. 11064 11065Output operands must specifically indicate which register an output 11066appears in after an @code{asm}. @samp{=f} is not allowed: the operand 11067constraints must select a class with a single register. 11068 11069@item 11070Output operands may not be ``inserted'' between existing stack registers. 11071Since no 387 opcode uses a read/write operand, all output operands 11072are dead before the @code{asm}, and are pushed by the @code{asm}. 11073It makes no sense to push anywhere but the top of the reg-stack. 11074 11075Output operands must start at the top of the reg-stack: output 11076operands may not ``skip'' a register. 11077 11078@item 11079Some @code{asm} statements may need extra stack space for internal 11080calculations. This can be guaranteed by clobbering stack registers 11081unrelated to the inputs and outputs. 11082 11083@end enumerate 11084 11085This @code{asm} 11086takes one input, which is internally popped, and produces two outputs. 11087 11088@smallexample 11089asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); 11090@end smallexample 11091 11092@noindent 11093This @code{asm} takes two inputs, which are popped by the @code{fyl2xp1} opcode, 11094and replaces them with one output. The @code{st(1)} clobber is necessary 11095for the compiler to know that @code{fyl2xp1} pops both inputs. 11096 11097@smallexample 11098asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); 11099@end smallexample 11100 11101@anchor{msp430Operandmodifiers} 11102@subsubsection MSP430 Operand Modifiers 11103 11104The list below describes the supported modifiers and their effects for MSP430. 11105 11106@multitable @columnfractions .10 .90 11107@headitem Modifier @tab Description 11108@item @code{A} @tab Select low 16-bits of the constant/register/memory operand. 11109@item @code{B} @tab Select high 16-bits of the constant/register/memory 11110operand. 11111@item @code{C} @tab Select bits 32-47 of the constant/register/memory operand. 11112@item @code{D} @tab Select bits 48-63 of the constant/register/memory operand. 11113@item @code{H} @tab Equivalent to @code{B} (for backwards compatibility). 11114@item @code{I} @tab Print the inverse (logical @code{NOT}) of the constant 11115value. 11116@item @code{J} @tab Print an integer without a @code{#} prefix. 11117@item @code{L} @tab Equivalent to @code{A} (for backwards compatibility). 11118@item @code{O} @tab Offset of the current frame from the top of the stack. 11119@item @code{Q} @tab Use the @code{A} instruction postfix. 11120@item @code{R} @tab Inverse of condition code, for unsigned comparisons. 11121@item @code{W} @tab Subtract 16 from the constant value. 11122@item @code{X} @tab Use the @code{X} instruction postfix. 11123@item @code{Y} @tab Subtract 4 from the constant value. 11124@item @code{Z} @tab Subtract 1 from the constant value. 11125@item @code{b} @tab Append @code{.B}, @code{.W} or @code{.A} to the 11126instruction, depending on the mode. 11127@item @code{d} @tab Offset 1 byte of a memory reference or constant value. 11128@item @code{e} @tab Offset 3 bytes of a memory reference or constant value. 11129@item @code{f} @tab Offset 5 bytes of a memory reference or constant value. 11130@item @code{g} @tab Offset 7 bytes of a memory reference or constant value. 11131@item @code{p} @tab Print the value of 2, raised to the power of the given 11132constant. Used to select the specified bit position. 11133@item @code{r} @tab Inverse of condition code, for signed comparisons. 11134@item @code{x} @tab Equivialent to @code{X}, but only for pointers. 11135@end multitable 11136 11137@lowersections 11138@include md.texi 11139@raisesections 11140 11141@node Asm Labels 11142@subsection Controlling Names Used in Assembler Code 11143@cindex assembler names for identifiers 11144@cindex names used in assembler code 11145@cindex identifiers, names in assembler code 11146 11147You can specify the name to be used in the assembler code for a C 11148function or variable by writing the @code{asm} (or @code{__asm__}) 11149keyword after the declarator. 11150It is up to you to make sure that the assembler names you choose do not 11151conflict with any other assembler symbols, or reference registers. 11152 11153@subsubheading Assembler names for data: 11154 11155This sample shows how to specify the assembler name for data: 11156 11157@smallexample 11158int foo asm ("myfoo") = 2; 11159@end smallexample 11160 11161@noindent 11162This specifies that the name to be used for the variable @code{foo} in 11163the assembler code should be @samp{myfoo} rather than the usual 11164@samp{_foo}. 11165 11166On systems where an underscore is normally prepended to the name of a C 11167variable, this feature allows you to define names for the 11168linker that do not start with an underscore. 11169 11170GCC does not support using this feature with a non-static local variable 11171since such variables do not have assembler names. If you are 11172trying to put the variable in a particular register, see 11173@ref{Explicit Register Variables}. 11174 11175@subsubheading Assembler names for functions: 11176 11177To specify the assembler name for functions, write a declaration for the 11178function before its definition and put @code{asm} there, like this: 11179 11180@smallexample 11181int func (int x, int y) asm ("MYFUNC"); 11182 11183int func (int x, int y) 11184@{ 11185 /* @r{@dots{}} */ 11186@end smallexample 11187 11188@noindent 11189This specifies that the name to be used for the function @code{func} in 11190the assembler code should be @code{MYFUNC}. 11191 11192@node Explicit Register Variables 11193@subsection Variables in Specified Registers 11194@anchor{Explicit Reg Vars} 11195@cindex explicit register variables 11196@cindex variables in specified registers 11197@cindex specified registers 11198 11199GNU C allows you to associate specific hardware registers with C 11200variables. In almost all cases, allowing the compiler to assign 11201registers produces the best code. However under certain unusual 11202circumstances, more precise control over the variable storage is 11203required. 11204 11205Both global and local variables can be associated with a register. The 11206consequences of performing this association are very different between 11207the two, as explained in the sections below. 11208 11209@menu 11210* Global Register Variables:: Variables declared at global scope. 11211* Local Register Variables:: Variables declared within a function. 11212@end menu 11213 11214@node Global Register Variables 11215@subsubsection Defining Global Register Variables 11216@anchor{Global Reg Vars} 11217@cindex global register variables 11218@cindex registers, global variables in 11219@cindex registers, global allocation 11220 11221You can define a global register variable and associate it with a specified 11222register like this: 11223 11224@smallexample 11225register int *foo asm ("r12"); 11226@end smallexample 11227 11228@noindent 11229Here @code{r12} is the name of the register that should be used. Note that 11230this is the same syntax used for defining local register variables, but for 11231a global variable the declaration appears outside a function. The 11232@code{register} keyword is required, and cannot be combined with 11233@code{static}. The register name must be a valid register name for the 11234target platform. 11235 11236Do not use type qualifiers such as @code{const} and @code{volatile}, as 11237the outcome may be contrary to expectations. In particular, using the 11238@code{volatile} qualifier does not fully prevent the compiler from 11239optimizing accesses to the register. 11240 11241Registers are a scarce resource on most systems and allowing the 11242compiler to manage their usage usually results in the best code. However, 11243under special circumstances it can make sense to reserve some globally. 11244For example this may be useful in programs such as programming language 11245interpreters that have a couple of global variables that are accessed 11246very often. 11247 11248After defining a global register variable, for the current compilation 11249unit: 11250 11251@itemize @bullet 11252@item If the register is a call-saved register, call ABI is affected: 11253the register will not be restored in function epilogue sequences after 11254the variable has been assigned. Therefore, functions cannot safely 11255return to callers that assume standard ABI. 11256@item Conversely, if the register is a call-clobbered register, making 11257calls to functions that use standard ABI may lose contents of the variable. 11258Such calls may be created by the compiler even if none are evident in 11259the original program, for example when libgcc functions are used to 11260make up for unavailable instructions. 11261@item Accesses to the variable may be optimized as usual and the register 11262remains available for allocation and use in any computations, provided that 11263observable values of the variable are not affected. 11264@item If the variable is referenced in inline assembly, the type of access 11265must be provided to the compiler via constraints (@pxref{Constraints}). 11266Accesses from basic asms are not supported. 11267@end itemize 11268 11269Note that these points @emph{only} apply to code that is compiled with the 11270definition. The behavior of code that is merely linked in (for example 11271code from libraries) is not affected. 11272 11273If you want to recompile source files that do not actually use your global 11274register variable so they do not use the specified register for any other 11275purpose, you need not actually add the global register declaration to 11276their source code. It suffices to specify the compiler option 11277@option{-ffixed-@var{reg}} (@pxref{Code Gen Options}) to reserve the 11278register. 11279 11280@subsubheading Declaring the variable 11281 11282Global register variables cannot have initial values, because an 11283executable file has no means to supply initial contents for a register. 11284 11285When selecting a register, choose one that is normally saved and 11286restored by function calls on your machine. This ensures that code 11287which is unaware of this reservation (such as library routines) will 11288restore it before returning. 11289 11290On machines with register windows, be sure to choose a global 11291register that is not affected magically by the function call mechanism. 11292 11293@subsubheading Using the variable 11294 11295@cindex @code{qsort}, and global register variables 11296When calling routines that are not aware of the reservation, be 11297cautious if those routines call back into code which uses them. As an 11298example, if you call the system library version of @code{qsort}, it may 11299clobber your registers during execution, but (if you have selected 11300appropriate registers) it will restore them before returning. However 11301it will @emph{not} restore them before calling @code{qsort}'s comparison 11302function. As a result, global values will not reliably be available to 11303the comparison function unless the @code{qsort} function itself is rebuilt. 11304 11305Similarly, it is not safe to access the global register variables from signal 11306handlers or from more than one thread of control. Unless you recompile 11307them specially for the task at hand, the system library routines may 11308temporarily use the register for other things. Furthermore, since the register 11309is not reserved exclusively for the variable, accessing it from handlers of 11310asynchronous signals may observe unrelated temporary values residing in the 11311register. 11312 11313@cindex register variable after @code{longjmp} 11314@cindex global register after @code{longjmp} 11315@cindex value after @code{longjmp} 11316@findex longjmp 11317@findex setjmp 11318On most machines, @code{longjmp} restores to each global register 11319variable the value it had at the time of the @code{setjmp}. On some 11320machines, however, @code{longjmp} does not change the value of global 11321register variables. To be portable, the function that called @code{setjmp} 11322should make other arrangements to save the values of the global register 11323variables, and to restore them in a @code{longjmp}. This way, the same 11324thing happens regardless of what @code{longjmp} does. 11325 11326@node Local Register Variables 11327@subsubsection Specifying Registers for Local Variables 11328@anchor{Local Reg Vars} 11329@cindex local variables, specifying registers 11330@cindex specifying registers for local variables 11331@cindex registers for local variables 11332 11333You can define a local register variable and associate it with a specified 11334register like this: 11335 11336@smallexample 11337register int *foo asm ("r12"); 11338@end smallexample 11339 11340@noindent 11341Here @code{r12} is the name of the register that should be used. Note 11342that this is the same syntax used for defining global register variables, 11343but for a local variable the declaration appears within a function. The 11344@code{register} keyword is required, and cannot be combined with 11345@code{static}. The register name must be a valid register name for the 11346target platform. 11347 11348Do not use type qualifiers such as @code{const} and @code{volatile}, as 11349the outcome may be contrary to expectations. In particular, when the 11350@code{const} qualifier is used, the compiler may substitute the 11351variable with its initializer in @code{asm} statements, which may cause 11352the corresponding operand to appear in a different register. 11353 11354As with global register variables, it is recommended that you choose 11355a register that is normally saved and restored by function calls on your 11356machine, so that calls to library routines will not clobber it. 11357 11358The only supported use for this feature is to specify registers 11359for input and output operands when calling Extended @code{asm} 11360(@pxref{Extended Asm}). This may be necessary if the constraints for a 11361particular machine don't provide sufficient control to select the desired 11362register. To force an operand into a register, create a local variable 11363and specify the register name after the variable's declaration. Then use 11364the local variable for the @code{asm} operand and specify any constraint 11365letter that matches the register: 11366 11367@smallexample 11368register int *p1 asm ("r0") = @dots{}; 11369register int *p2 asm ("r1") = @dots{}; 11370register int *result asm ("r0"); 11371asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2)); 11372@end smallexample 11373 11374@emph{Warning:} In the above example, be aware that a register (for example 11375@code{r0}) can be call-clobbered by subsequent code, including function 11376calls and library calls for arithmetic operators on other variables (for 11377example the initialization of @code{p2}). In this case, use temporary 11378variables for expressions between the register assignments: 11379 11380@smallexample 11381int t1 = @dots{}; 11382register int *p1 asm ("r0") = @dots{}; 11383register int *p2 asm ("r1") = t1; 11384register int *result asm ("r0"); 11385asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2)); 11386@end smallexample 11387 11388Defining a register variable does not reserve the register. Other than 11389when invoking the Extended @code{asm}, the contents of the specified 11390register are not guaranteed. For this reason, the following uses 11391are explicitly @emph{not} supported. If they appear to work, it is only 11392happenstance, and may stop working as intended due to (seemingly) 11393unrelated changes in surrounding code, or even minor changes in the 11394optimization of a future version of gcc: 11395 11396@itemize @bullet 11397@item Passing parameters to or from Basic @code{asm} 11398@item Passing parameters to or from Extended @code{asm} without using input 11399or output operands. 11400@item Passing parameters to or from routines written in assembler (or 11401other languages) using non-standard calling conventions. 11402@end itemize 11403 11404Some developers use Local Register Variables in an attempt to improve 11405gcc's allocation of registers, especially in large functions. In this 11406case the register name is essentially a hint to the register allocator. 11407While in some instances this can generate better code, improvements are 11408subject to the whims of the allocator/optimizers. Since there are no 11409guarantees that your improvements won't be lost, this usage of Local 11410Register Variables is discouraged. 11411 11412On the MIPS platform, there is related use for local register variables 11413with slightly different characteristics (@pxref{MIPS Coprocessors,, 11414Defining coprocessor specifics for MIPS targets, gccint, 11415GNU Compiler Collection (GCC) Internals}). 11416 11417@node Size of an asm 11418@subsection Size of an @code{asm} 11419 11420Some targets require that GCC track the size of each instruction used 11421in order to generate correct code. Because the final length of the 11422code produced by an @code{asm} statement is only known by the 11423assembler, GCC must make an estimate as to how big it will be. It 11424does this by counting the number of instructions in the pattern of the 11425@code{asm} and multiplying that by the length of the longest 11426instruction supported by that processor. (When working out the number 11427of instructions, it assumes that any occurrence of a newline or of 11428whatever statement separator character is supported by the assembler --- 11429typically @samp{;} --- indicates the end of an instruction.) 11430 11431Normally, GCC's estimate is adequate to ensure that correct 11432code is generated, but it is possible to confuse the compiler if you use 11433pseudo instructions or assembler macros that expand into multiple real 11434instructions, or if you use assembler directives that expand to more 11435space in the object file than is needed for a single instruction. 11436If this happens then the assembler may produce a diagnostic saying that 11437a label is unreachable. 11438 11439@cindex @code{asm inline} 11440This size is also used for inlining decisions. If you use @code{asm inline} 11441instead of just @code{asm}, then for inlining purposes the size of the asm 11442is taken as the minimum size, ignoring how many instructions GCC thinks it is. 11443 11444@node Alternate Keywords 11445@section Alternate Keywords 11446@cindex alternate keywords 11447@cindex keywords, alternate 11448 11449@option{-ansi} and the various @option{-std} options disable certain 11450keywords. This causes trouble when you want to use GNU C extensions, or 11451a general-purpose header file that should be usable by all programs, 11452including ISO C programs. The keywords @code{asm}, @code{typeof} and 11453@code{inline} are not available in programs compiled with 11454@option{-ansi} or @option{-std} (although @code{inline} can be used in a 11455program compiled with @option{-std=c99} or a later standard). The 11456ISO C99 keyword 11457@code{restrict} is only available when @option{-std=gnu99} (which will 11458eventually be the default) or @option{-std=c99} (or the equivalent 11459@option{-std=iso9899:1999}), or an option for a later standard 11460version, is used. 11461 11462The way to solve these problems is to put @samp{__} at the beginning and 11463end of each problematical keyword. For example, use @code{__asm__} 11464instead of @code{asm}, and @code{__inline__} instead of @code{inline}. 11465 11466Other C compilers won't accept these alternative keywords; if you want to 11467compile with another compiler, you can define the alternate keywords as 11468macros to replace them with the customary keywords. It looks like this: 11469 11470@smallexample 11471#ifndef __GNUC__ 11472#define __asm__ asm 11473#endif 11474@end smallexample 11475 11476@findex __extension__ 11477@opindex pedantic 11478@option{-pedantic} and other options cause warnings for many GNU C extensions. 11479You can 11480prevent such warnings within one expression by writing 11481@code{__extension__} before the expression. @code{__extension__} has no 11482effect aside from this. 11483 11484@node Incomplete Enums 11485@section Incomplete @code{enum} Types 11486 11487You can define an @code{enum} tag without specifying its possible values. 11488This results in an incomplete type, much like what you get if you write 11489@code{struct foo} without describing the elements. A later declaration 11490that does specify the possible values completes the type. 11491 11492You cannot allocate variables or storage using the type while it is 11493incomplete. However, you can work with pointers to that type. 11494 11495This extension may not be very useful, but it makes the handling of 11496@code{enum} more consistent with the way @code{struct} and @code{union} 11497are handled. 11498 11499This extension is not supported by GNU C++. 11500 11501@node Function Names 11502@section Function Names as Strings 11503@cindex @code{__func__} identifier 11504@cindex @code{__FUNCTION__} identifier 11505@cindex @code{__PRETTY_FUNCTION__} identifier 11506 11507GCC provides three magic constants that hold the name of the current 11508function as a string. In C++11 and later modes, all three are treated 11509as constant expressions and can be used in @code{constexpr} constexts. 11510The first of these constants is @code{__func__}, which is part of 11511the C99 standard: 11512 11513The identifier @code{__func__} is implicitly declared by the translator 11514as if, immediately following the opening brace of each function 11515definition, the declaration 11516 11517@smallexample 11518static const char __func__[] = "function-name"; 11519@end smallexample 11520 11521@noindent 11522appeared, where function-name is the name of the lexically-enclosing 11523function. This name is the unadorned name of the function. As an 11524extension, at file (or, in C++, namespace scope), @code{__func__} 11525evaluates to the empty string. 11526 11527@code{__FUNCTION__} is another name for @code{__func__}, provided for 11528backward compatibility with old versions of GCC. 11529 11530In C, @code{__PRETTY_FUNCTION__} is yet another name for 11531@code{__func__}, except that at file scope (or, in C++, namespace scope), 11532it evaluates to the string @code{"top level"}. In addition, in C++, 11533@code{__PRETTY_FUNCTION__} contains the signature of the function as 11534well as its bare name. For example, this program: 11535 11536@smallexample 11537extern "C" int printf (const char *, ...); 11538 11539class a @{ 11540 public: 11541 void sub (int i) 11542 @{ 11543 printf ("__FUNCTION__ = %s\n", __FUNCTION__); 11544 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); 11545 @} 11546@}; 11547 11548int 11549main (void) 11550@{ 11551 a ax; 11552 ax.sub (0); 11553 return 0; 11554@} 11555@end smallexample 11556 11557@noindent 11558gives this output: 11559 11560@smallexample 11561__FUNCTION__ = sub 11562__PRETTY_FUNCTION__ = void a::sub(int) 11563@end smallexample 11564 11565These identifiers are variables, not preprocessor macros, and may not 11566be used to initialize @code{char} arrays or be concatenated with string 11567literals. 11568 11569@node Return Address 11570@section Getting the Return or Frame Address of a Function 11571 11572These functions may be used to get information about the callers of a 11573function. 11574 11575@deftypefn {Built-in Function} {void *} __builtin_return_address (unsigned int @var{level}) 11576This function returns the return address of the current function, or of 11577one of its callers. The @var{level} argument is number of frames to 11578scan up the call stack. A value of @code{0} yields the return address 11579of the current function, a value of @code{1} yields the return address 11580of the caller of the current function, and so forth. When inlining 11581the expected behavior is that the function returns the address of 11582the function that is returned to. To work around this behavior use 11583the @code{noinline} function attribute. 11584 11585The @var{level} argument must be a constant integer. 11586 11587On some machines it may be impossible to determine the return address of 11588any function other than the current one; in such cases, or when the top 11589of the stack has been reached, this function returns an unspecified 11590value. In addition, @code{__builtin_frame_address} may be used 11591to determine if the top of the stack has been reached. 11592 11593Additional post-processing of the returned value may be needed, see 11594@code{__builtin_extract_return_addr}. 11595 11596The stored representation of the return address in memory may be different 11597from the address returned by @code{__builtin_return_address}. For example, 11598on AArch64 the stored address may be mangled with return address signing 11599whereas the address returned by @code{__builtin_return_address} is not. 11600 11601Calling this function with a nonzero argument can have unpredictable 11602effects, including crashing the calling program. As a result, calls 11603that are considered unsafe are diagnosed when the @option{-Wframe-address} 11604option is in effect. Such calls should only be made in debugging 11605situations. 11606 11607On targets where code addresses are representable as @code{void *}, 11608@smallexample 11609void *addr = __builtin_extract_return_addr (__builtin_return_address (0)); 11610@end smallexample 11611gives the code address where the current function would return. For example, 11612such an address may be used with @code{dladdr} or other interfaces that work 11613with code addresses. 11614@end deftypefn 11615 11616@deftypefn {Built-in Function} {void *} __builtin_extract_return_addr (void *@var{addr}) 11617The address as returned by @code{__builtin_return_address} may have to be fed 11618through this function to get the actual encoded address. For example, on the 1161931-bit S/390 platform the highest bit has to be masked out, or on SPARC 11620platforms an offset has to be added for the true next instruction to be 11621executed. 11622 11623If no fixup is needed, this function simply passes through @var{addr}. 11624@end deftypefn 11625 11626@deftypefn {Built-in Function} {void *} __builtin_frob_return_addr (void *@var{addr}) 11627This function does the reverse of @code{__builtin_extract_return_addr}. 11628@end deftypefn 11629 11630@deftypefn {Built-in Function} {void *} __builtin_frame_address (unsigned int @var{level}) 11631This function is similar to @code{__builtin_return_address}, but it 11632returns the address of the function frame rather than the return address 11633of the function. Calling @code{__builtin_frame_address} with a value of 11634@code{0} yields the frame address of the current function, a value of 11635@code{1} yields the frame address of the caller of the current function, 11636and so forth. 11637 11638The frame is the area on the stack that holds local variables and saved 11639registers. The frame address is normally the address of the first word 11640pushed on to the stack by the function. However, the exact definition 11641depends upon the processor and the calling convention. If the processor 11642has a dedicated frame pointer register, and the function has a frame, 11643then @code{__builtin_frame_address} returns the value of the frame 11644pointer register. 11645 11646On some machines it may be impossible to determine the frame address of 11647any function other than the current one; in such cases, or when the top 11648of the stack has been reached, this function returns @code{0} if 11649the first frame pointer is properly initialized by the startup code. 11650 11651Calling this function with a nonzero argument can have unpredictable 11652effects, including crashing the calling program. As a result, calls 11653that are considered unsafe are diagnosed when the @option{-Wframe-address} 11654option is in effect. Such calls should only be made in debugging 11655situations. 11656@end deftypefn 11657 11658@node Vector Extensions 11659@section Using Vector Instructions through Built-in Functions 11660 11661On some targets, the instruction set contains SIMD vector instructions which 11662operate on multiple values contained in one large register at the same time. 11663For example, on the x86 the MMX, 3DNow!@: and SSE extensions can be used 11664this way. 11665 11666The first step in using these extensions is to provide the necessary data 11667types. This should be done using an appropriate @code{typedef}: 11668 11669@smallexample 11670typedef int v4si __attribute__ ((vector_size (16))); 11671@end smallexample 11672 11673@noindent 11674The @code{int} type specifies the @dfn{base type}, while the attribute specifies 11675the vector size for the variable, measured in bytes. For example, the 11676declaration above causes the compiler to set the mode for the @code{v4si} 11677type to be 16 bytes wide and divided into @code{int} sized units. For 11678a 32-bit @code{int} this means a vector of 4 units of 4 bytes, and the 11679corresponding mode of @code{foo} is @acronym{V4SI}. 11680 11681The @code{vector_size} attribute is only applicable to integral and 11682floating scalars, although arrays, pointers, and function return values 11683are allowed in conjunction with this construct. Only sizes that are 11684positive power-of-two multiples of the base type size are currently allowed. 11685 11686All the basic integer types can be used as base types, both as signed 11687and as unsigned: @code{char}, @code{short}, @code{int}, @code{long}, 11688@code{long long}. In addition, @code{float} and @code{double} can be 11689used to build floating-point vector types. 11690 11691Specifying a combination that is not valid for the current architecture 11692causes GCC to synthesize the instructions using a narrower mode. 11693For example, if you specify a variable of type @code{V4SI} and your 11694architecture does not allow for this specific SIMD type, GCC 11695produces code that uses 4 @code{SIs}. 11696 11697The types defined in this manner can be used with a subset of normal C 11698operations. Currently, GCC allows using the following operators 11699on these types: @code{+, -, *, /, unary minus, ^, |, &, ~, %}@. 11700 11701The operations behave like C++ @code{valarrays}. Addition is defined as 11702the addition of the corresponding elements of the operands. For 11703example, in the code below, each of the 4 elements in @var{a} is 11704added to the corresponding 4 elements in @var{b} and the resulting 11705vector is stored in @var{c}. 11706 11707@smallexample 11708typedef int v4si __attribute__ ((vector_size (16))); 11709 11710v4si a, b, c; 11711 11712c = a + b; 11713@end smallexample 11714 11715Subtraction, multiplication, division, and the logical operations 11716operate in a similar manner. Likewise, the result of using the unary 11717minus or complement operators on a vector type is a vector whose 11718elements are the negative or complemented values of the corresponding 11719elements in the operand. 11720 11721It is possible to use shifting operators @code{<<}, @code{>>} on 11722integer-type vectors. The operation is defined as following: @code{@{a0, 11723a1, @dots{}, an@} >> @{b0, b1, @dots{}, bn@} == @{a0 >> b0, a1 >> b1, 11724@dots{}, an >> bn@}}@. Vector operands must have the same number of 11725elements. 11726 11727For convenience, it is allowed to use a binary vector operation 11728where one operand is a scalar. In that case the compiler transforms 11729the scalar operand into a vector where each element is the scalar from 11730the operation. The transformation happens only if the scalar could be 11731safely converted to the vector-element type. 11732Consider the following code. 11733 11734@smallexample 11735typedef int v4si __attribute__ ((vector_size (16))); 11736 11737v4si a, b, c; 11738long l; 11739 11740a = b + 1; /* a = b + @{1,1,1,1@}; */ 11741a = 2 * b; /* a = @{2,2,2,2@} * b; */ 11742 11743a = l + a; /* Error, cannot convert long to int. */ 11744@end smallexample 11745 11746Vectors can be subscripted as if the vector were an array with 11747the same number of elements and base type. Out of bound accesses 11748invoke undefined behavior at run time. Warnings for out of bound 11749accesses for vector subscription can be enabled with 11750@option{-Warray-bounds}. 11751 11752Vector comparison is supported with standard comparison 11753operators: @code{==, !=, <, <=, >, >=}. Comparison operands can be 11754vector expressions of integer-type or real-type. Comparison between 11755integer-type vectors and real-type vectors are not supported. The 11756result of the comparison is a vector of the same width and number of 11757elements as the comparison operands with a signed integral element 11758type. 11759 11760Vectors are compared element-wise producing 0 when comparison is false 11761and -1 (constant of the appropriate type where all bits are set) 11762otherwise. Consider the following example. 11763 11764@smallexample 11765typedef int v4si __attribute__ ((vector_size (16))); 11766 11767v4si a = @{1,2,3,4@}; 11768v4si b = @{3,2,1,4@}; 11769v4si c; 11770 11771c = a > b; /* The result would be @{0, 0,-1, 0@} */ 11772c = a == b; /* The result would be @{0,-1, 0,-1@} */ 11773@end smallexample 11774 11775In C++, the ternary operator @code{?:} is available. @code{a?b:c}, where 11776@code{b} and @code{c} are vectors of the same type and @code{a} is an 11777integer vector with the same number of elements of the same size as @code{b} 11778and @code{c}, computes all three arguments and creates a vector 11779@code{@{a[0]?b[0]:c[0], a[1]?b[1]:c[1], @dots{}@}}. Note that unlike in 11780OpenCL, @code{a} is thus interpreted as @code{a != 0} and not @code{a < 0}. 11781As in the case of binary operations, this syntax is also accepted when 11782one of @code{b} or @code{c} is a scalar that is then transformed into a 11783vector. If both @code{b} and @code{c} are scalars and the type of 11784@code{true?b:c} has the same size as the element type of @code{a}, then 11785@code{b} and @code{c} are converted to a vector type whose elements have 11786this type and with the same number of elements as @code{a}. 11787 11788In C++, the logic operators @code{!, &&, ||} are available for vectors. 11789@code{!v} is equivalent to @code{v == 0}, @code{a && b} is equivalent to 11790@code{a!=0 & b!=0} and @code{a || b} is equivalent to @code{a!=0 | b!=0}. 11791For mixed operations between a scalar @code{s} and a vector @code{v}, 11792@code{s && v} is equivalent to @code{s?v!=0:0} (the evaluation is 11793short-circuit) and @code{v && s} is equivalent to @code{v!=0 & (s?-1:0)}. 11794 11795@findex __builtin_shuffle 11796Vector shuffling is available using functions 11797@code{__builtin_shuffle (vec, mask)} and 11798@code{__builtin_shuffle (vec0, vec1, mask)}. 11799Both functions construct a permutation of elements from one or two 11800vectors and return a vector of the same type as the input vector(s). 11801The @var{mask} is an integral vector with the same width (@var{W}) 11802and element count (@var{N}) as the output vector. 11803 11804The elements of the input vectors are numbered in memory ordering of 11805@var{vec0} beginning at 0 and @var{vec1} beginning at @var{N}. The 11806elements of @var{mask} are considered modulo @var{N} in the single-operand 11807case and modulo @math{2*@var{N}} in the two-operand case. 11808 11809Consider the following example, 11810 11811@smallexample 11812typedef int v4si __attribute__ ((vector_size (16))); 11813 11814v4si a = @{1,2,3,4@}; 11815v4si b = @{5,6,7,8@}; 11816v4si mask1 = @{0,1,1,3@}; 11817v4si mask2 = @{0,4,2,5@}; 11818v4si res; 11819 11820res = __builtin_shuffle (a, mask1); /* res is @{1,2,2,4@} */ 11821res = __builtin_shuffle (a, b, mask2); /* res is @{1,5,3,6@} */ 11822@end smallexample 11823 11824Note that @code{__builtin_shuffle} is intentionally semantically 11825compatible with the OpenCL @code{shuffle} and @code{shuffle2} functions. 11826 11827You can declare variables and use them in function calls and returns, as 11828well as in assignments and some casts. You can specify a vector type as 11829a return type for a function. Vector types can also be used as function 11830arguments. It is possible to cast from one vector type to another, 11831provided they are of the same size (in fact, you can also cast vectors 11832to and from other datatypes of the same size). 11833 11834You cannot operate between vectors of different lengths or different 11835signedness without a cast. 11836 11837@findex __builtin_convertvector 11838Vector conversion is available using the 11839@code{__builtin_convertvector (vec, vectype)} 11840function. @var{vec} must be an expression with integral or floating 11841vector type and @var{vectype} an integral or floating vector type with the 11842same number of elements. The result has @var{vectype} type and value of 11843a C cast of every element of @var{vec} to the element type of @var{vectype}. 11844 11845Consider the following example, 11846@smallexample 11847typedef int v4si __attribute__ ((vector_size (16))); 11848typedef float v4sf __attribute__ ((vector_size (16))); 11849typedef double v4df __attribute__ ((vector_size (32))); 11850typedef unsigned long long v4di __attribute__ ((vector_size (32))); 11851 11852v4si a = @{1,-2,3,-4@}; 11853v4sf b = @{1.5f,-2.5f,3.f,7.f@}; 11854v4di c = @{1ULL,5ULL,0ULL,10ULL@}; 11855v4sf d = __builtin_convertvector (a, v4sf); /* d is @{1.f,-2.f,3.f,-4.f@} */ 11856/* Equivalent of: 11857 v4sf d = @{ (float)a[0], (float)a[1], (float)a[2], (float)a[3] @}; */ 11858v4df e = __builtin_convertvector (a, v4df); /* e is @{1.,-2.,3.,-4.@} */ 11859v4df f = __builtin_convertvector (b, v4df); /* f is @{1.5,-2.5,3.,7.@} */ 11860v4si g = __builtin_convertvector (f, v4si); /* g is @{1,-2,3,7@} */ 11861v4si h = __builtin_convertvector (c, v4si); /* h is @{1,5,0,10@} */ 11862@end smallexample 11863 11864@cindex vector types, using with x86 intrinsics 11865Sometimes it is desirable to write code using a mix of generic vector 11866operations (for clarity) and machine-specific vector intrinsics (to 11867access vector instructions that are not exposed via generic built-ins). 11868On x86, intrinsic functions for integer vectors typically use the same 11869vector type @code{__m128i} irrespective of how they interpret the vector, 11870making it necessary to cast their arguments and return values from/to 11871other vector types. In C, you can make use of a @code{union} type: 11872@c In C++ such type punning via a union is not allowed by the language 11873@smallexample 11874#include <immintrin.h> 11875 11876typedef unsigned char u8x16 __attribute__ ((vector_size (16))); 11877typedef unsigned int u32x4 __attribute__ ((vector_size (16))); 11878 11879typedef union @{ 11880 __m128i mm; 11881 u8x16 u8; 11882 u32x4 u32; 11883@} v128; 11884@end smallexample 11885 11886@noindent 11887for variables that can be used with both built-in operators and x86 11888intrinsics: 11889 11890@smallexample 11891v128 x, y = @{ 0 @}; 11892memcpy (&x, ptr, sizeof x); 11893y.u8 += 0x80; 11894x.mm = _mm_adds_epu8 (x.mm, y.mm); 11895x.u32 &= 0xffffff; 11896 11897/* Instead of a variable, a compound literal may be used to pass the 11898 return value of an intrinsic call to a function expecting the union: */ 11899v128 foo (v128); 11900x = foo ((v128) @{_mm_adds_epu8 (x.mm, y.mm)@}); 11901@c This could be done implicitly with __attribute__((transparent_union)), 11902@c but GCC does not accept it for unions of vector types (PR 88955). 11903@end smallexample 11904 11905@node Offsetof 11906@section Support for @code{offsetof} 11907@findex __builtin_offsetof 11908 11909GCC implements for both C and C++ a syntactic extension to implement 11910the @code{offsetof} macro. 11911 11912@smallexample 11913primary: 11914 "__builtin_offsetof" "(" @code{typename} "," offsetof_member_designator ")" 11915 11916offsetof_member_designator: 11917 @code{identifier} 11918 | offsetof_member_designator "." @code{identifier} 11919 | offsetof_member_designator "[" @code{expr} "]" 11920@end smallexample 11921 11922This extension is sufficient such that 11923 11924@smallexample 11925#define offsetof(@var{type}, @var{member}) __builtin_offsetof (@var{type}, @var{member}) 11926@end smallexample 11927 11928@noindent 11929is a suitable definition of the @code{offsetof} macro. In C++, @var{type} 11930may be dependent. In either case, @var{member} may consist of a single 11931identifier, or a sequence of member accesses and array references. 11932 11933@node __sync Builtins 11934@section Legacy @code{__sync} Built-in Functions for Atomic Memory Access 11935 11936The following built-in functions 11937are intended to be compatible with those described 11938in the @cite{Intel Itanium Processor-specific Application Binary Interface}, 11939section 7.4. As such, they depart from normal GCC practice by not using 11940the @samp{__builtin_} prefix and also by being overloaded so that they 11941work on multiple types. 11942 11943The definition given in the Intel documentation allows only for the use of 11944the types @code{int}, @code{long}, @code{long long} or their unsigned 11945counterparts. GCC allows any scalar type that is 1, 2, 4 or 8 bytes in 11946size other than the C type @code{_Bool} or the C++ type @code{bool}. 11947Operations on pointer arguments are performed as if the operands were 11948of the @code{uintptr_t} type. That is, they are not scaled by the size 11949of the type to which the pointer points. 11950 11951These functions are implemented in terms of the @samp{__atomic} 11952builtins (@pxref{__atomic Builtins}). They should not be used for new 11953code which should use the @samp{__atomic} builtins instead. 11954 11955Not all operations are supported by all target processors. If a particular 11956operation cannot be implemented on the target processor, a warning is 11957generated and a call to an external function is generated. The external 11958function carries the same name as the built-in version, 11959with an additional suffix 11960@samp{_@var{n}} where @var{n} is the size of the data type. 11961 11962@c ??? Should we have a mechanism to suppress this warning? This is almost 11963@c useful for implementing the operation under the control of an external 11964@c mutex. 11965 11966In most cases, these built-in functions are considered a @dfn{full barrier}. 11967That is, 11968no memory operand is moved across the operation, either forward or 11969backward. Further, instructions are issued as necessary to prevent the 11970processor from speculating loads across the operation and from queuing stores 11971after the operation. 11972 11973All of the routines are described in the Intel documentation to take 11974``an optional list of variables protected by the memory barrier''. It's 11975not clear what is meant by that; it could mean that @emph{only} the 11976listed variables are protected, or it could mean a list of additional 11977variables to be protected. The list is ignored by GCC which treats it as 11978empty. GCC interprets an empty list as meaning that all globally 11979accessible variables should be protected. 11980 11981@table @code 11982@item @var{type} __sync_fetch_and_add (@var{type} *ptr, @var{type} value, ...) 11983@itemx @var{type} __sync_fetch_and_sub (@var{type} *ptr, @var{type} value, ...) 11984@itemx @var{type} __sync_fetch_and_or (@var{type} *ptr, @var{type} value, ...) 11985@itemx @var{type} __sync_fetch_and_and (@var{type} *ptr, @var{type} value, ...) 11986@itemx @var{type} __sync_fetch_and_xor (@var{type} *ptr, @var{type} value, ...) 11987@itemx @var{type} __sync_fetch_and_nand (@var{type} *ptr, @var{type} value, ...) 11988@findex __sync_fetch_and_add 11989@findex __sync_fetch_and_sub 11990@findex __sync_fetch_and_or 11991@findex __sync_fetch_and_and 11992@findex __sync_fetch_and_xor 11993@findex __sync_fetch_and_nand 11994These built-in functions perform the operation suggested by the name, and 11995returns the value that had previously been in memory. That is, operations 11996on integer operands have the following semantics. Operations on pointer 11997arguments are performed as if the operands were of the @code{uintptr_t} 11998type. That is, they are not scaled by the size of the type to which 11999the pointer points. 12000 12001@smallexample 12002@{ tmp = *ptr; *ptr @var{op}= value; return tmp; @} 12003@{ tmp = *ptr; *ptr = ~(tmp & value); return tmp; @} // nand 12004@end smallexample 12005 12006The object pointed to by the first argument must be of integer or pointer 12007type. It must not be a boolean type. 12008 12009@emph{Note:} GCC 4.4 and later implement @code{__sync_fetch_and_nand} 12010as @code{*ptr = ~(tmp & value)} instead of @code{*ptr = ~tmp & value}. 12011 12012@item @var{type} __sync_add_and_fetch (@var{type} *ptr, @var{type} value, ...) 12013@itemx @var{type} __sync_sub_and_fetch (@var{type} *ptr, @var{type} value, ...) 12014@itemx @var{type} __sync_or_and_fetch (@var{type} *ptr, @var{type} value, ...) 12015@itemx @var{type} __sync_and_and_fetch (@var{type} *ptr, @var{type} value, ...) 12016@itemx @var{type} __sync_xor_and_fetch (@var{type} *ptr, @var{type} value, ...) 12017@itemx @var{type} __sync_nand_and_fetch (@var{type} *ptr, @var{type} value, ...) 12018@findex __sync_add_and_fetch 12019@findex __sync_sub_and_fetch 12020@findex __sync_or_and_fetch 12021@findex __sync_and_and_fetch 12022@findex __sync_xor_and_fetch 12023@findex __sync_nand_and_fetch 12024These built-in functions perform the operation suggested by the name, and 12025return the new value. That is, operations on integer operands have 12026the following semantics. Operations on pointer operands are performed as 12027if the operand's type were @code{uintptr_t}. 12028 12029@smallexample 12030@{ *ptr @var{op}= value; return *ptr; @} 12031@{ *ptr = ~(*ptr & value); return *ptr; @} // nand 12032@end smallexample 12033 12034The same constraints on arguments apply as for the corresponding 12035@code{__sync_op_and_fetch} built-in functions. 12036 12037@emph{Note:} GCC 4.4 and later implement @code{__sync_nand_and_fetch} 12038as @code{*ptr = ~(*ptr & value)} instead of 12039@code{*ptr = ~*ptr & value}. 12040 12041@item bool __sync_bool_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...) 12042@itemx @var{type} __sync_val_compare_and_swap (@var{type} *ptr, @var{type} oldval, @var{type} newval, ...) 12043@findex __sync_bool_compare_and_swap 12044@findex __sync_val_compare_and_swap 12045These built-in functions perform an atomic compare and swap. 12046That is, if the current 12047value of @code{*@var{ptr}} is @var{oldval}, then write @var{newval} into 12048@code{*@var{ptr}}. 12049 12050The ``bool'' version returns @code{true} if the comparison is successful and 12051@var{newval} is written. The ``val'' version returns the contents 12052of @code{*@var{ptr}} before the operation. 12053 12054@item __sync_synchronize (...) 12055@findex __sync_synchronize 12056This built-in function issues a full memory barrier. 12057 12058@item @var{type} __sync_lock_test_and_set (@var{type} *ptr, @var{type} value, ...) 12059@findex __sync_lock_test_and_set 12060This built-in function, as described by Intel, is not a traditional test-and-set 12061operation, but rather an atomic exchange operation. It writes @var{value} 12062into @code{*@var{ptr}}, and returns the previous contents of 12063@code{*@var{ptr}}. 12064 12065Many targets have only minimal support for such locks, and do not support 12066a full exchange operation. In this case, a target may support reduced 12067functionality here by which the @emph{only} valid value to store is the 12068immediate constant 1. The exact value actually stored in @code{*@var{ptr}} 12069is implementation defined. 12070 12071This built-in function is not a full barrier, 12072but rather an @dfn{acquire barrier}. 12073This means that references after the operation cannot move to (or be 12074speculated to) before the operation, but previous memory stores may not 12075be globally visible yet, and previous memory loads may not yet be 12076satisfied. 12077 12078@item void __sync_lock_release (@var{type} *ptr, ...) 12079@findex __sync_lock_release 12080This built-in function releases the lock acquired by 12081@code{__sync_lock_test_and_set}. 12082Normally this means writing the constant 0 to @code{*@var{ptr}}. 12083 12084This built-in function is not a full barrier, 12085but rather a @dfn{release barrier}. 12086This means that all previous memory stores are globally visible, and all 12087previous memory loads have been satisfied, but following memory reads 12088are not prevented from being speculated to before the barrier. 12089@end table 12090 12091@node __atomic Builtins 12092@section Built-in Functions for Memory Model Aware Atomic Operations 12093 12094The following built-in functions approximately match the requirements 12095for the C++11 memory model. They are all 12096identified by being prefixed with @samp{__atomic} and most are 12097overloaded so that they work with multiple types. 12098 12099These functions are intended to replace the legacy @samp{__sync} 12100builtins. The main difference is that the memory order that is requested 12101is a parameter to the functions. New code should always use the 12102@samp{__atomic} builtins rather than the @samp{__sync} builtins. 12103 12104Note that the @samp{__atomic} builtins assume that programs will 12105conform to the C++11 memory model. In particular, they assume 12106that programs are free of data races. See the C++11 standard for 12107detailed requirements. 12108 12109The @samp{__atomic} builtins can be used with any integral scalar or 12110pointer type that is 1, 2, 4, or 8 bytes in length. 16-byte integral 12111types are also allowed if @samp{__int128} (@pxref{__int128}) is 12112supported by the architecture. 12113 12114The four non-arithmetic functions (load, store, exchange, and 12115compare_exchange) all have a generic version as well. This generic 12116version works on any data type. It uses the lock-free built-in function 12117if the specific data type size makes that possible; otherwise, an 12118external call is left to be resolved at run time. This external call is 12119the same format with the addition of a @samp{size_t} parameter inserted 12120as the first parameter indicating the size of the object being pointed to. 12121All objects must be the same size. 12122 12123There are 6 different memory orders that can be specified. These map 12124to the C++11 memory orders with the same names, see the C++11 standard 12125or the @uref{http://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync,GCC wiki 12126on atomic synchronization} for detailed definitions. Individual 12127targets may also support additional memory orders for use on specific 12128architectures. Refer to the target documentation for details of 12129these. 12130 12131An atomic operation can both constrain code motion and 12132be mapped to hardware instructions for synchronization between threads 12133(e.g., a fence). To which extent this happens is controlled by the 12134memory orders, which are listed here in approximately ascending order of 12135strength. The description of each memory order is only meant to roughly 12136illustrate the effects and is not a specification; see the C++11 12137memory model for precise semantics. 12138 12139@table @code 12140@item __ATOMIC_RELAXED 12141Implies no inter-thread ordering constraints. 12142@item __ATOMIC_CONSUME 12143This is currently implemented using the stronger @code{__ATOMIC_ACQUIRE} 12144memory order because of a deficiency in C++11's semantics for 12145@code{memory_order_consume}. 12146@item __ATOMIC_ACQUIRE 12147Creates an inter-thread happens-before constraint from the release (or 12148stronger) semantic store to this acquire load. Can prevent hoisting 12149of code to before the operation. 12150@item __ATOMIC_RELEASE 12151Creates an inter-thread happens-before constraint to acquire (or stronger) 12152semantic loads that read from this release store. Can prevent sinking 12153of code to after the operation. 12154@item __ATOMIC_ACQ_REL 12155Combines the effects of both @code{__ATOMIC_ACQUIRE} and 12156@code{__ATOMIC_RELEASE}. 12157@item __ATOMIC_SEQ_CST 12158Enforces total ordering with all other @code{__ATOMIC_SEQ_CST} operations. 12159@end table 12160 12161Note that in the C++11 memory model, @emph{fences} (e.g., 12162@samp{__atomic_thread_fence}) take effect in combination with other 12163atomic operations on specific memory locations (e.g., atomic loads); 12164operations on specific memory locations do not necessarily affect other 12165operations in the same way. 12166 12167Target architectures are encouraged to provide their own patterns for 12168each of the atomic built-in functions. If no target is provided, the original 12169non-memory model set of @samp{__sync} atomic built-in functions are 12170used, along with any required synchronization fences surrounding it in 12171order to achieve the proper behavior. Execution in this case is subject 12172to the same restrictions as those built-in functions. 12173 12174If there is no pattern or mechanism to provide a lock-free instruction 12175sequence, a call is made to an external routine with the same parameters 12176to be resolved at run time. 12177 12178When implementing patterns for these built-in functions, the memory order 12179parameter can be ignored as long as the pattern implements the most 12180restrictive @code{__ATOMIC_SEQ_CST} memory order. Any of the other memory 12181orders execute correctly with this memory order but they may not execute as 12182efficiently as they could with a more appropriate implementation of the 12183relaxed requirements. 12184 12185Note that the C++11 standard allows for the memory order parameter to be 12186determined at run time rather than at compile time. These built-in 12187functions map any run-time value to @code{__ATOMIC_SEQ_CST} rather 12188than invoke a runtime library call or inline a switch statement. This is 12189standard compliant, safe, and the simplest approach for now. 12190 12191The memory order parameter is a signed int, but only the lower 16 bits are 12192reserved for the memory order. The remainder of the signed int is reserved 12193for target use and should be 0. Use of the predefined atomic values 12194ensures proper usage. 12195 12196@deftypefn {Built-in Function} @var{type} __atomic_load_n (@var{type} *ptr, int memorder) 12197This built-in function implements an atomic load operation. It returns the 12198contents of @code{*@var{ptr}}. 12199 12200The valid memory order variants are 12201@code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE}, 12202and @code{__ATOMIC_CONSUME}. 12203 12204@end deftypefn 12205 12206@deftypefn {Built-in Function} void __atomic_load (@var{type} *ptr, @var{type} *ret, int memorder) 12207This is the generic version of an atomic load. It returns the 12208contents of @code{*@var{ptr}} in @code{*@var{ret}}. 12209 12210@end deftypefn 12211 12212@deftypefn {Built-in Function} void __atomic_store_n (@var{type} *ptr, @var{type} val, int memorder) 12213This built-in function implements an atomic store operation. It writes 12214@code{@var{val}} into @code{*@var{ptr}}. 12215 12216The valid memory order variants are 12217@code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and @code{__ATOMIC_RELEASE}. 12218 12219@end deftypefn 12220 12221@deftypefn {Built-in Function} void __atomic_store (@var{type} *ptr, @var{type} *val, int memorder) 12222This is the generic version of an atomic store. It stores the value 12223of @code{*@var{val}} into @code{*@var{ptr}}. 12224 12225@end deftypefn 12226 12227@deftypefn {Built-in Function} @var{type} __atomic_exchange_n (@var{type} *ptr, @var{type} val, int memorder) 12228This built-in function implements an atomic exchange operation. It writes 12229@var{val} into @code{*@var{ptr}}, and returns the previous contents of 12230@code{*@var{ptr}}. 12231 12232The valid memory order variants are 12233@code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, @code{__ATOMIC_ACQUIRE}, 12234@code{__ATOMIC_RELEASE}, and @code{__ATOMIC_ACQ_REL}. 12235 12236@end deftypefn 12237 12238@deftypefn {Built-in Function} void __atomic_exchange (@var{type} *ptr, @var{type} *val, @var{type} *ret, int memorder) 12239This is the generic version of an atomic exchange. It stores the 12240contents of @code{*@var{val}} into @code{*@var{ptr}}. The original value 12241of @code{*@var{ptr}} is copied into @code{*@var{ret}}. 12242 12243@end deftypefn 12244 12245@deftypefn {Built-in Function} bool __atomic_compare_exchange_n (@var{type} *ptr, @var{type} *expected, @var{type} desired, bool weak, int success_memorder, int failure_memorder) 12246This built-in function implements an atomic compare and exchange operation. 12247This compares the contents of @code{*@var{ptr}} with the contents of 12248@code{*@var{expected}}. If equal, the operation is a @emph{read-modify-write} 12249operation that writes @var{desired} into @code{*@var{ptr}}. If they are not 12250equal, the operation is a @emph{read} and the current contents of 12251@code{*@var{ptr}} are written into @code{*@var{expected}}. @var{weak} is @code{true} 12252for weak compare_exchange, which may fail spuriously, and @code{false} for 12253the strong variation, which never fails spuriously. Many targets 12254only offer the strong variation and ignore the parameter. When in doubt, use 12255the strong variation. 12256 12257If @var{desired} is written into @code{*@var{ptr}} then @code{true} is returned 12258and memory is affected according to the 12259memory order specified by @var{success_memorder}. There are no 12260restrictions on what memory order can be used here. 12261 12262Otherwise, @code{false} is returned and memory is affected according 12263to @var{failure_memorder}. This memory order cannot be 12264@code{__ATOMIC_RELEASE} nor @code{__ATOMIC_ACQ_REL}. It also cannot be a 12265stronger order than that specified by @var{success_memorder}. 12266 12267@end deftypefn 12268 12269@deftypefn {Built-in Function} bool __atomic_compare_exchange (@var{type} *ptr, @var{type} *expected, @var{type} *desired, bool weak, int success_memorder, int failure_memorder) 12270This built-in function implements the generic version of 12271@code{__atomic_compare_exchange}. The function is virtually identical to 12272@code{__atomic_compare_exchange_n}, except the desired value is also a 12273pointer. 12274 12275@end deftypefn 12276 12277@deftypefn {Built-in Function} @var{type} __atomic_add_fetch (@var{type} *ptr, @var{type} val, int memorder) 12278@deftypefnx {Built-in Function} @var{type} __atomic_sub_fetch (@var{type} *ptr, @var{type} val, int memorder) 12279@deftypefnx {Built-in Function} @var{type} __atomic_and_fetch (@var{type} *ptr, @var{type} val, int memorder) 12280@deftypefnx {Built-in Function} @var{type} __atomic_xor_fetch (@var{type} *ptr, @var{type} val, int memorder) 12281@deftypefnx {Built-in Function} @var{type} __atomic_or_fetch (@var{type} *ptr, @var{type} val, int memorder) 12282@deftypefnx {Built-in Function} @var{type} __atomic_nand_fetch (@var{type} *ptr, @var{type} val, int memorder) 12283These built-in functions perform the operation suggested by the name, and 12284return the result of the operation. Operations on pointer arguments are 12285performed as if the operands were of the @code{uintptr_t} type. That is, 12286they are not scaled by the size of the type to which the pointer points. 12287 12288@smallexample 12289@{ *ptr @var{op}= val; return *ptr; @} 12290@{ *ptr = ~(*ptr & val); return *ptr; @} // nand 12291@end smallexample 12292 12293The object pointed to by the first argument must be of integer or pointer 12294type. It must not be a boolean type. All memory orders are valid. 12295 12296@end deftypefn 12297 12298@deftypefn {Built-in Function} @var{type} __atomic_fetch_add (@var{type} *ptr, @var{type} val, int memorder) 12299@deftypefnx {Built-in Function} @var{type} __atomic_fetch_sub (@var{type} *ptr, @var{type} val, int memorder) 12300@deftypefnx {Built-in Function} @var{type} __atomic_fetch_and (@var{type} *ptr, @var{type} val, int memorder) 12301@deftypefnx {Built-in Function} @var{type} __atomic_fetch_xor (@var{type} *ptr, @var{type} val, int memorder) 12302@deftypefnx {Built-in Function} @var{type} __atomic_fetch_or (@var{type} *ptr, @var{type} val, int memorder) 12303@deftypefnx {Built-in Function} @var{type} __atomic_fetch_nand (@var{type} *ptr, @var{type} val, int memorder) 12304These built-in functions perform the operation suggested by the name, and 12305return the value that had previously been in @code{*@var{ptr}}. Operations 12306on pointer arguments are performed as if the operands were of 12307the @code{uintptr_t} type. That is, they are not scaled by the size of 12308the type to which the pointer points. 12309 12310@smallexample 12311@{ tmp = *ptr; *ptr @var{op}= val; return tmp; @} 12312@{ tmp = *ptr; *ptr = ~(*ptr & val); return tmp; @} // nand 12313@end smallexample 12314 12315The same constraints on arguments apply as for the corresponding 12316@code{__atomic_op_fetch} built-in functions. All memory orders are valid. 12317 12318@end deftypefn 12319 12320@deftypefn {Built-in Function} bool __atomic_test_and_set (void *ptr, int memorder) 12321 12322This built-in function performs an atomic test-and-set operation on 12323the byte at @code{*@var{ptr}}. The byte is set to some implementation 12324defined nonzero ``set'' value and the return value is @code{true} if and only 12325if the previous contents were ``set''. 12326It should be only used for operands of type @code{bool} or @code{char}. For 12327other types only part of the value may be set. 12328 12329All memory orders are valid. 12330 12331@end deftypefn 12332 12333@deftypefn {Built-in Function} void __atomic_clear (bool *ptr, int memorder) 12334 12335This built-in function performs an atomic clear operation on 12336@code{*@var{ptr}}. After the operation, @code{*@var{ptr}} contains 0. 12337It should be only used for operands of type @code{bool} or @code{char} and 12338in conjunction with @code{__atomic_test_and_set}. 12339For other types it may only clear partially. If the type is not @code{bool} 12340prefer using @code{__atomic_store}. 12341 12342The valid memory order variants are 12343@code{__ATOMIC_RELAXED}, @code{__ATOMIC_SEQ_CST}, and 12344@code{__ATOMIC_RELEASE}. 12345 12346@end deftypefn 12347 12348@deftypefn {Built-in Function} void __atomic_thread_fence (int memorder) 12349 12350This built-in function acts as a synchronization fence between threads 12351based on the specified memory order. 12352 12353All memory orders are valid. 12354 12355@end deftypefn 12356 12357@deftypefn {Built-in Function} void __atomic_signal_fence (int memorder) 12358 12359This built-in function acts as a synchronization fence between a thread 12360and signal handlers based in the same thread. 12361 12362All memory orders are valid. 12363 12364@end deftypefn 12365 12366@deftypefn {Built-in Function} bool __atomic_always_lock_free (size_t size, void *ptr) 12367 12368This built-in function returns @code{true} if objects of @var{size} bytes always 12369generate lock-free atomic instructions for the target architecture. 12370@var{size} must resolve to a compile-time constant and the result also 12371resolves to a compile-time constant. 12372 12373@var{ptr} is an optional pointer to the object that may be used to determine 12374alignment. A value of 0 indicates typical alignment should be used. The 12375compiler may also ignore this parameter. 12376 12377@smallexample 12378if (__atomic_always_lock_free (sizeof (long long), 0)) 12379@end smallexample 12380 12381@end deftypefn 12382 12383@deftypefn {Built-in Function} bool __atomic_is_lock_free (size_t size, void *ptr) 12384 12385This built-in function returns @code{true} if objects of @var{size} bytes always 12386generate lock-free atomic instructions for the target architecture. If 12387the built-in function is not known to be lock-free, a call is made to a 12388runtime routine named @code{__atomic_is_lock_free}. 12389 12390@var{ptr} is an optional pointer to the object that may be used to determine 12391alignment. A value of 0 indicates typical alignment should be used. The 12392compiler may also ignore this parameter. 12393@end deftypefn 12394 12395@node Integer Overflow Builtins 12396@section Built-in Functions to Perform Arithmetic with Overflow Checking 12397 12398The following built-in functions allow performing simple arithmetic operations 12399together with checking whether the operations overflowed. 12400 12401@deftypefn {Built-in Function} bool __builtin_add_overflow (@var{type1} a, @var{type2} b, @var{type3} *res) 12402@deftypefnx {Built-in Function} bool __builtin_sadd_overflow (int a, int b, int *res) 12403@deftypefnx {Built-in Function} bool __builtin_saddl_overflow (long int a, long int b, long int *res) 12404@deftypefnx {Built-in Function} bool __builtin_saddll_overflow (long long int a, long long int b, long long int *res) 12405@deftypefnx {Built-in Function} bool __builtin_uadd_overflow (unsigned int a, unsigned int b, unsigned int *res) 12406@deftypefnx {Built-in Function} bool __builtin_uaddl_overflow (unsigned long int a, unsigned long int b, unsigned long int *res) 12407@deftypefnx {Built-in Function} bool __builtin_uaddll_overflow (unsigned long long int a, unsigned long long int b, unsigned long long int *res) 12408 12409These built-in functions promote the first two operands into infinite precision signed 12410type and perform addition on those promoted operands. The result is then 12411cast to the type the third pointer argument points to and stored there. 12412If the stored result is equal to the infinite precision result, the built-in 12413functions return @code{false}, otherwise they return @code{true}. As the addition is 12414performed in infinite signed precision, these built-in functions have fully defined 12415behavior for all argument values. 12416 12417The first built-in function allows arbitrary integral types for operands and 12418the result type must be pointer to some integral type other than enumerated or 12419boolean type, the rest of the built-in functions have explicit integer types. 12420 12421The compiler will attempt to use hardware instructions to implement 12422these built-in functions where possible, like conditional jump on overflow 12423after addition, conditional jump on carry etc. 12424 12425@end deftypefn 12426 12427@deftypefn {Built-in Function} bool __builtin_sub_overflow (@var{type1} a, @var{type2} b, @var{type3} *res) 12428@deftypefnx {Built-in Function} bool __builtin_ssub_overflow (int a, int b, int *res) 12429@deftypefnx {Built-in Function} bool __builtin_ssubl_overflow (long int a, long int b, long int *res) 12430@deftypefnx {Built-in Function} bool __builtin_ssubll_overflow (long long int a, long long int b, long long int *res) 12431@deftypefnx {Built-in Function} bool __builtin_usub_overflow (unsigned int a, unsigned int b, unsigned int *res) 12432@deftypefnx {Built-in Function} bool __builtin_usubl_overflow (unsigned long int a, unsigned long int b, unsigned long int *res) 12433@deftypefnx {Built-in Function} bool __builtin_usubll_overflow (unsigned long long int a, unsigned long long int b, unsigned long long int *res) 12434 12435These built-in functions are similar to the add overflow checking built-in 12436functions above, except they perform subtraction, subtract the second argument 12437from the first one, instead of addition. 12438 12439@end deftypefn 12440 12441@deftypefn {Built-in Function} bool __builtin_mul_overflow (@var{type1} a, @var{type2} b, @var{type3} *res) 12442@deftypefnx {Built-in Function} bool __builtin_smul_overflow (int a, int b, int *res) 12443@deftypefnx {Built-in Function} bool __builtin_smull_overflow (long int a, long int b, long int *res) 12444@deftypefnx {Built-in Function} bool __builtin_smulll_overflow (long long int a, long long int b, long long int *res) 12445@deftypefnx {Built-in Function} bool __builtin_umul_overflow (unsigned int a, unsigned int b, unsigned int *res) 12446@deftypefnx {Built-in Function} bool __builtin_umull_overflow (unsigned long int a, unsigned long int b, unsigned long int *res) 12447@deftypefnx {Built-in Function} bool __builtin_umulll_overflow (unsigned long long int a, unsigned long long int b, unsigned long long int *res) 12448 12449These built-in functions are similar to the add overflow checking built-in 12450functions above, except they perform multiplication, instead of addition. 12451 12452@end deftypefn 12453 12454The following built-in functions allow checking if simple arithmetic operation 12455would overflow. 12456 12457@deftypefn {Built-in Function} bool __builtin_add_overflow_p (@var{type1} a, @var{type2} b, @var{type3} c) 12458@deftypefnx {Built-in Function} bool __builtin_sub_overflow_p (@var{type1} a, @var{type2} b, @var{type3} c) 12459@deftypefnx {Built-in Function} bool __builtin_mul_overflow_p (@var{type1} a, @var{type2} b, @var{type3} c) 12460 12461These built-in functions are similar to @code{__builtin_add_overflow}, 12462@code{__builtin_sub_overflow}, or @code{__builtin_mul_overflow}, except that 12463they don't store the result of the arithmetic operation anywhere and the 12464last argument is not a pointer, but some expression with integral type other 12465than enumerated or boolean type. 12466 12467The built-in functions promote the first two operands into infinite precision signed type 12468and perform addition on those promoted operands. The result is then 12469cast to the type of the third argument. If the cast result is equal to the infinite 12470precision result, the built-in functions return @code{false}, otherwise they return @code{true}. 12471The value of the third argument is ignored, just the side effects in the third argument 12472are evaluated, and no integral argument promotions are performed on the last argument. 12473If the third argument is a bit-field, the type used for the result cast has the 12474precision and signedness of the given bit-field, rather than precision and signedness 12475of the underlying type. 12476 12477For example, the following macro can be used to portably check, at 12478compile-time, whether or not adding two constant integers will overflow, 12479and perform the addition only when it is known to be safe and not to trigger 12480a @option{-Woverflow} warning. 12481 12482@smallexample 12483#define INT_ADD_OVERFLOW_P(a, b) \ 12484 __builtin_add_overflow_p (a, b, (__typeof__ ((a) + (b))) 0) 12485 12486enum @{ 12487 A = INT_MAX, B = 3, 12488 C = INT_ADD_OVERFLOW_P (A, B) ? 0 : A + B, 12489 D = __builtin_add_overflow_p (1, SCHAR_MAX, (signed char) 0) 12490@}; 12491@end smallexample 12492 12493The compiler will attempt to use hardware instructions to implement 12494these built-in functions where possible, like conditional jump on overflow 12495after addition, conditional jump on carry etc. 12496 12497@end deftypefn 12498 12499@node x86 specific memory model extensions for transactional memory 12500@section x86-Specific Memory Model Extensions for Transactional Memory 12501 12502The x86 architecture supports additional memory ordering flags 12503to mark critical sections for hardware lock elision. 12504These must be specified in addition to an existing memory order to 12505atomic intrinsics. 12506 12507@table @code 12508@item __ATOMIC_HLE_ACQUIRE 12509Start lock elision on a lock variable. 12510Memory order must be @code{__ATOMIC_ACQUIRE} or stronger. 12511@item __ATOMIC_HLE_RELEASE 12512End lock elision on a lock variable. 12513Memory order must be @code{__ATOMIC_RELEASE} or stronger. 12514@end table 12515 12516When a lock acquire fails, it is required for good performance to abort 12517the transaction quickly. This can be done with a @code{_mm_pause}. 12518 12519@smallexample 12520#include <immintrin.h> // For _mm_pause 12521 12522int lockvar; 12523 12524/* Acquire lock with lock elision */ 12525while (__atomic_exchange_n(&lockvar, 1, __ATOMIC_ACQUIRE|__ATOMIC_HLE_ACQUIRE)) 12526 _mm_pause(); /* Abort failed transaction */ 12527... 12528/* Free lock with lock elision */ 12529__atomic_store_n(&lockvar, 0, __ATOMIC_RELEASE|__ATOMIC_HLE_RELEASE); 12530@end smallexample 12531 12532@node Object Size Checking 12533@section Object Size Checking Built-in Functions 12534@findex __builtin_object_size 12535@findex __builtin___memcpy_chk 12536@findex __builtin___mempcpy_chk 12537@findex __builtin___memmove_chk 12538@findex __builtin___memset_chk 12539@findex __builtin___strcpy_chk 12540@findex __builtin___stpcpy_chk 12541@findex __builtin___strncpy_chk 12542@findex __builtin___strcat_chk 12543@findex __builtin___strncat_chk 12544@findex __builtin___sprintf_chk 12545@findex __builtin___snprintf_chk 12546@findex __builtin___vsprintf_chk 12547@findex __builtin___vsnprintf_chk 12548@findex __builtin___printf_chk 12549@findex __builtin___vprintf_chk 12550@findex __builtin___fprintf_chk 12551@findex __builtin___vfprintf_chk 12552 12553GCC implements a limited buffer overflow protection mechanism that can 12554prevent some buffer overflow attacks by determining the sizes of objects 12555into which data is about to be written and preventing the writes when 12556the size isn't sufficient. The built-in functions described below yield 12557the best results when used together and when optimization is enabled. 12558For example, to detect object sizes across function boundaries or to 12559follow pointer assignments through non-trivial control flow they rely 12560on various optimization passes enabled with @option{-O2}. However, to 12561a limited extent, they can be used without optimization as well. 12562 12563@deftypefn {Built-in Function} {size_t} __builtin_object_size (const void * @var{ptr}, int @var{type}) 12564is a built-in construct that returns a constant number of bytes from 12565@var{ptr} to the end of the object @var{ptr} pointer points to 12566(if known at compile time). To determine the sizes of dynamically allocated 12567objects the function relies on the allocation functions called to obtain 12568the storage to be declared with the @code{alloc_size} attribute (@pxref{Common 12569Function Attributes}). @code{__builtin_object_size} never evaluates 12570its arguments for side effects. If there are any side effects in them, it 12571returns @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0} 12572for @var{type} 2 or 3. If there are multiple objects @var{ptr} can 12573point to and all of them are known at compile time, the returned number 12574is the maximum of remaining byte counts in those objects if @var{type} & 2 is 125750 and minimum if nonzero. If it is not possible to determine which objects 12576@var{ptr} points to at compile time, @code{__builtin_object_size} should 12577return @code{(size_t) -1} for @var{type} 0 or 1 and @code{(size_t) 0} 12578for @var{type} 2 or 3. 12579 12580@var{type} is an integer constant from 0 to 3. If the least significant 12581bit is clear, objects are whole variables, if it is set, a closest 12582surrounding subobject is considered the object a pointer points to. 12583The second bit determines if maximum or minimum of remaining bytes 12584is computed. 12585 12586@smallexample 12587struct V @{ char buf1[10]; int b; char buf2[10]; @} var; 12588char *p = &var.buf1[1], *q = &var.b; 12589 12590/* Here the object p points to is var. */ 12591assert (__builtin_object_size (p, 0) == sizeof (var) - 1); 12592/* The subobject p points to is var.buf1. */ 12593assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1); 12594/* The object q points to is var. */ 12595assert (__builtin_object_size (q, 0) 12596 == (char *) (&var + 1) - (char *) &var.b); 12597/* The subobject q points to is var.b. */ 12598assert (__builtin_object_size (q, 1) == sizeof (var.b)); 12599@end smallexample 12600@end deftypefn 12601 12602There are built-in functions added for many common string operation 12603functions, e.g., for @code{memcpy} @code{__builtin___memcpy_chk} 12604built-in is provided. This built-in has an additional last argument, 12605which is the number of bytes remaining in the object the @var{dest} 12606argument points to or @code{(size_t) -1} if the size is not known. 12607 12608The built-in functions are optimized into the normal string functions 12609like @code{memcpy} if the last argument is @code{(size_t) -1} or if 12610it is known at compile time that the destination object will not 12611be overflowed. If the compiler can determine at compile time that the 12612object will always be overflowed, it issues a warning. 12613 12614The intended use can be e.g.@: 12615 12616@smallexample 12617#undef memcpy 12618#define bos0(dest) __builtin_object_size (dest, 0) 12619#define memcpy(dest, src, n) \ 12620 __builtin___memcpy_chk (dest, src, n, bos0 (dest)) 12621 12622char *volatile p; 12623char buf[10]; 12624/* It is unknown what object p points to, so this is optimized 12625 into plain memcpy - no checking is possible. */ 12626memcpy (p, "abcde", n); 12627/* Destination is known and length too. It is known at compile 12628 time there will be no overflow. */ 12629memcpy (&buf[5], "abcde", 5); 12630/* Destination is known, but the length is not known at compile time. 12631 This will result in __memcpy_chk call that can check for overflow 12632 at run time. */ 12633memcpy (&buf[5], "abcde", n); 12634/* Destination is known and it is known at compile time there will 12635 be overflow. There will be a warning and __memcpy_chk call that 12636 will abort the program at run time. */ 12637memcpy (&buf[6], "abcde", 5); 12638@end smallexample 12639 12640Such built-in functions are provided for @code{memcpy}, @code{mempcpy}, 12641@code{memmove}, @code{memset}, @code{strcpy}, @code{stpcpy}, @code{strncpy}, 12642@code{strcat} and @code{strncat}. 12643 12644There are also checking built-in functions for formatted output functions. 12645@smallexample 12646int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...); 12647int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os, 12648 const char *fmt, ...); 12649int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt, 12650 va_list ap); 12651int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os, 12652 const char *fmt, va_list ap); 12653@end smallexample 12654 12655The added @var{flag} argument is passed unchanged to @code{__sprintf_chk} 12656etc.@: functions and can contain implementation specific flags on what 12657additional security measures the checking function might take, such as 12658handling @code{%n} differently. 12659 12660The @var{os} argument is the object size @var{s} points to, like in the 12661other built-in functions. There is a small difference in the behavior 12662though, if @var{os} is @code{(size_t) -1}, the built-in functions are 12663optimized into the non-checking functions only if @var{flag} is 0, otherwise 12664the checking function is called with @var{os} argument set to 12665@code{(size_t) -1}. 12666 12667In addition to this, there are checking built-in functions 12668@code{__builtin___printf_chk}, @code{__builtin___vprintf_chk}, 12669@code{__builtin___fprintf_chk} and @code{__builtin___vfprintf_chk}. 12670These have just one additional argument, @var{flag}, right before 12671format string @var{fmt}. If the compiler is able to optimize them to 12672@code{fputc} etc.@: functions, it does, otherwise the checking function 12673is called and the @var{flag} argument passed to it. 12674 12675@node Other Builtins 12676@section Other Built-in Functions Provided by GCC 12677@cindex built-in functions 12678@findex __builtin_alloca 12679@findex __builtin_alloca_with_align 12680@findex __builtin_alloca_with_align_and_max 12681@findex __builtin_call_with_static_chain 12682@findex __builtin_extend_pointer 12683@findex __builtin_fpclassify 12684@findex __builtin_has_attribute 12685@findex __builtin_isfinite 12686@findex __builtin_isnormal 12687@findex __builtin_isgreater 12688@findex __builtin_isgreaterequal 12689@findex __builtin_isinf_sign 12690@findex __builtin_isless 12691@findex __builtin_islessequal 12692@findex __builtin_islessgreater 12693@findex __builtin_isunordered 12694@findex __builtin_object_size 12695@findex __builtin_powi 12696@findex __builtin_powif 12697@findex __builtin_powil 12698@findex __builtin_speculation_safe_value 12699@findex _Exit 12700@findex _exit 12701@findex abort 12702@findex abs 12703@findex acos 12704@findex acosf 12705@findex acosh 12706@findex acoshf 12707@findex acoshl 12708@findex acosl 12709@findex alloca 12710@findex asin 12711@findex asinf 12712@findex asinh 12713@findex asinhf 12714@findex asinhl 12715@findex asinl 12716@findex atan 12717@findex atan2 12718@findex atan2f 12719@findex atan2l 12720@findex atanf 12721@findex atanh 12722@findex atanhf 12723@findex atanhl 12724@findex atanl 12725@findex bcmp 12726@findex bzero 12727@findex cabs 12728@findex cabsf 12729@findex cabsl 12730@findex cacos 12731@findex cacosf 12732@findex cacosh 12733@findex cacoshf 12734@findex cacoshl 12735@findex cacosl 12736@findex calloc 12737@findex carg 12738@findex cargf 12739@findex cargl 12740@findex casin 12741@findex casinf 12742@findex casinh 12743@findex casinhf 12744@findex casinhl 12745@findex casinl 12746@findex catan 12747@findex catanf 12748@findex catanh 12749@findex catanhf 12750@findex catanhl 12751@findex catanl 12752@findex cbrt 12753@findex cbrtf 12754@findex cbrtl 12755@findex ccos 12756@findex ccosf 12757@findex ccosh 12758@findex ccoshf 12759@findex ccoshl 12760@findex ccosl 12761@findex ceil 12762@findex ceilf 12763@findex ceill 12764@findex cexp 12765@findex cexpf 12766@findex cexpl 12767@findex cimag 12768@findex cimagf 12769@findex cimagl 12770@findex clog 12771@findex clogf 12772@findex clogl 12773@findex clog10 12774@findex clog10f 12775@findex clog10l 12776@findex conj 12777@findex conjf 12778@findex conjl 12779@findex copysign 12780@findex copysignf 12781@findex copysignl 12782@findex cos 12783@findex cosf 12784@findex cosh 12785@findex coshf 12786@findex coshl 12787@findex cosl 12788@findex cpow 12789@findex cpowf 12790@findex cpowl 12791@findex cproj 12792@findex cprojf 12793@findex cprojl 12794@findex creal 12795@findex crealf 12796@findex creall 12797@findex csin 12798@findex csinf 12799@findex csinh 12800@findex csinhf 12801@findex csinhl 12802@findex csinl 12803@findex csqrt 12804@findex csqrtf 12805@findex csqrtl 12806@findex ctan 12807@findex ctanf 12808@findex ctanh 12809@findex ctanhf 12810@findex ctanhl 12811@findex ctanl 12812@findex dcgettext 12813@findex dgettext 12814@findex drem 12815@findex dremf 12816@findex dreml 12817@findex erf 12818@findex erfc 12819@findex erfcf 12820@findex erfcl 12821@findex erff 12822@findex erfl 12823@findex exit 12824@findex exp 12825@findex exp10 12826@findex exp10f 12827@findex exp10l 12828@findex exp2 12829@findex exp2f 12830@findex exp2l 12831@findex expf 12832@findex expl 12833@findex expm1 12834@findex expm1f 12835@findex expm1l 12836@findex fabs 12837@findex fabsf 12838@findex fabsl 12839@findex fdim 12840@findex fdimf 12841@findex fdiml 12842@findex ffs 12843@findex floor 12844@findex floorf 12845@findex floorl 12846@findex fma 12847@findex fmaf 12848@findex fmal 12849@findex fmax 12850@findex fmaxf 12851@findex fmaxl 12852@findex fmin 12853@findex fminf 12854@findex fminl 12855@findex fmod 12856@findex fmodf 12857@findex fmodl 12858@findex fprintf 12859@findex fprintf_unlocked 12860@findex fputs 12861@findex fputs_unlocked 12862@findex free 12863@findex frexp 12864@findex frexpf 12865@findex frexpl 12866@findex fscanf 12867@findex gamma 12868@findex gammaf 12869@findex gammal 12870@findex gamma_r 12871@findex gammaf_r 12872@findex gammal_r 12873@findex gettext 12874@findex hypot 12875@findex hypotf 12876@findex hypotl 12877@findex ilogb 12878@findex ilogbf 12879@findex ilogbl 12880@findex imaxabs 12881@findex index 12882@findex isalnum 12883@findex isalpha 12884@findex isascii 12885@findex isblank 12886@findex iscntrl 12887@findex isdigit 12888@findex isgraph 12889@findex islower 12890@findex isprint 12891@findex ispunct 12892@findex isspace 12893@findex isupper 12894@findex iswalnum 12895@findex iswalpha 12896@findex iswblank 12897@findex iswcntrl 12898@findex iswdigit 12899@findex iswgraph 12900@findex iswlower 12901@findex iswprint 12902@findex iswpunct 12903@findex iswspace 12904@findex iswupper 12905@findex iswxdigit 12906@findex isxdigit 12907@findex j0 12908@findex j0f 12909@findex j0l 12910@findex j1 12911@findex j1f 12912@findex j1l 12913@findex jn 12914@findex jnf 12915@findex jnl 12916@findex labs 12917@findex ldexp 12918@findex ldexpf 12919@findex ldexpl 12920@findex lgamma 12921@findex lgammaf 12922@findex lgammal 12923@findex lgamma_r 12924@findex lgammaf_r 12925@findex lgammal_r 12926@findex llabs 12927@findex llrint 12928@findex llrintf 12929@findex llrintl 12930@findex llround 12931@findex llroundf 12932@findex llroundl 12933@findex log 12934@findex log10 12935@findex log10f 12936@findex log10l 12937@findex log1p 12938@findex log1pf 12939@findex log1pl 12940@findex log2 12941@findex log2f 12942@findex log2l 12943@findex logb 12944@findex logbf 12945@findex logbl 12946@findex logf 12947@findex logl 12948@findex lrint 12949@findex lrintf 12950@findex lrintl 12951@findex lround 12952@findex lroundf 12953@findex lroundl 12954@findex malloc 12955@findex memchr 12956@findex memcmp 12957@findex memcpy 12958@findex mempcpy 12959@findex memset 12960@findex modf 12961@findex modff 12962@findex modfl 12963@findex nearbyint 12964@findex nearbyintf 12965@findex nearbyintl 12966@findex nextafter 12967@findex nextafterf 12968@findex nextafterl 12969@findex nexttoward 12970@findex nexttowardf 12971@findex nexttowardl 12972@findex pow 12973@findex pow10 12974@findex pow10f 12975@findex pow10l 12976@findex powf 12977@findex powl 12978@findex printf 12979@findex printf_unlocked 12980@findex putchar 12981@findex puts 12982@findex realloc 12983@findex remainder 12984@findex remainderf 12985@findex remainderl 12986@findex remquo 12987@findex remquof 12988@findex remquol 12989@findex rindex 12990@findex rint 12991@findex rintf 12992@findex rintl 12993@findex round 12994@findex roundf 12995@findex roundl 12996@findex scalb 12997@findex scalbf 12998@findex scalbl 12999@findex scalbln 13000@findex scalblnf 13001@findex scalblnf 13002@findex scalbn 13003@findex scalbnf 13004@findex scanfnl 13005@findex signbit 13006@findex signbitf 13007@findex signbitl 13008@findex signbitd32 13009@findex signbitd64 13010@findex signbitd128 13011@findex significand 13012@findex significandf 13013@findex significandl 13014@findex sin 13015@findex sincos 13016@findex sincosf 13017@findex sincosl 13018@findex sinf 13019@findex sinh 13020@findex sinhf 13021@findex sinhl 13022@findex sinl 13023@findex snprintf 13024@findex sprintf 13025@findex sqrt 13026@findex sqrtf 13027@findex sqrtl 13028@findex sscanf 13029@findex stpcpy 13030@findex stpncpy 13031@findex strcasecmp 13032@findex strcat 13033@findex strchr 13034@findex strcmp 13035@findex strcpy 13036@findex strcspn 13037@findex strdup 13038@findex strfmon 13039@findex strftime 13040@findex strlen 13041@findex strncasecmp 13042@findex strncat 13043@findex strncmp 13044@findex strncpy 13045@findex strndup 13046@findex strnlen 13047@findex strpbrk 13048@findex strrchr 13049@findex strspn 13050@findex strstr 13051@findex tan 13052@findex tanf 13053@findex tanh 13054@findex tanhf 13055@findex tanhl 13056@findex tanl 13057@findex tgamma 13058@findex tgammaf 13059@findex tgammal 13060@findex toascii 13061@findex tolower 13062@findex toupper 13063@findex towlower 13064@findex towupper 13065@findex trunc 13066@findex truncf 13067@findex truncl 13068@findex vfprintf 13069@findex vfscanf 13070@findex vprintf 13071@findex vscanf 13072@findex vsnprintf 13073@findex vsprintf 13074@findex vsscanf 13075@findex y0 13076@findex y0f 13077@findex y0l 13078@findex y1 13079@findex y1f 13080@findex y1l 13081@findex yn 13082@findex ynf 13083@findex ynl 13084 13085GCC provides a large number of built-in functions other than the ones 13086mentioned above. Some of these are for internal use in the processing 13087of exceptions or variable-length argument lists and are not 13088documented here because they may change from time to time; we do not 13089recommend general use of these functions. 13090 13091The remaining functions are provided for optimization purposes. 13092 13093With the exception of built-ins that have library equivalents such as 13094the standard C library functions discussed below, or that expand to 13095library calls, GCC built-in functions are always expanded inline and 13096thus do not have corresponding entry points and their address cannot 13097be obtained. Attempting to use them in an expression other than 13098a function call results in a compile-time error. 13099 13100@opindex fno-builtin 13101GCC includes built-in versions of many of the functions in the standard 13102C library. These functions come in two forms: one whose names start with 13103the @code{__builtin_} prefix, and the other without. Both forms have the 13104same type (including prototype), the same address (when their address is 13105taken), and the same meaning as the C library functions even if you specify 13106the @option{-fno-builtin} option @pxref{C Dialect Options}). Many of these 13107functions are only optimized in certain cases; if they are not optimized in 13108a particular case, a call to the library function is emitted. 13109 13110@opindex ansi 13111@opindex std 13112Outside strict ISO C mode (@option{-ansi}, @option{-std=c90}, 13113@option{-std=c99} or @option{-std=c11}), the functions 13114@code{_exit}, @code{alloca}, @code{bcmp}, @code{bzero}, 13115@code{dcgettext}, @code{dgettext}, @code{dremf}, @code{dreml}, 13116@code{drem}, @code{exp10f}, @code{exp10l}, @code{exp10}, @code{ffsll}, 13117@code{ffsl}, @code{ffs}, @code{fprintf_unlocked}, 13118@code{fputs_unlocked}, @code{gammaf}, @code{gammal}, @code{gamma}, 13119@code{gammaf_r}, @code{gammal_r}, @code{gamma_r}, @code{gettext}, 13120@code{index}, @code{isascii}, @code{j0f}, @code{j0l}, @code{j0}, 13121@code{j1f}, @code{j1l}, @code{j1}, @code{jnf}, @code{jnl}, @code{jn}, 13122@code{lgammaf_r}, @code{lgammal_r}, @code{lgamma_r}, @code{mempcpy}, 13123@code{pow10f}, @code{pow10l}, @code{pow10}, @code{printf_unlocked}, 13124@code{rindex}, @code{roundeven}, @code{roundevenf}, @code{roundevenl}, 13125@code{scalbf}, @code{scalbl}, @code{scalb}, 13126@code{signbit}, @code{signbitf}, @code{signbitl}, @code{signbitd32}, 13127@code{signbitd64}, @code{signbitd128}, @code{significandf}, 13128@code{significandl}, @code{significand}, @code{sincosf}, 13129@code{sincosl}, @code{sincos}, @code{stpcpy}, @code{stpncpy}, 13130@code{strcasecmp}, @code{strdup}, @code{strfmon}, @code{strncasecmp}, 13131@code{strndup}, @code{strnlen}, @code{toascii}, @code{y0f}, @code{y0l}, 13132@code{y0}, @code{y1f}, @code{y1l}, @code{y1}, @code{ynf}, @code{ynl} and 13133@code{yn} 13134may be handled as built-in functions. 13135All these functions have corresponding versions 13136prefixed with @code{__builtin_}, which may be used even in strict C90 13137mode. 13138 13139The ISO C99 functions 13140@code{_Exit}, @code{acoshf}, @code{acoshl}, @code{acosh}, @code{asinhf}, 13141@code{asinhl}, @code{asinh}, @code{atanhf}, @code{atanhl}, @code{atanh}, 13142@code{cabsf}, @code{cabsl}, @code{cabs}, @code{cacosf}, @code{cacoshf}, 13143@code{cacoshl}, @code{cacosh}, @code{cacosl}, @code{cacos}, 13144@code{cargf}, @code{cargl}, @code{carg}, @code{casinf}, @code{casinhf}, 13145@code{casinhl}, @code{casinh}, @code{casinl}, @code{casin}, 13146@code{catanf}, @code{catanhf}, @code{catanhl}, @code{catanh}, 13147@code{catanl}, @code{catan}, @code{cbrtf}, @code{cbrtl}, @code{cbrt}, 13148@code{ccosf}, @code{ccoshf}, @code{ccoshl}, @code{ccosh}, @code{ccosl}, 13149@code{ccos}, @code{cexpf}, @code{cexpl}, @code{cexp}, @code{cimagf}, 13150@code{cimagl}, @code{cimag}, @code{clogf}, @code{clogl}, @code{clog}, 13151@code{conjf}, @code{conjl}, @code{conj}, @code{copysignf}, @code{copysignl}, 13152@code{copysign}, @code{cpowf}, @code{cpowl}, @code{cpow}, @code{cprojf}, 13153@code{cprojl}, @code{cproj}, @code{crealf}, @code{creall}, @code{creal}, 13154@code{csinf}, @code{csinhf}, @code{csinhl}, @code{csinh}, @code{csinl}, 13155@code{csin}, @code{csqrtf}, @code{csqrtl}, @code{csqrt}, @code{ctanf}, 13156@code{ctanhf}, @code{ctanhl}, @code{ctanh}, @code{ctanl}, @code{ctan}, 13157@code{erfcf}, @code{erfcl}, @code{erfc}, @code{erff}, @code{erfl}, 13158@code{erf}, @code{exp2f}, @code{exp2l}, @code{exp2}, @code{expm1f}, 13159@code{expm1l}, @code{expm1}, @code{fdimf}, @code{fdiml}, @code{fdim}, 13160@code{fmaf}, @code{fmal}, @code{fmaxf}, @code{fmaxl}, @code{fmax}, 13161@code{fma}, @code{fminf}, @code{fminl}, @code{fmin}, @code{hypotf}, 13162@code{hypotl}, @code{hypot}, @code{ilogbf}, @code{ilogbl}, @code{ilogb}, 13163@code{imaxabs}, @code{isblank}, @code{iswblank}, @code{lgammaf}, 13164@code{lgammal}, @code{lgamma}, @code{llabs}, @code{llrintf}, @code{llrintl}, 13165@code{llrint}, @code{llroundf}, @code{llroundl}, @code{llround}, 13166@code{log1pf}, @code{log1pl}, @code{log1p}, @code{log2f}, @code{log2l}, 13167@code{log2}, @code{logbf}, @code{logbl}, @code{logb}, @code{lrintf}, 13168@code{lrintl}, @code{lrint}, @code{lroundf}, @code{lroundl}, 13169@code{lround}, @code{nearbyintf}, @code{nearbyintl}, @code{nearbyint}, 13170@code{nextafterf}, @code{nextafterl}, @code{nextafter}, 13171@code{nexttowardf}, @code{nexttowardl}, @code{nexttoward}, 13172@code{remainderf}, @code{remainderl}, @code{remainder}, @code{remquof}, 13173@code{remquol}, @code{remquo}, @code{rintf}, @code{rintl}, @code{rint}, 13174@code{roundf}, @code{roundl}, @code{round}, @code{scalblnf}, 13175@code{scalblnl}, @code{scalbln}, @code{scalbnf}, @code{scalbnl}, 13176@code{scalbn}, @code{snprintf}, @code{tgammaf}, @code{tgammal}, 13177@code{tgamma}, @code{truncf}, @code{truncl}, @code{trunc}, 13178@code{vfscanf}, @code{vscanf}, @code{vsnprintf} and @code{vsscanf} 13179are handled as built-in functions 13180except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}). 13181 13182There are also built-in versions of the ISO C99 functions 13183@code{acosf}, @code{acosl}, @code{asinf}, @code{asinl}, @code{atan2f}, 13184@code{atan2l}, @code{atanf}, @code{atanl}, @code{ceilf}, @code{ceill}, 13185@code{cosf}, @code{coshf}, @code{coshl}, @code{cosl}, @code{expf}, 13186@code{expl}, @code{fabsf}, @code{fabsl}, @code{floorf}, @code{floorl}, 13187@code{fmodf}, @code{fmodl}, @code{frexpf}, @code{frexpl}, @code{ldexpf}, 13188@code{ldexpl}, @code{log10f}, @code{log10l}, @code{logf}, @code{logl}, 13189@code{modfl}, @code{modf}, @code{powf}, @code{powl}, @code{sinf}, 13190@code{sinhf}, @code{sinhl}, @code{sinl}, @code{sqrtf}, @code{sqrtl}, 13191@code{tanf}, @code{tanhf}, @code{tanhl} and @code{tanl} 13192that are recognized in any mode since ISO C90 reserves these names for 13193the purpose to which ISO C99 puts them. All these functions have 13194corresponding versions prefixed with @code{__builtin_}. 13195 13196There are also built-in functions @code{__builtin_fabsf@var{n}}, 13197@code{__builtin_fabsf@var{n}x}, @code{__builtin_copysignf@var{n}} and 13198@code{__builtin_copysignf@var{n}x}, corresponding to the TS 18661-3 13199functions @code{fabsf@var{n}}, @code{fabsf@var{n}x}, 13200@code{copysignf@var{n}} and @code{copysignf@var{n}x}, for supported 13201types @code{_Float@var{n}} and @code{_Float@var{n}x}. 13202 13203There are also GNU extension functions @code{clog10}, @code{clog10f} and 13204@code{clog10l} which names are reserved by ISO C99 for future use. 13205All these functions have versions prefixed with @code{__builtin_}. 13206 13207The ISO C94 functions 13208@code{iswalnum}, @code{iswalpha}, @code{iswcntrl}, @code{iswdigit}, 13209@code{iswgraph}, @code{iswlower}, @code{iswprint}, @code{iswpunct}, 13210@code{iswspace}, @code{iswupper}, @code{iswxdigit}, @code{towlower} and 13211@code{towupper} 13212are handled as built-in functions 13213except in strict ISO C90 mode (@option{-ansi} or @option{-std=c90}). 13214 13215The ISO C90 functions 13216@code{abort}, @code{abs}, @code{acos}, @code{asin}, @code{atan2}, 13217@code{atan}, @code{calloc}, @code{ceil}, @code{cosh}, @code{cos}, 13218@code{exit}, @code{exp}, @code{fabs}, @code{floor}, @code{fmod}, 13219@code{fprintf}, @code{fputs}, @code{free}, @code{frexp}, @code{fscanf}, 13220@code{isalnum}, @code{isalpha}, @code{iscntrl}, @code{isdigit}, 13221@code{isgraph}, @code{islower}, @code{isprint}, @code{ispunct}, 13222@code{isspace}, @code{isupper}, @code{isxdigit}, @code{tolower}, 13223@code{toupper}, @code{labs}, @code{ldexp}, @code{log10}, @code{log}, 13224@code{malloc}, @code{memchr}, @code{memcmp}, @code{memcpy}, 13225@code{memset}, @code{modf}, @code{pow}, @code{printf}, @code{putchar}, 13226@code{puts}, @code{realloc}, @code{scanf}, @code{sinh}, @code{sin}, 13227@code{snprintf}, @code{sprintf}, @code{sqrt}, @code{sscanf}, @code{strcat}, 13228@code{strchr}, @code{strcmp}, @code{strcpy}, @code{strcspn}, 13229@code{strlen}, @code{strncat}, @code{strncmp}, @code{strncpy}, 13230@code{strpbrk}, @code{strrchr}, @code{strspn}, @code{strstr}, 13231@code{tanh}, @code{tan}, @code{vfprintf}, @code{vprintf} and @code{vsprintf} 13232are all recognized as built-in functions unless 13233@option{-fno-builtin} is specified (or @option{-fno-builtin-@var{function}} 13234is specified for an individual function). All of these functions have 13235corresponding versions prefixed with @code{__builtin_}. 13236 13237GCC provides built-in versions of the ISO C99 floating-point comparison 13238macros that avoid raising exceptions for unordered operands. They have 13239the same names as the standard macros ( @code{isgreater}, 13240@code{isgreaterequal}, @code{isless}, @code{islessequal}, 13241@code{islessgreater}, and @code{isunordered}) , with @code{__builtin_} 13242prefixed. We intend for a library implementor to be able to simply 13243@code{#define} each standard macro to its built-in equivalent. 13244In the same fashion, GCC provides @code{fpclassify}, @code{isfinite}, 13245@code{isinf_sign}, @code{isnormal} and @code{signbit} built-ins used with 13246@code{__builtin_} prefixed. The @code{isinf} and @code{isnan} 13247built-in functions appear both with and without the @code{__builtin_} prefix. 13248 13249@deftypefn {Built-in Function} void *__builtin_alloca (size_t size) 13250The @code{__builtin_alloca} function must be called at block scope. 13251The function allocates an object @var{size} bytes large on the stack 13252of the calling function. The object is aligned on the default stack 13253alignment boundary for the target determined by the 13254@code{__BIGGEST_ALIGNMENT__} macro. The @code{__builtin_alloca} 13255function returns a pointer to the first byte of the allocated object. 13256The lifetime of the allocated object ends just before the calling 13257function returns to its caller. This is so even when 13258@code{__builtin_alloca} is called within a nested block. 13259 13260For example, the following function allocates eight objects of @code{n} 13261bytes each on the stack, storing a pointer to each in consecutive elements 13262of the array @code{a}. It then passes the array to function @code{g} 13263which can safely use the storage pointed to by each of the array elements. 13264 13265@smallexample 13266void f (unsigned n) 13267@{ 13268 void *a [8]; 13269 for (int i = 0; i != 8; ++i) 13270 a [i] = __builtin_alloca (n); 13271 13272 g (a, n); // @r{safe} 13273@} 13274@end smallexample 13275 13276Since the @code{__builtin_alloca} function doesn't validate its argument 13277it is the responsibility of its caller to make sure the argument doesn't 13278cause it to exceed the stack size limit. 13279The @code{__builtin_alloca} function is provided to make it possible to 13280allocate on the stack arrays of bytes with an upper bound that may be 13281computed at run time. Since C99 Variable Length Arrays offer 13282similar functionality under a portable, more convenient, and safer 13283interface they are recommended instead, in both C99 and C++ programs 13284where GCC provides them as an extension. 13285@xref{Variable Length}, for details. 13286 13287@end deftypefn 13288 13289@deftypefn {Built-in Function} void *__builtin_alloca_with_align (size_t size, size_t alignment) 13290The @code{__builtin_alloca_with_align} function must be called at block 13291scope. The function allocates an object @var{size} bytes large on 13292the stack of the calling function. The allocated object is aligned on 13293the boundary specified by the argument @var{alignment} whose unit is given 13294in bits (not bytes). The @var{size} argument must be positive and not 13295exceed the stack size limit. The @var{alignment} argument must be a constant 13296integer expression that evaluates to a power of 2 greater than or equal to 13297@code{CHAR_BIT} and less than some unspecified maximum. Invocations 13298with other values are rejected with an error indicating the valid bounds. 13299The function returns a pointer to the first byte of the allocated object. 13300The lifetime of the allocated object ends at the end of the block in which 13301the function was called. The allocated storage is released no later than 13302just before the calling function returns to its caller, but may be released 13303at the end of the block in which the function was called. 13304 13305For example, in the following function the call to @code{g} is unsafe 13306because when @code{overalign} is non-zero, the space allocated by 13307@code{__builtin_alloca_with_align} may have been released at the end 13308of the @code{if} statement in which it was called. 13309 13310@smallexample 13311void f (unsigned n, bool overalign) 13312@{ 13313 void *p; 13314 if (overalign) 13315 p = __builtin_alloca_with_align (n, 64 /* bits */); 13316 else 13317 p = __builtin_alloc (n); 13318 13319 g (p, n); // @r{unsafe} 13320@} 13321@end smallexample 13322 13323Since the @code{__builtin_alloca_with_align} function doesn't validate its 13324@var{size} argument it is the responsibility of its caller to make sure 13325the argument doesn't cause it to exceed the stack size limit. 13326The @code{__builtin_alloca_with_align} function is provided to make 13327it possible to allocate on the stack overaligned arrays of bytes with 13328an upper bound that may be computed at run time. Since C99 13329Variable Length Arrays offer the same functionality under 13330a portable, more convenient, and safer interface they are recommended 13331instead, in both C99 and C++ programs where GCC provides them as 13332an extension. @xref{Variable Length}, for details. 13333 13334@end deftypefn 13335 13336@deftypefn {Built-in Function} void *__builtin_alloca_with_align_and_max (size_t size, size_t alignment, size_t max_size) 13337Similar to @code{__builtin_alloca_with_align} but takes an extra argument 13338specifying an upper bound for @var{size} in case its value cannot be computed 13339at compile time, for use by @option{-fstack-usage}, @option{-Wstack-usage} 13340and @option{-Walloca-larger-than}. @var{max_size} must be a constant integer 13341expression, it has no effect on code generation and no attempt is made to 13342check its compatibility with @var{size}. 13343 13344@end deftypefn 13345 13346@deftypefn {Built-in Function} bool __builtin_has_attribute (@var{type-or-expression}, @var{attribute}) 13347The @code{__builtin_has_attribute} function evaluates to an integer constant 13348expression equal to @code{true} if the symbol or type referenced by 13349the @var{type-or-expression} argument has been declared with 13350the @var{attribute} referenced by the second argument. For 13351an @var{type-or-expression} argument that does not reference a symbol, 13352since attributes do not apply to expressions the built-in consider 13353the type of the argument. Neither argument is evaluated. 13354The @var{type-or-expression} argument is subject to the same 13355restrictions as the argument to @code{typeof} (@pxref{Typeof}). The 13356@var{attribute} argument is an attribute name optionally followed by 13357a comma-separated list of arguments enclosed in parentheses. Both forms 13358of attribute names---with and without double leading and trailing 13359underscores---are recognized. @xref{Attribute Syntax}, for details. 13360When no attribute arguments are specified for an attribute that expects 13361one or more arguments the function returns @code{true} if 13362@var{type-or-expression} has been declared with the attribute regardless 13363of the attribute argument values. Arguments provided for an attribute 13364that expects some are validated and matched up to the provided number. 13365The function returns @code{true} if all provided arguments match. For 13366example, the first call to the function below evaluates to @code{true} 13367because @code{x} is declared with the @code{aligned} attribute but 13368the second call evaluates to @code{false} because @code{x} is declared 13369@code{aligned (8)} and not @code{aligned (4)}. 13370 13371@smallexample 13372__attribute__ ((aligned (8))) int x; 13373_Static_assert (__builtin_has_attribute (x, aligned), "aligned"); 13374_Static_assert (!__builtin_has_attribute (x, aligned (4)), "aligned (4)"); 13375@end smallexample 13376 13377Due to a limitation the @code{__builtin_has_attribute} function returns 13378@code{false} for the @code{mode} attribute even if the type or variable 13379referenced by the @var{type-or-expression} argument was declared with one. 13380The function is also not supported with labels, and in C with enumerators. 13381 13382Note that unlike the @code{__has_attribute} preprocessor operator which 13383is suitable for use in @code{#if} preprocessing directives 13384@code{__builtin_has_attribute} is an intrinsic function that is not 13385recognized in such contexts. 13386 13387@end deftypefn 13388 13389@deftypefn {Built-in Function} @var{type} __builtin_speculation_safe_value (@var{type} val, @var{type} failval) 13390 13391This built-in function can be used to help mitigate against unsafe 13392speculative execution. @var{type} may be any integral type or any 13393pointer type. 13394 13395@enumerate 13396@item 13397If the CPU is not speculatively executing the code, then @var{val} 13398is returned. 13399@item 13400If the CPU is executing speculatively then either: 13401@itemize 13402@item 13403The function may cause execution to pause until it is known that the 13404code is no-longer being executed speculatively (in which case 13405@var{val} can be returned, as above); or 13406@item 13407The function may use target-dependent speculation tracking state to cause 13408@var{failval} to be returned when it is known that speculative 13409execution has incorrectly predicted a conditional branch operation. 13410@end itemize 13411@end enumerate 13412 13413The second argument, @var{failval}, is optional and defaults to zero 13414if omitted. 13415 13416GCC defines the preprocessor macro 13417@code{__HAVE_BUILTIN_SPECULATION_SAFE_VALUE} for targets that have been 13418updated to support this builtin. 13419 13420The built-in function can be used where a variable appears to be used in a 13421safe way, but the CPU, due to speculative execution may temporarily ignore 13422the bounds checks. Consider, for example, the following function: 13423 13424@smallexample 13425int array[500]; 13426int f (unsigned untrusted_index) 13427@{ 13428 if (untrusted_index < 500) 13429 return array[untrusted_index]; 13430 return 0; 13431@} 13432@end smallexample 13433 13434If the function is called repeatedly with @code{untrusted_index} less 13435than the limit of 500, then a branch predictor will learn that the 13436block of code that returns a value stored in @code{array} will be 13437executed. If the function is subsequently called with an 13438out-of-range value it will still try to execute that block of code 13439first until the CPU determines that the prediction was incorrect 13440(the CPU will unwind any incorrect operations at that point). 13441However, depending on how the result of the function is used, it might be 13442possible to leave traces in the cache that can reveal what was stored 13443at the out-of-bounds location. The built-in function can be used to 13444provide some protection against leaking data in this way by changing 13445the code to: 13446 13447@smallexample 13448int array[500]; 13449int f (unsigned untrusted_index) 13450@{ 13451 if (untrusted_index < 500) 13452 return array[__builtin_speculation_safe_value (untrusted_index)]; 13453 return 0; 13454@} 13455@end smallexample 13456 13457The built-in function will either cause execution to stall until the 13458conditional branch has been fully resolved, or it may permit 13459speculative execution to continue, but using 0 instead of 13460@code{untrusted_value} if that exceeds the limit. 13461 13462If accessing any memory location is potentially unsafe when speculative 13463execution is incorrect, then the code can be rewritten as 13464 13465@smallexample 13466int array[500]; 13467int f (unsigned untrusted_index) 13468@{ 13469 if (untrusted_index < 500) 13470 return *__builtin_speculation_safe_value (&array[untrusted_index], NULL); 13471 return 0; 13472@} 13473@end smallexample 13474 13475which will cause a @code{NULL} pointer to be used for the unsafe case. 13476 13477@end deftypefn 13478 13479@deftypefn {Built-in Function} int __builtin_types_compatible_p (@var{type1}, @var{type2}) 13480 13481You can use the built-in function @code{__builtin_types_compatible_p} to 13482determine whether two types are the same. 13483 13484This built-in function returns 1 if the unqualified versions of the 13485types @var{type1} and @var{type2} (which are types, not expressions) are 13486compatible, 0 otherwise. The result of this built-in function can be 13487used in integer constant expressions. 13488 13489This built-in function ignores top level qualifiers (e.g., @code{const}, 13490@code{volatile}). For example, @code{int} is equivalent to @code{const 13491int}. 13492 13493The type @code{int[]} and @code{int[5]} are compatible. On the other 13494hand, @code{int} and @code{char *} are not compatible, even if the size 13495of their types, on the particular architecture are the same. Also, the 13496amount of pointer indirection is taken into account when determining 13497similarity. Consequently, @code{short *} is not similar to 13498@code{short **}. Furthermore, two types that are typedefed are 13499considered compatible if their underlying types are compatible. 13500 13501An @code{enum} type is not considered to be compatible with another 13502@code{enum} type even if both are compatible with the same integer 13503type; this is what the C standard specifies. 13504For example, @code{enum @{foo, bar@}} is not similar to 13505@code{enum @{hot, dog@}}. 13506 13507You typically use this function in code whose execution varies 13508depending on the arguments' types. For example: 13509 13510@smallexample 13511#define foo(x) \ 13512 (@{ \ 13513 typeof (x) tmp = (x); \ 13514 if (__builtin_types_compatible_p (typeof (x), long double)) \ 13515 tmp = foo_long_double (tmp); \ 13516 else if (__builtin_types_compatible_p (typeof (x), double)) \ 13517 tmp = foo_double (tmp); \ 13518 else if (__builtin_types_compatible_p (typeof (x), float)) \ 13519 tmp = foo_float (tmp); \ 13520 else \ 13521 abort (); \ 13522 tmp; \ 13523 @}) 13524@end smallexample 13525 13526@emph{Note:} This construct is only available for C@. 13527 13528@end deftypefn 13529 13530@deftypefn {Built-in Function} @var{type} __builtin_call_with_static_chain (@var{call_exp}, @var{pointer_exp}) 13531 13532The @var{call_exp} expression must be a function call, and the 13533@var{pointer_exp} expression must be a pointer. The @var{pointer_exp} 13534is passed to the function call in the target's static chain location. 13535The result of builtin is the result of the function call. 13536 13537@emph{Note:} This builtin is only available for C@. 13538This builtin can be used to call Go closures from C. 13539 13540@end deftypefn 13541 13542@deftypefn {Built-in Function} @var{type} __builtin_choose_expr (@var{const_exp}, @var{exp1}, @var{exp2}) 13543 13544You can use the built-in function @code{__builtin_choose_expr} to 13545evaluate code depending on the value of a constant expression. This 13546built-in function returns @var{exp1} if @var{const_exp}, which is an 13547integer constant expression, is nonzero. Otherwise it returns @var{exp2}. 13548 13549This built-in function is analogous to the @samp{? :} operator in C, 13550except that the expression returned has its type unaltered by promotion 13551rules. Also, the built-in function does not evaluate the expression 13552that is not chosen. For example, if @var{const_exp} evaluates to @code{true}, 13553@var{exp2} is not evaluated even if it has side effects. 13554 13555This built-in function can return an lvalue if the chosen argument is an 13556lvalue. 13557 13558If @var{exp1} is returned, the return type is the same as @var{exp1}'s 13559type. Similarly, if @var{exp2} is returned, its return type is the same 13560as @var{exp2}. 13561 13562Example: 13563 13564@smallexample 13565#define foo(x) \ 13566 __builtin_choose_expr ( \ 13567 __builtin_types_compatible_p (typeof (x), double), \ 13568 foo_double (x), \ 13569 __builtin_choose_expr ( \ 13570 __builtin_types_compatible_p (typeof (x), float), \ 13571 foo_float (x), \ 13572 /* @r{The void expression results in a compile-time error} \ 13573 @r{when assigning the result to something.} */ \ 13574 (void)0)) 13575@end smallexample 13576 13577@emph{Note:} This construct is only available for C@. Furthermore, the 13578unused expression (@var{exp1} or @var{exp2} depending on the value of 13579@var{const_exp}) may still generate syntax errors. This may change in 13580future revisions. 13581 13582@end deftypefn 13583 13584@deftypefn {Built-in Function} @var{type} __builtin_tgmath (@var{functions}, @var{arguments}) 13585 13586The built-in function @code{__builtin_tgmath}, available only for C 13587and Objective-C, calls a function determined according to the rules of 13588@code{<tgmath.h>} macros. It is intended to be used in 13589implementations of that header, so that expansions of macros from that 13590header only expand each of their arguments once, to avoid problems 13591when calls to such macros are nested inside the arguments of other 13592calls to such macros; in addition, it results in better diagnostics 13593for invalid calls to @code{<tgmath.h>} macros than implementations 13594using other GNU C language features. For example, the @code{pow} 13595type-generic macro might be defined as: 13596 13597@smallexample 13598#define pow(a, b) __builtin_tgmath (powf, pow, powl, \ 13599 cpowf, cpow, cpowl, a, b) 13600@end smallexample 13601 13602The arguments to @code{__builtin_tgmath} are at least two pointers to 13603functions, followed by the arguments to the type-generic macro (which 13604will be passed as arguments to the selected function). All the 13605pointers to functions must be pointers to prototyped functions, none 13606of which may have variable arguments, and all of which must have the 13607same number of parameters; the number of parameters of the first 13608function determines how many arguments to @code{__builtin_tgmath} are 13609interpreted as function pointers, and how many as the arguments to the 13610called function. 13611 13612The types of the specified functions must all be different, but 13613related to each other in the same way as a set of functions that may 13614be selected between by a macro in @code{<tgmath.h>}. This means that 13615the functions are parameterized by a floating-point type @var{t}, 13616different for each such function. The function return types may all 13617be the same type, or they may be @var{t} for each function, or they 13618may be the real type corresponding to @var{t} for each function (if 13619some of the types @var{t} are complex). Likewise, for each parameter 13620position, the type of the parameter in that position may always be the 13621same type, or may be @var{t} for each function (this case must apply 13622for at least one parameter position), or may be the real type 13623corresponding to @var{t} for each function. 13624 13625The standard rules for @code{<tgmath.h>} macros are used to find a 13626common type @var{u} from the types of the arguments for parameters 13627whose types vary between the functions; complex integer types (a GNU 13628extension) are treated like @code{_Complex double} for this purpose 13629(or @code{_Complex _Float64} if all the function return types are the 13630same @code{_Float@var{n}} or @code{_Float@var{n}x} type). 13631If the function return types vary, or are all the same integer type, 13632the function called is the one for which @var{t} is @var{u}, and it is 13633an error if there is no such function. If the function return types 13634are all the same floating-point type, the type-generic macro is taken 13635to be one of those from TS 18661 that rounds the result to a narrower 13636type; if there is a function for which @var{t} is @var{u}, it is 13637called, and otherwise the first function, if any, for which @var{t} 13638has at least the range and precision of @var{u} is called, and it is 13639an error if there is no such function. 13640 13641@end deftypefn 13642 13643@deftypefn {Built-in Function} @var{type} __builtin_complex (@var{real}, @var{imag}) 13644 13645The built-in function @code{__builtin_complex} is provided for use in 13646implementing the ISO C11 macros @code{CMPLXF}, @code{CMPLX} and 13647@code{CMPLXL}. @var{real} and @var{imag} must have the same type, a 13648real binary floating-point type, and the result has the corresponding 13649complex type with real and imaginary parts @var{real} and @var{imag}. 13650Unlike @samp{@var{real} + I * @var{imag}}, this works even when 13651infinities, NaNs and negative zeros are involved. 13652 13653@end deftypefn 13654 13655@deftypefn {Built-in Function} int __builtin_constant_p (@var{exp}) 13656You can use the built-in function @code{__builtin_constant_p} to 13657determine if a value is known to be constant at compile time and hence 13658that GCC can perform constant-folding on expressions involving that 13659value. The argument of the function is the value to test. The function 13660returns the integer 1 if the argument is known to be a compile-time 13661constant and 0 if it is not known to be a compile-time constant. A 13662return of 0 does not indicate that the value is @emph{not} a constant, 13663but merely that GCC cannot prove it is a constant with the specified 13664value of the @option{-O} option. 13665 13666You typically use this function in an embedded application where 13667memory is a critical resource. If you have some complex calculation, 13668you may want it to be folded if it involves constants, but need to call 13669a function if it does not. For example: 13670 13671@smallexample 13672#define Scale_Value(X) \ 13673 (__builtin_constant_p (X) \ 13674 ? ((X) * SCALE + OFFSET) : Scale (X)) 13675@end smallexample 13676 13677You may use this built-in function in either a macro or an inline 13678function. However, if you use it in an inlined function and pass an 13679argument of the function as the argument to the built-in, GCC 13680never returns 1 when you call the inline function with a string constant 13681or compound literal (@pxref{Compound Literals}) and does not return 1 13682when you pass a constant numeric value to the inline function unless you 13683specify the @option{-O} option. 13684 13685You may also use @code{__builtin_constant_p} in initializers for static 13686data. For instance, you can write 13687 13688@smallexample 13689static const int table[] = @{ 13690 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1, 13691 /* @r{@dots{}} */ 13692@}; 13693@end smallexample 13694 13695@noindent 13696This is an acceptable initializer even if @var{EXPRESSION} is not a 13697constant expression, including the case where 13698@code{__builtin_constant_p} returns 1 because @var{EXPRESSION} can be 13699folded to a constant but @var{EXPRESSION} contains operands that are 13700not otherwise permitted in a static initializer (for example, 13701@code{0 && foo ()}). GCC must be more conservative about evaluating the 13702built-in in this case, because it has no opportunity to perform 13703optimization. 13704@end deftypefn 13705 13706@deftypefn {Built-in Function} bool __builtin_is_constant_evaluated (void) 13707The @code{__builtin_is_constant_evaluated} function is available only 13708in C++. The built-in is intended to be used by implementations of 13709the @code{std::is_constant_evaluated} C++ function. Programs should make 13710use of the latter function rather than invoking the built-in directly. 13711 13712The main use case of the built-in is to determine whether a @code{constexpr} 13713function is being called in a @code{constexpr} context. A call to 13714the function evaluates to a core constant expression with the value 13715@code{true} if and only if it occurs within the evaluation of an expression 13716or conversion that is manifestly constant-evaluated as defined in the C++ 13717standard. Manifestly constant-evaluated contexts include constant-expressions, 13718the conditions of @code{constexpr if} statements, constraint-expressions, and 13719initializers of variables usable in constant expressions. For more details 13720refer to the latest revision of the C++ standard. 13721@end deftypefn 13722 13723@deftypefn {Built-in Function} void __builtin_clear_padding (@var{ptr}) 13724The built-in function @code{__builtin_clear_padding} function clears 13725padding bits inside of the object representation of object pointed by 13726@var{ptr}, which has to be a pointer. The value representation of the 13727object is not affected. The type of the object is assumed to be the type 13728the pointer points to. Inside of a union, the only cleared bits are 13729bits that are padding bits for all the union members. 13730 13731This built-in-function is useful if the padding bits of an object might 13732have intederminate values and the object representation needs to be 13733bitwise compared to some other object, for example for atomic operations. 13734@end deftypefn 13735 13736@deftypefn {Built-in Function} @var{type} __builtin_bit_cast (@var{type}, @var{arg}) 13737The @code{__builtin_bit_cast} function is available only 13738in C++. The built-in is intended to be used by implementations of 13739the @code{std::bit_cast} C++ template function. Programs should make 13740use of the latter function rather than invoking the built-in directly. 13741 13742This built-in function allows reinterpreting the bits of the @var{arg} 13743argument as if it had type @var{type}. @var{type} and the type of the 13744@var{arg} argument need to be trivially copyable types with the same size. 13745When manifestly constant-evaluated, it performs extra diagnostics required 13746for @code{std::bit_cast} and returns a constant expression if @var{arg} 13747is a constant expression. For more details 13748refer to the latest revision of the C++ standard. 13749@end deftypefn 13750 13751@deftypefn {Built-in Function} long __builtin_expect (long @var{exp}, long @var{c}) 13752@opindex fprofile-arcs 13753You may use @code{__builtin_expect} to provide the compiler with 13754branch prediction information. In general, you should prefer to 13755use actual profile feedback for this (@option{-fprofile-arcs}), as 13756programmers are notoriously bad at predicting how their programs 13757actually perform. However, there are applications in which this 13758data is hard to collect. 13759 13760The return value is the value of @var{exp}, which should be an integral 13761expression. The semantics of the built-in are that it is expected that 13762@var{exp} == @var{c}. For example: 13763 13764@smallexample 13765if (__builtin_expect (x, 0)) 13766 foo (); 13767@end smallexample 13768 13769@noindent 13770indicates that we do not expect to call @code{foo}, since 13771we expect @code{x} to be zero. Since you are limited to integral 13772expressions for @var{exp}, you should use constructions such as 13773 13774@smallexample 13775if (__builtin_expect (ptr != NULL, 1)) 13776 foo (*ptr); 13777@end smallexample 13778 13779@noindent 13780when testing pointer or floating-point values. 13781 13782For the purposes of branch prediction optimizations, the probability that 13783a @code{__builtin_expect} expression is @code{true} is controlled by GCC's 13784@code{builtin-expect-probability} parameter, which defaults to 90%. 13785 13786You can also use @code{__builtin_expect_with_probability} to explicitly 13787assign a probability value to individual expressions. If the built-in 13788is used in a loop construct, the provided probability will influence 13789the expected number of iterations made by loop optimizations. 13790@end deftypefn 13791 13792@deftypefn {Built-in Function} long __builtin_expect_with_probability 13793(long @var{exp}, long @var{c}, double @var{probability}) 13794 13795This function has the same semantics as @code{__builtin_expect}, 13796but the caller provides the expected probability that @var{exp} == @var{c}. 13797The last argument, @var{probability}, is a floating-point value in the 13798range 0.0 to 1.0, inclusive. The @var{probability} argument must be 13799constant floating-point expression. 13800@end deftypefn 13801 13802@deftypefn {Built-in Function} void __builtin_trap (void) 13803This function causes the program to exit abnormally. GCC implements 13804this function by using a target-dependent mechanism (such as 13805intentionally executing an illegal instruction) or by calling 13806@code{abort}. The mechanism used may vary from release to release so 13807you should not rely on any particular implementation. 13808@end deftypefn 13809 13810@deftypefn {Built-in Function} void __builtin_unreachable (void) 13811If control flow reaches the point of the @code{__builtin_unreachable}, 13812the program is undefined. It is useful in situations where the 13813compiler cannot deduce the unreachability of the code. 13814 13815One such case is immediately following an @code{asm} statement that 13816either never terminates, or one that transfers control elsewhere 13817and never returns. In this example, without the 13818@code{__builtin_unreachable}, GCC issues a warning that control 13819reaches the end of a non-void function. It also generates code 13820to return after the @code{asm}. 13821 13822@smallexample 13823int f (int c, int v) 13824@{ 13825 if (c) 13826 @{ 13827 return v; 13828 @} 13829 else 13830 @{ 13831 asm("jmp error_handler"); 13832 __builtin_unreachable (); 13833 @} 13834@} 13835@end smallexample 13836 13837@noindent 13838Because the @code{asm} statement unconditionally transfers control out 13839of the function, control never reaches the end of the function 13840body. The @code{__builtin_unreachable} is in fact unreachable and 13841communicates this fact to the compiler. 13842 13843Another use for @code{__builtin_unreachable} is following a call a 13844function that never returns but that is not declared 13845@code{__attribute__((noreturn))}, as in this example: 13846 13847@smallexample 13848void function_that_never_returns (void); 13849 13850int g (int c) 13851@{ 13852 if (c) 13853 @{ 13854 return 1; 13855 @} 13856 else 13857 @{ 13858 function_that_never_returns (); 13859 __builtin_unreachable (); 13860 @} 13861@} 13862@end smallexample 13863 13864@end deftypefn 13865 13866@deftypefn {Built-in Function} {void *} __builtin_assume_aligned (const void *@var{exp}, size_t @var{align}, ...) 13867This function returns its first argument, and allows the compiler 13868to assume that the returned pointer is at least @var{align} bytes 13869aligned. This built-in can have either two or three arguments, 13870if it has three, the third argument should have integer type, and 13871if it is nonzero means misalignment offset. For example: 13872 13873@smallexample 13874void *x = __builtin_assume_aligned (arg, 16); 13875@end smallexample 13876 13877@noindent 13878means that the compiler can assume @code{x}, set to @code{arg}, is at least 1387916-byte aligned, while: 13880 13881@smallexample 13882void *x = __builtin_assume_aligned (arg, 32, 8); 13883@end smallexample 13884 13885@noindent 13886means that the compiler can assume for @code{x}, set to @code{arg}, that 13887@code{(char *) x - 8} is 32-byte aligned. 13888@end deftypefn 13889 13890@deftypefn {Built-in Function} int __builtin_LINE () 13891This function is the equivalent of the preprocessor @code{__LINE__} 13892macro and returns a constant integer expression that evaluates to 13893the line number of the invocation of the built-in. When used as a C++ 13894default argument for a function @var{F}, it returns the line number 13895of the call to @var{F}. 13896@end deftypefn 13897 13898@deftypefn {Built-in Function} {const char *} __builtin_FUNCTION () 13899This function is the equivalent of the @code{__FUNCTION__} symbol 13900and returns an address constant pointing to the name of the function 13901from which the built-in was invoked, or the empty string if 13902the invocation is not at function scope. When used as a C++ default 13903argument for a function @var{F}, it returns the name of @var{F}'s 13904caller or the empty string if the call was not made at function 13905scope. 13906@end deftypefn 13907 13908@deftypefn {Built-in Function} {const char *} __builtin_FILE () 13909This function is the equivalent of the preprocessor @code{__FILE__} 13910macro and returns an address constant pointing to the file name 13911containing the invocation of the built-in, or the empty string if 13912the invocation is not at function scope. When used as a C++ default 13913argument for a function @var{F}, it returns the file name of the call 13914to @var{F} or the empty string if the call was not made at function 13915scope. 13916 13917For example, in the following, each call to function @code{foo} will 13918print a line similar to @code{"file.c:123: foo: message"} with the name 13919of the file and the line number of the @code{printf} call, the name of 13920the function @code{foo}, followed by the word @code{message}. 13921 13922@smallexample 13923const char* 13924function (const char *func = __builtin_FUNCTION ()) 13925@{ 13926 return func; 13927@} 13928 13929void foo (void) 13930@{ 13931 printf ("%s:%i: %s: message\n", file (), line (), function ()); 13932@} 13933@end smallexample 13934 13935@end deftypefn 13936 13937@deftypefn {Built-in Function} void __builtin___clear_cache (void *@var{begin}, void *@var{end}) 13938This function is used to flush the processor's instruction cache for 13939the region of memory between @var{begin} inclusive and @var{end} 13940exclusive. Some targets require that the instruction cache be 13941flushed, after modifying memory containing code, in order to obtain 13942deterministic behavior. 13943 13944If the target does not require instruction cache flushes, 13945@code{__builtin___clear_cache} has no effect. Otherwise either 13946instructions are emitted in-line to clear the instruction cache or a 13947call to the @code{__clear_cache} function in libgcc is made. 13948@end deftypefn 13949 13950@deftypefn {Built-in Function} void __builtin_prefetch (const void *@var{addr}, ...) 13951This function is used to minimize cache-miss latency by moving data into 13952a cache before it is accessed. 13953You can insert calls to @code{__builtin_prefetch} into code for which 13954you know addresses of data in memory that is likely to be accessed soon. 13955If the target supports them, data prefetch instructions are generated. 13956If the prefetch is done early enough before the access then the data will 13957be in the cache by the time it is accessed. 13958 13959The value of @var{addr} is the address of the memory to prefetch. 13960There are two optional arguments, @var{rw} and @var{locality}. 13961The value of @var{rw} is a compile-time constant one or zero; one 13962means that the prefetch is preparing for a write to the memory address 13963and zero, the default, means that the prefetch is preparing for a read. 13964The value @var{locality} must be a compile-time constant integer between 13965zero and three. A value of zero means that the data has no temporal 13966locality, so it need not be left in the cache after the access. A value 13967of three means that the data has a high degree of temporal locality and 13968should be left in all levels of cache possible. Values of one and two 13969mean, respectively, a low or moderate degree of temporal locality. The 13970default is three. 13971 13972@smallexample 13973for (i = 0; i < n; i++) 13974 @{ 13975 a[i] = a[i] + b[i]; 13976 __builtin_prefetch (&a[i+j], 1, 1); 13977 __builtin_prefetch (&b[i+j], 0, 1); 13978 /* @r{@dots{}} */ 13979 @} 13980@end smallexample 13981 13982Data prefetch does not generate faults if @var{addr} is invalid, but 13983the address expression itself must be valid. For example, a prefetch 13984of @code{p->next} does not fault if @code{p->next} is not a valid 13985address, but evaluation faults if @code{p} is not a valid address. 13986 13987If the target does not support data prefetch, the address expression 13988is evaluated if it includes side effects but no other code is generated 13989and GCC does not issue a warning. 13990@end deftypefn 13991 13992@deftypefn {Built-in Function}{size_t} __builtin_object_size (const void * @var{ptr}, int @var{type}) 13993Returns the size of an object pointed to by @var{ptr}. @xref{Object Size 13994Checking}, for a detailed description of the function. 13995@end deftypefn 13996 13997@deftypefn {Built-in Function} double __builtin_huge_val (void) 13998Returns a positive infinity, if supported by the floating-point format, 13999else @code{DBL_MAX}. This function is suitable for implementing the 14000ISO C macro @code{HUGE_VAL}. 14001@end deftypefn 14002 14003@deftypefn {Built-in Function} float __builtin_huge_valf (void) 14004Similar to @code{__builtin_huge_val}, except the return type is @code{float}. 14005@end deftypefn 14006 14007@deftypefn {Built-in Function} {long double} __builtin_huge_vall (void) 14008Similar to @code{__builtin_huge_val}, except the return 14009type is @code{long double}. 14010@end deftypefn 14011 14012@deftypefn {Built-in Function} _Float@var{n} __builtin_huge_valf@var{n} (void) 14013Similar to @code{__builtin_huge_val}, except the return type is 14014@code{_Float@var{n}}. 14015@end deftypefn 14016 14017@deftypefn {Built-in Function} _Float@var{n}x __builtin_huge_valf@var{n}x (void) 14018Similar to @code{__builtin_huge_val}, except the return type is 14019@code{_Float@var{n}x}. 14020@end deftypefn 14021 14022@deftypefn {Built-in Function} int __builtin_fpclassify (int, int, int, int, int, ...) 14023This built-in implements the C99 fpclassify functionality. The first 14024five int arguments should be the target library's notion of the 14025possible FP classes and are used for return values. They must be 14026constant values and they must appear in this order: @code{FP_NAN}, 14027@code{FP_INFINITE}, @code{FP_NORMAL}, @code{FP_SUBNORMAL} and 14028@code{FP_ZERO}. The ellipsis is for exactly one floating-point value 14029to classify. GCC treats the last argument as type-generic, which 14030means it does not do default promotion from float to double. 14031@end deftypefn 14032 14033@deftypefn {Built-in Function} double __builtin_inf (void) 14034Similar to @code{__builtin_huge_val}, except a warning is generated 14035if the target floating-point format does not support infinities. 14036@end deftypefn 14037 14038@deftypefn {Built-in Function} _Decimal32 __builtin_infd32 (void) 14039Similar to @code{__builtin_inf}, except the return type is @code{_Decimal32}. 14040@end deftypefn 14041 14042@deftypefn {Built-in Function} _Decimal64 __builtin_infd64 (void) 14043Similar to @code{__builtin_inf}, except the return type is @code{_Decimal64}. 14044@end deftypefn 14045 14046@deftypefn {Built-in Function} _Decimal128 __builtin_infd128 (void) 14047Similar to @code{__builtin_inf}, except the return type is @code{_Decimal128}. 14048@end deftypefn 14049 14050@deftypefn {Built-in Function} float __builtin_inff (void) 14051Similar to @code{__builtin_inf}, except the return type is @code{float}. 14052This function is suitable for implementing the ISO C99 macro @code{INFINITY}. 14053@end deftypefn 14054 14055@deftypefn {Built-in Function} {long double} __builtin_infl (void) 14056Similar to @code{__builtin_inf}, except the return 14057type is @code{long double}. 14058@end deftypefn 14059 14060@deftypefn {Built-in Function} _Float@var{n} __builtin_inff@var{n} (void) 14061Similar to @code{__builtin_inf}, except the return 14062type is @code{_Float@var{n}}. 14063@end deftypefn 14064 14065@deftypefn {Built-in Function} _Float@var{n} __builtin_inff@var{n}x (void) 14066Similar to @code{__builtin_inf}, except the return 14067type is @code{_Float@var{n}x}. 14068@end deftypefn 14069 14070@deftypefn {Built-in Function} int __builtin_isinf_sign (...) 14071Similar to @code{isinf}, except the return value is -1 for 14072an argument of @code{-Inf} and 1 for an argument of @code{+Inf}. 14073Note while the parameter list is an 14074ellipsis, this function only accepts exactly one floating-point 14075argument. GCC treats this parameter as type-generic, which means it 14076does not do default promotion from float to double. 14077@end deftypefn 14078 14079@deftypefn {Built-in Function} double __builtin_nan (const char *str) 14080This is an implementation of the ISO C99 function @code{nan}. 14081 14082Since ISO C99 defines this function in terms of @code{strtod}, which we 14083do not implement, a description of the parsing is in order. The string 14084is parsed as by @code{strtol}; that is, the base is recognized by 14085leading @samp{0} or @samp{0x} prefixes. The number parsed is placed 14086in the significand such that the least significant bit of the number 14087is at the least significant bit of the significand. The number is 14088truncated to fit the significand field provided. The significand is 14089forced to be a quiet NaN@. 14090 14091This function, if given a string literal all of which would have been 14092consumed by @code{strtol}, is evaluated early enough that it is considered a 14093compile-time constant. 14094@end deftypefn 14095 14096@deftypefn {Built-in Function} _Decimal32 __builtin_nand32 (const char *str) 14097Similar to @code{__builtin_nan}, except the return type is @code{_Decimal32}. 14098@end deftypefn 14099 14100@deftypefn {Built-in Function} _Decimal64 __builtin_nand64 (const char *str) 14101Similar to @code{__builtin_nan}, except the return type is @code{_Decimal64}. 14102@end deftypefn 14103 14104@deftypefn {Built-in Function} _Decimal128 __builtin_nand128 (const char *str) 14105Similar to @code{__builtin_nan}, except the return type is @code{_Decimal128}. 14106@end deftypefn 14107 14108@deftypefn {Built-in Function} float __builtin_nanf (const char *str) 14109Similar to @code{__builtin_nan}, except the return type is @code{float}. 14110@end deftypefn 14111 14112@deftypefn {Built-in Function} {long double} __builtin_nanl (const char *str) 14113Similar to @code{__builtin_nan}, except the return type is @code{long double}. 14114@end deftypefn 14115 14116@deftypefn {Built-in Function} _Float@var{n} __builtin_nanf@var{n} (const char *str) 14117Similar to @code{__builtin_nan}, except the return type is 14118@code{_Float@var{n}}. 14119@end deftypefn 14120 14121@deftypefn {Built-in Function} _Float@var{n}x __builtin_nanf@var{n}x (const char *str) 14122Similar to @code{__builtin_nan}, except the return type is 14123@code{_Float@var{n}x}. 14124@end deftypefn 14125 14126@deftypefn {Built-in Function} double __builtin_nans (const char *str) 14127Similar to @code{__builtin_nan}, except the significand is forced 14128to be a signaling NaN@. The @code{nans} function is proposed by 14129@uref{http://www.open-std.org/jtc1/sc22/wg14/www/docs/n965.htm,,WG14 N965}. 14130@end deftypefn 14131 14132@deftypefn {Built-in Function} _Decimal32 __builtin_nansd32 (const char *str) 14133Similar to @code{__builtin_nans}, except the return type is @code{_Decimal32}. 14134@end deftypefn 14135 14136@deftypefn {Built-in Function} _Decimal64 __builtin_nansd64 (const char *str) 14137Similar to @code{__builtin_nans}, except the return type is @code{_Decimal64}. 14138@end deftypefn 14139 14140@deftypefn {Built-in Function} _Decimal128 __builtin_nansd128 (const char *str) 14141Similar to @code{__builtin_nans}, except the return type is @code{_Decimal128}. 14142@end deftypefn 14143 14144@deftypefn {Built-in Function} float __builtin_nansf (const char *str) 14145Similar to @code{__builtin_nans}, except the return type is @code{float}. 14146@end deftypefn 14147 14148@deftypefn {Built-in Function} {long double} __builtin_nansl (const char *str) 14149Similar to @code{__builtin_nans}, except the return type is @code{long double}. 14150@end deftypefn 14151 14152@deftypefn {Built-in Function} _Float@var{n} __builtin_nansf@var{n} (const char *str) 14153Similar to @code{__builtin_nans}, except the return type is 14154@code{_Float@var{n}}. 14155@end deftypefn 14156 14157@deftypefn {Built-in Function} _Float@var{n}x __builtin_nansf@var{n}x (const char *str) 14158Similar to @code{__builtin_nans}, except the return type is 14159@code{_Float@var{n}x}. 14160@end deftypefn 14161 14162@deftypefn {Built-in Function} int __builtin_ffs (int x) 14163Returns one plus the index of the least significant 1-bit of @var{x}, or 14164if @var{x} is zero, returns zero. 14165@end deftypefn 14166 14167@deftypefn {Built-in Function} int __builtin_clz (unsigned int x) 14168Returns the number of leading 0-bits in @var{x}, starting at the most 14169significant bit position. If @var{x} is 0, the result is undefined. 14170@end deftypefn 14171 14172@deftypefn {Built-in Function} int __builtin_ctz (unsigned int x) 14173Returns the number of trailing 0-bits in @var{x}, starting at the least 14174significant bit position. If @var{x} is 0, the result is undefined. 14175@end deftypefn 14176 14177@deftypefn {Built-in Function} int __builtin_clrsb (int x) 14178Returns the number of leading redundant sign bits in @var{x}, i.e.@: the 14179number of bits following the most significant bit that are identical 14180to it. There are no special cases for 0 or other values. 14181@end deftypefn 14182 14183@deftypefn {Built-in Function} int __builtin_popcount (unsigned int x) 14184Returns the number of 1-bits in @var{x}. 14185@end deftypefn 14186 14187@deftypefn {Built-in Function} int __builtin_parity (unsigned int x) 14188Returns the parity of @var{x}, i.e.@: the number of 1-bits in @var{x} 14189modulo 2. 14190@end deftypefn 14191 14192@deftypefn {Built-in Function} int __builtin_ffsl (long) 14193Similar to @code{__builtin_ffs}, except the argument type is 14194@code{long}. 14195@end deftypefn 14196 14197@deftypefn {Built-in Function} int __builtin_clzl (unsigned long) 14198Similar to @code{__builtin_clz}, except the argument type is 14199@code{unsigned long}. 14200@end deftypefn 14201 14202@deftypefn {Built-in Function} int __builtin_ctzl (unsigned long) 14203Similar to @code{__builtin_ctz}, except the argument type is 14204@code{unsigned long}. 14205@end deftypefn 14206 14207@deftypefn {Built-in Function} int __builtin_clrsbl (long) 14208Similar to @code{__builtin_clrsb}, except the argument type is 14209@code{long}. 14210@end deftypefn 14211 14212@deftypefn {Built-in Function} int __builtin_popcountl (unsigned long) 14213Similar to @code{__builtin_popcount}, except the argument type is 14214@code{unsigned long}. 14215@end deftypefn 14216 14217@deftypefn {Built-in Function} int __builtin_parityl (unsigned long) 14218Similar to @code{__builtin_parity}, except the argument type is 14219@code{unsigned long}. 14220@end deftypefn 14221 14222@deftypefn {Built-in Function} int __builtin_ffsll (long long) 14223Similar to @code{__builtin_ffs}, except the argument type is 14224@code{long long}. 14225@end deftypefn 14226 14227@deftypefn {Built-in Function} int __builtin_clzll (unsigned long long) 14228Similar to @code{__builtin_clz}, except the argument type is 14229@code{unsigned long long}. 14230@end deftypefn 14231 14232@deftypefn {Built-in Function} int __builtin_ctzll (unsigned long long) 14233Similar to @code{__builtin_ctz}, except the argument type is 14234@code{unsigned long long}. 14235@end deftypefn 14236 14237@deftypefn {Built-in Function} int __builtin_clrsbll (long long) 14238Similar to @code{__builtin_clrsb}, except the argument type is 14239@code{long long}. 14240@end deftypefn 14241 14242@deftypefn {Built-in Function} int __builtin_popcountll (unsigned long long) 14243Similar to @code{__builtin_popcount}, except the argument type is 14244@code{unsigned long long}. 14245@end deftypefn 14246 14247@deftypefn {Built-in Function} int __builtin_parityll (unsigned long long) 14248Similar to @code{__builtin_parity}, except the argument type is 14249@code{unsigned long long}. 14250@end deftypefn 14251 14252@deftypefn {Built-in Function} double __builtin_powi (double, int) 14253Returns the first argument raised to the power of the second. Unlike the 14254@code{pow} function no guarantees about precision and rounding are made. 14255@end deftypefn 14256 14257@deftypefn {Built-in Function} float __builtin_powif (float, int) 14258Similar to @code{__builtin_powi}, except the argument and return types 14259are @code{float}. 14260@end deftypefn 14261 14262@deftypefn {Built-in Function} {long double} __builtin_powil (long double, int) 14263Similar to @code{__builtin_powi}, except the argument and return types 14264are @code{long double}. 14265@end deftypefn 14266 14267@deftypefn {Built-in Function} uint16_t __builtin_bswap16 (uint16_t x) 14268Returns @var{x} with the order of the bytes reversed; for example, 14269@code{0xaabb} becomes @code{0xbbaa}. Byte here always means 14270exactly 8 bits. 14271@end deftypefn 14272 14273@deftypefn {Built-in Function} uint32_t __builtin_bswap32 (uint32_t x) 14274Similar to @code{__builtin_bswap16}, except the argument and return types 14275are 32-bit. 14276@end deftypefn 14277 14278@deftypefn {Built-in Function} uint64_t __builtin_bswap64 (uint64_t x) 14279Similar to @code{__builtin_bswap32}, except the argument and return types 14280are 64-bit. 14281@end deftypefn 14282 14283@deftypefn {Built-in Function} uint128_t __builtin_bswap128 (uint128_t x) 14284Similar to @code{__builtin_bswap64}, except the argument and return types 14285are 128-bit. Only supported on targets when 128-bit types are supported. 14286@end deftypefn 14287 14288 14289@deftypefn {Built-in Function} Pmode __builtin_extend_pointer (void * x) 14290On targets where the user visible pointer size is smaller than the size 14291of an actual hardware address this function returns the extended user 14292pointer. Targets where this is true included ILP32 mode on x86_64 or 14293Aarch64. This function is mainly useful when writing inline assembly 14294code. 14295@end deftypefn 14296 14297@deftypefn {Built-in Function} int __builtin_goacc_parlevel_id (int x) 14298Returns the openacc gang, worker or vector id depending on whether @var{x} is 142990, 1 or 2. 14300@end deftypefn 14301 14302@deftypefn {Built-in Function} int __builtin_goacc_parlevel_size (int x) 14303Returns the openacc gang, worker or vector size depending on whether @var{x} is 143040, 1 or 2. 14305@end deftypefn 14306 14307@node Target Builtins 14308@section Built-in Functions Specific to Particular Target Machines 14309 14310On some target machines, GCC supports many built-in functions specific 14311to those machines. Generally these generate calls to specific machine 14312instructions, but allow the compiler to schedule those calls. 14313 14314@menu 14315* AArch64 Built-in Functions:: 14316* Alpha Built-in Functions:: 14317* Altera Nios II Built-in Functions:: 14318* ARC Built-in Functions:: 14319* ARC SIMD Built-in Functions:: 14320* ARM iWMMXt Built-in Functions:: 14321* ARM C Language Extensions (ACLE):: 14322* ARM Floating Point Status and Control Intrinsics:: 14323* ARM ARMv8-M Security Extensions:: 14324* AVR Built-in Functions:: 14325* Blackfin Built-in Functions:: 14326* BPF Built-in Functions:: 14327* FR-V Built-in Functions:: 14328* MIPS DSP Built-in Functions:: 14329* MIPS Paired-Single Support:: 14330* MIPS Loongson Built-in Functions:: 14331* MIPS SIMD Architecture (MSA) Support:: 14332* Other MIPS Built-in Functions:: 14333* MSP430 Built-in Functions:: 14334* NDS32 Built-in Functions:: 14335* picoChip Built-in Functions:: 14336* Basic PowerPC Built-in Functions:: 14337* PowerPC AltiVec/VSX Built-in Functions:: 14338* PowerPC Hardware Transactional Memory Built-in Functions:: 14339* PowerPC Atomic Memory Operation Functions:: 14340* PowerPC Matrix-Multiply Assist Built-in Functions:: 14341* PRU Built-in Functions:: 14342* RISC-V Built-in Functions:: 14343* RX Built-in Functions:: 14344* S/390 System z Built-in Functions:: 14345* SH Built-in Functions:: 14346* SPARC VIS Built-in Functions:: 14347* TI C6X Built-in Functions:: 14348* TILE-Gx Built-in Functions:: 14349* TILEPro Built-in Functions:: 14350* x86 Built-in Functions:: 14351* x86 transactional memory intrinsics:: 14352* x86 control-flow protection intrinsics:: 14353@end menu 14354 14355@node AArch64 Built-in Functions 14356@subsection AArch64 Built-in Functions 14357 14358These built-in functions are available for the AArch64 family of 14359processors. 14360@smallexample 14361unsigned int __builtin_aarch64_get_fpcr () 14362void __builtin_aarch64_set_fpcr (unsigned int) 14363unsigned int __builtin_aarch64_get_fpsr () 14364void __builtin_aarch64_set_fpsr (unsigned int) 14365 14366unsigned long long __builtin_aarch64_get_fpcr64 () 14367void __builtin_aarch64_set_fpcr64 (unsigned long long) 14368unsigned long long __builtin_aarch64_get_fpsr64 () 14369void __builtin_aarch64_set_fpsr64 (unsigned long long) 14370@end smallexample 14371 14372@node Alpha Built-in Functions 14373@subsection Alpha Built-in Functions 14374 14375These built-in functions are available for the Alpha family of 14376processors, depending on the command-line switches used. 14377 14378The following built-in functions are always available. They 14379all generate the machine instruction that is part of the name. 14380 14381@smallexample 14382long __builtin_alpha_implver (void) 14383long __builtin_alpha_rpcc (void) 14384long __builtin_alpha_amask (long) 14385long __builtin_alpha_cmpbge (long, long) 14386long __builtin_alpha_extbl (long, long) 14387long __builtin_alpha_extwl (long, long) 14388long __builtin_alpha_extll (long, long) 14389long __builtin_alpha_extql (long, long) 14390long __builtin_alpha_extwh (long, long) 14391long __builtin_alpha_extlh (long, long) 14392long __builtin_alpha_extqh (long, long) 14393long __builtin_alpha_insbl (long, long) 14394long __builtin_alpha_inswl (long, long) 14395long __builtin_alpha_insll (long, long) 14396long __builtin_alpha_insql (long, long) 14397long __builtin_alpha_inswh (long, long) 14398long __builtin_alpha_inslh (long, long) 14399long __builtin_alpha_insqh (long, long) 14400long __builtin_alpha_mskbl (long, long) 14401long __builtin_alpha_mskwl (long, long) 14402long __builtin_alpha_mskll (long, long) 14403long __builtin_alpha_mskql (long, long) 14404long __builtin_alpha_mskwh (long, long) 14405long __builtin_alpha_msklh (long, long) 14406long __builtin_alpha_mskqh (long, long) 14407long __builtin_alpha_umulh (long, long) 14408long __builtin_alpha_zap (long, long) 14409long __builtin_alpha_zapnot (long, long) 14410@end smallexample 14411 14412The following built-in functions are always with @option{-mmax} 14413or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{pca56} or 14414later. They all generate the machine instruction that is part 14415of the name. 14416 14417@smallexample 14418long __builtin_alpha_pklb (long) 14419long __builtin_alpha_pkwb (long) 14420long __builtin_alpha_unpkbl (long) 14421long __builtin_alpha_unpkbw (long) 14422long __builtin_alpha_minub8 (long, long) 14423long __builtin_alpha_minsb8 (long, long) 14424long __builtin_alpha_minuw4 (long, long) 14425long __builtin_alpha_minsw4 (long, long) 14426long __builtin_alpha_maxub8 (long, long) 14427long __builtin_alpha_maxsb8 (long, long) 14428long __builtin_alpha_maxuw4 (long, long) 14429long __builtin_alpha_maxsw4 (long, long) 14430long __builtin_alpha_perr (long, long) 14431@end smallexample 14432 14433The following built-in functions are always with @option{-mcix} 14434or @option{-mcpu=@var{cpu}} where @var{cpu} is @code{ev67} or 14435later. They all generate the machine instruction that is part 14436of the name. 14437 14438@smallexample 14439long __builtin_alpha_cttz (long) 14440long __builtin_alpha_ctlz (long) 14441long __builtin_alpha_ctpop (long) 14442@end smallexample 14443 14444The following built-in functions are available on systems that use the OSF/1 14445PALcode. Normally they invoke the @code{rduniq} and @code{wruniq} 14446PAL calls, but when invoked with @option{-mtls-kernel}, they invoke 14447@code{rdval} and @code{wrval}. 14448 14449@smallexample 14450void *__builtin_thread_pointer (void) 14451void __builtin_set_thread_pointer (void *) 14452@end smallexample 14453 14454@node Altera Nios II Built-in Functions 14455@subsection Altera Nios II Built-in Functions 14456 14457These built-in functions are available for the Altera Nios II 14458family of processors. 14459 14460The following built-in functions are always available. They 14461all generate the machine instruction that is part of the name. 14462 14463@example 14464int __builtin_ldbio (volatile const void *) 14465int __builtin_ldbuio (volatile const void *) 14466int __builtin_ldhio (volatile const void *) 14467int __builtin_ldhuio (volatile const void *) 14468int __builtin_ldwio (volatile const void *) 14469void __builtin_stbio (volatile void *, int) 14470void __builtin_sthio (volatile void *, int) 14471void __builtin_stwio (volatile void *, int) 14472void __builtin_sync (void) 14473int __builtin_rdctl (int) 14474int __builtin_rdprs (int, int) 14475void __builtin_wrctl (int, int) 14476void __builtin_flushd (volatile void *) 14477void __builtin_flushda (volatile void *) 14478int __builtin_wrpie (int); 14479void __builtin_eni (int); 14480int __builtin_ldex (volatile const void *) 14481int __builtin_stex (volatile void *, int) 14482int __builtin_ldsex (volatile const void *) 14483int __builtin_stsex (volatile void *, int) 14484@end example 14485 14486The following built-in functions are always available. They 14487all generate a Nios II Custom Instruction. The name of the 14488function represents the types that the function takes and 14489returns. The letter before the @code{n} is the return type 14490or void if absent. The @code{n} represents the first parameter 14491to all the custom instructions, the custom instruction number. 14492The two letters after the @code{n} represent the up to two 14493parameters to the function. 14494 14495The letters represent the following data types: 14496@table @code 14497@item <no letter> 14498@code{void} for return type and no parameter for parameter types. 14499 14500@item i 14501@code{int} for return type and parameter type 14502 14503@item f 14504@code{float} for return type and parameter type 14505 14506@item p 14507@code{void *} for return type and parameter type 14508 14509@end table 14510 14511And the function names are: 14512@example 14513void __builtin_custom_n (void) 14514void __builtin_custom_ni (int) 14515void __builtin_custom_nf (float) 14516void __builtin_custom_np (void *) 14517void __builtin_custom_nii (int, int) 14518void __builtin_custom_nif (int, float) 14519void __builtin_custom_nip (int, void *) 14520void __builtin_custom_nfi (float, int) 14521void __builtin_custom_nff (float, float) 14522void __builtin_custom_nfp (float, void *) 14523void __builtin_custom_npi (void *, int) 14524void __builtin_custom_npf (void *, float) 14525void __builtin_custom_npp (void *, void *) 14526int __builtin_custom_in (void) 14527int __builtin_custom_ini (int) 14528int __builtin_custom_inf (float) 14529int __builtin_custom_inp (void *) 14530int __builtin_custom_inii (int, int) 14531int __builtin_custom_inif (int, float) 14532int __builtin_custom_inip (int, void *) 14533int __builtin_custom_infi (float, int) 14534int __builtin_custom_inff (float, float) 14535int __builtin_custom_infp (float, void *) 14536int __builtin_custom_inpi (void *, int) 14537int __builtin_custom_inpf (void *, float) 14538int __builtin_custom_inpp (void *, void *) 14539float __builtin_custom_fn (void) 14540float __builtin_custom_fni (int) 14541float __builtin_custom_fnf (float) 14542float __builtin_custom_fnp (void *) 14543float __builtin_custom_fnii (int, int) 14544float __builtin_custom_fnif (int, float) 14545float __builtin_custom_fnip (int, void *) 14546float __builtin_custom_fnfi (float, int) 14547float __builtin_custom_fnff (float, float) 14548float __builtin_custom_fnfp (float, void *) 14549float __builtin_custom_fnpi (void *, int) 14550float __builtin_custom_fnpf (void *, float) 14551float __builtin_custom_fnpp (void *, void *) 14552void * __builtin_custom_pn (void) 14553void * __builtin_custom_pni (int) 14554void * __builtin_custom_pnf (float) 14555void * __builtin_custom_pnp (void *) 14556void * __builtin_custom_pnii (int, int) 14557void * __builtin_custom_pnif (int, float) 14558void * __builtin_custom_pnip (int, void *) 14559void * __builtin_custom_pnfi (float, int) 14560void * __builtin_custom_pnff (float, float) 14561void * __builtin_custom_pnfp (float, void *) 14562void * __builtin_custom_pnpi (void *, int) 14563void * __builtin_custom_pnpf (void *, float) 14564void * __builtin_custom_pnpp (void *, void *) 14565@end example 14566 14567@node ARC Built-in Functions 14568@subsection ARC Built-in Functions 14569 14570The following built-in functions are provided for ARC targets. The 14571built-ins generate the corresponding assembly instructions. In the 14572examples given below, the generated code often requires an operand or 14573result to be in a register. Where necessary further code will be 14574generated to ensure this is true, but for brevity this is not 14575described in each case. 14576 14577@emph{Note:} Using a built-in to generate an instruction not supported 14578by a target may cause problems. At present the compiler is not 14579guaranteed to detect such misuse, and as a result an internal compiler 14580error may be generated. 14581 14582@deftypefn {Built-in Function} int __builtin_arc_aligned (void *@var{val}, int @var{alignval}) 14583Return 1 if @var{val} is known to have the byte alignment given 14584by @var{alignval}, otherwise return 0. 14585Note that this is different from 14586@smallexample 14587__alignof__(*(char *)@var{val}) >= alignval 14588@end smallexample 14589because __alignof__ sees only the type of the dereference, whereas 14590__builtin_arc_align uses alignment information from the pointer 14591as well as from the pointed-to type. 14592The information available will depend on optimization level. 14593@end deftypefn 14594 14595@deftypefn {Built-in Function} void __builtin_arc_brk (void) 14596Generates 14597@example 14598brk 14599@end example 14600@end deftypefn 14601 14602@deftypefn {Built-in Function} {unsigned int} __builtin_arc_core_read (unsigned int @var{regno}) 14603The operand is the number of a register to be read. Generates: 14604@example 14605mov @var{dest}, r@var{regno} 14606@end example 14607where the value in @var{dest} will be the result returned from the 14608built-in. 14609@end deftypefn 14610 14611@deftypefn {Built-in Function} void __builtin_arc_core_write (unsigned int @var{regno}, unsigned int @var{val}) 14612The first operand is the number of a register to be written, the 14613second operand is a compile time constant to write into that 14614register. Generates: 14615@example 14616mov r@var{regno}, @var{val} 14617@end example 14618@end deftypefn 14619 14620@deftypefn {Built-in Function} int __builtin_arc_divaw (int @var{a}, int @var{b}) 14621Only available if either @option{-mcpu=ARC700} or @option{-meA} is set. 14622Generates: 14623@example 14624divaw @var{dest}, @var{a}, @var{b} 14625@end example 14626where the value in @var{dest} will be the result returned from the 14627built-in. 14628@end deftypefn 14629 14630@deftypefn {Built-in Function} void __builtin_arc_flag (unsigned int @var{a}) 14631Generates 14632@example 14633flag @var{a} 14634@end example 14635@end deftypefn 14636 14637@deftypefn {Built-in Function} {unsigned int} __builtin_arc_lr (unsigned int @var{auxr}) 14638The operand, @var{auxv}, is the address of an auxiliary register and 14639must be a compile time constant. Generates: 14640@example 14641lr @var{dest}, [@var{auxr}] 14642@end example 14643Where the value in @var{dest} will be the result returned from the 14644built-in. 14645@end deftypefn 14646 14647@deftypefn {Built-in Function} void __builtin_arc_mul64 (int @var{a}, int @var{b}) 14648Only available with @option{-mmul64}. Generates: 14649@example 14650mul64 @var{a}, @var{b} 14651@end example 14652@end deftypefn 14653 14654@deftypefn {Built-in Function} void __builtin_arc_mulu64 (unsigned int @var{a}, unsigned int @var{b}) 14655Only available with @option{-mmul64}. Generates: 14656@example 14657mulu64 @var{a}, @var{b} 14658@end example 14659@end deftypefn 14660 14661@deftypefn {Built-in Function} void __builtin_arc_nop (void) 14662Generates: 14663@example 14664nop 14665@end example 14666@end deftypefn 14667 14668@deftypefn {Built-in Function} int __builtin_arc_norm (int @var{src}) 14669Only valid if the @samp{norm} instruction is available through the 14670@option{-mnorm} option or by default with @option{-mcpu=ARC700}. 14671Generates: 14672@example 14673norm @var{dest}, @var{src} 14674@end example 14675Where the value in @var{dest} will be the result returned from the 14676built-in. 14677@end deftypefn 14678 14679@deftypefn {Built-in Function} {short int} __builtin_arc_normw (short int @var{src}) 14680Only valid if the @samp{normw} instruction is available through the 14681@option{-mnorm} option or by default with @option{-mcpu=ARC700}. 14682Generates: 14683@example 14684normw @var{dest}, @var{src} 14685@end example 14686Where the value in @var{dest} will be the result returned from the 14687built-in. 14688@end deftypefn 14689 14690@deftypefn {Built-in Function} void __builtin_arc_rtie (void) 14691Generates: 14692@example 14693rtie 14694@end example 14695@end deftypefn 14696 14697@deftypefn {Built-in Function} void __builtin_arc_sleep (int @var{a} 14698Generates: 14699@example 14700sleep @var{a} 14701@end example 14702@end deftypefn 14703 14704@deftypefn {Built-in Function} void __builtin_arc_sr (unsigned int @var{auxr}, unsigned int @var{val}) 14705The first argument, @var{auxv}, is the address of an auxiliary 14706register, the second argument, @var{val}, is a compile time constant 14707to be written to the register. Generates: 14708@example 14709sr @var{auxr}, [@var{val}] 14710@end example 14711@end deftypefn 14712 14713@deftypefn {Built-in Function} int __builtin_arc_swap (int @var{src}) 14714Only valid with @option{-mswap}. Generates: 14715@example 14716swap @var{dest}, @var{src} 14717@end example 14718Where the value in @var{dest} will be the result returned from the 14719built-in. 14720@end deftypefn 14721 14722@deftypefn {Built-in Function} void __builtin_arc_swi (void) 14723Generates: 14724@example 14725swi 14726@end example 14727@end deftypefn 14728 14729@deftypefn {Built-in Function} void __builtin_arc_sync (void) 14730Only available with @option{-mcpu=ARC700}. Generates: 14731@example 14732sync 14733@end example 14734@end deftypefn 14735 14736@deftypefn {Built-in Function} void __builtin_arc_trap_s (unsigned int @var{c}) 14737Only available with @option{-mcpu=ARC700}. Generates: 14738@example 14739trap_s @var{c} 14740@end example 14741@end deftypefn 14742 14743@deftypefn {Built-in Function} void __builtin_arc_unimp_s (void) 14744Only available with @option{-mcpu=ARC700}. Generates: 14745@example 14746unimp_s 14747@end example 14748@end deftypefn 14749 14750The instructions generated by the following builtins are not 14751considered as candidates for scheduling. They are not moved around by 14752the compiler during scheduling, and thus can be expected to appear 14753where they are put in the C code: 14754@example 14755__builtin_arc_brk() 14756__builtin_arc_core_read() 14757__builtin_arc_core_write() 14758__builtin_arc_flag() 14759__builtin_arc_lr() 14760__builtin_arc_sleep() 14761__builtin_arc_sr() 14762__builtin_arc_swi() 14763@end example 14764 14765@node ARC SIMD Built-in Functions 14766@subsection ARC SIMD Built-in Functions 14767 14768SIMD builtins provided by the compiler can be used to generate the 14769vector instructions. This section describes the available builtins 14770and their usage in programs. With the @option{-msimd} option, the 14771compiler provides 128-bit vector types, which can be specified using 14772the @code{vector_size} attribute. The header file @file{arc-simd.h} 14773can be included to use the following predefined types: 14774@example 14775typedef int __v4si __attribute__((vector_size(16))); 14776typedef short __v8hi __attribute__((vector_size(16))); 14777@end example 14778 14779These types can be used to define 128-bit variables. The built-in 14780functions listed in the following section can be used on these 14781variables to generate the vector operations. 14782 14783For all builtins, @code{__builtin_arc_@var{someinsn}}, the header file 14784@file{arc-simd.h} also provides equivalent macros called 14785@code{_@var{someinsn}} that can be used for programming ease and 14786improved readability. The following macros for DMA control are also 14787provided: 14788@example 14789#define _setup_dma_in_channel_reg _vdiwr 14790#define _setup_dma_out_channel_reg _vdowr 14791@end example 14792 14793The following is a complete list of all the SIMD built-ins provided 14794for ARC, grouped by calling signature. 14795 14796The following take two @code{__v8hi} arguments and return a 14797@code{__v8hi} result: 14798@example 14799__v8hi __builtin_arc_vaddaw (__v8hi, __v8hi) 14800__v8hi __builtin_arc_vaddw (__v8hi, __v8hi) 14801__v8hi __builtin_arc_vand (__v8hi, __v8hi) 14802__v8hi __builtin_arc_vandaw (__v8hi, __v8hi) 14803__v8hi __builtin_arc_vavb (__v8hi, __v8hi) 14804__v8hi __builtin_arc_vavrb (__v8hi, __v8hi) 14805__v8hi __builtin_arc_vbic (__v8hi, __v8hi) 14806__v8hi __builtin_arc_vbicaw (__v8hi, __v8hi) 14807__v8hi __builtin_arc_vdifaw (__v8hi, __v8hi) 14808__v8hi __builtin_arc_vdifw (__v8hi, __v8hi) 14809__v8hi __builtin_arc_veqw (__v8hi, __v8hi) 14810__v8hi __builtin_arc_vh264f (__v8hi, __v8hi) 14811__v8hi __builtin_arc_vh264ft (__v8hi, __v8hi) 14812__v8hi __builtin_arc_vh264fw (__v8hi, __v8hi) 14813__v8hi __builtin_arc_vlew (__v8hi, __v8hi) 14814__v8hi __builtin_arc_vltw (__v8hi, __v8hi) 14815__v8hi __builtin_arc_vmaxaw (__v8hi, __v8hi) 14816__v8hi __builtin_arc_vmaxw (__v8hi, __v8hi) 14817__v8hi __builtin_arc_vminaw (__v8hi, __v8hi) 14818__v8hi __builtin_arc_vminw (__v8hi, __v8hi) 14819__v8hi __builtin_arc_vmr1aw (__v8hi, __v8hi) 14820__v8hi __builtin_arc_vmr1w (__v8hi, __v8hi) 14821__v8hi __builtin_arc_vmr2aw (__v8hi, __v8hi) 14822__v8hi __builtin_arc_vmr2w (__v8hi, __v8hi) 14823__v8hi __builtin_arc_vmr3aw (__v8hi, __v8hi) 14824__v8hi __builtin_arc_vmr3w (__v8hi, __v8hi) 14825__v8hi __builtin_arc_vmr4aw (__v8hi, __v8hi) 14826__v8hi __builtin_arc_vmr4w (__v8hi, __v8hi) 14827__v8hi __builtin_arc_vmr5aw (__v8hi, __v8hi) 14828__v8hi __builtin_arc_vmr5w (__v8hi, __v8hi) 14829__v8hi __builtin_arc_vmr6aw (__v8hi, __v8hi) 14830__v8hi __builtin_arc_vmr6w (__v8hi, __v8hi) 14831__v8hi __builtin_arc_vmr7aw (__v8hi, __v8hi) 14832__v8hi __builtin_arc_vmr7w (__v8hi, __v8hi) 14833__v8hi __builtin_arc_vmrb (__v8hi, __v8hi) 14834__v8hi __builtin_arc_vmulaw (__v8hi, __v8hi) 14835__v8hi __builtin_arc_vmulfaw (__v8hi, __v8hi) 14836__v8hi __builtin_arc_vmulfw (__v8hi, __v8hi) 14837__v8hi __builtin_arc_vmulw (__v8hi, __v8hi) 14838__v8hi __builtin_arc_vnew (__v8hi, __v8hi) 14839__v8hi __builtin_arc_vor (__v8hi, __v8hi) 14840__v8hi __builtin_arc_vsubaw (__v8hi, __v8hi) 14841__v8hi __builtin_arc_vsubw (__v8hi, __v8hi) 14842__v8hi __builtin_arc_vsummw (__v8hi, __v8hi) 14843__v8hi __builtin_arc_vvc1f (__v8hi, __v8hi) 14844__v8hi __builtin_arc_vvc1ft (__v8hi, __v8hi) 14845__v8hi __builtin_arc_vxor (__v8hi, __v8hi) 14846__v8hi __builtin_arc_vxoraw (__v8hi, __v8hi) 14847@end example 14848 14849The following take one @code{__v8hi} and one @code{int} argument and return a 14850@code{__v8hi} result: 14851 14852@example 14853__v8hi __builtin_arc_vbaddw (__v8hi, int) 14854__v8hi __builtin_arc_vbmaxw (__v8hi, int) 14855__v8hi __builtin_arc_vbminw (__v8hi, int) 14856__v8hi __builtin_arc_vbmulaw (__v8hi, int) 14857__v8hi __builtin_arc_vbmulfw (__v8hi, int) 14858__v8hi __builtin_arc_vbmulw (__v8hi, int) 14859__v8hi __builtin_arc_vbrsubw (__v8hi, int) 14860__v8hi __builtin_arc_vbsubw (__v8hi, int) 14861@end example 14862 14863The following take one @code{__v8hi} argument and one @code{int} argument which 14864must be a 3-bit compile time constant indicating a register number 14865I0-I7. They return a @code{__v8hi} result. 14866@example 14867__v8hi __builtin_arc_vasrw (__v8hi, const int) 14868__v8hi __builtin_arc_vsr8 (__v8hi, const int) 14869__v8hi __builtin_arc_vsr8aw (__v8hi, const int) 14870@end example 14871 14872The following take one @code{__v8hi} argument and one @code{int} 14873argument which must be a 6-bit compile time constant. They return a 14874@code{__v8hi} result. 14875@example 14876__v8hi __builtin_arc_vasrpwbi (__v8hi, const int) 14877__v8hi __builtin_arc_vasrrpwbi (__v8hi, const int) 14878__v8hi __builtin_arc_vasrrwi (__v8hi, const int) 14879__v8hi __builtin_arc_vasrsrwi (__v8hi, const int) 14880__v8hi __builtin_arc_vasrwi (__v8hi, const int) 14881__v8hi __builtin_arc_vsr8awi (__v8hi, const int) 14882__v8hi __builtin_arc_vsr8i (__v8hi, const int) 14883@end example 14884 14885The following take one @code{__v8hi} argument and one @code{int} argument which 14886must be a 8-bit compile time constant. They return a @code{__v8hi} 14887result. 14888@example 14889__v8hi __builtin_arc_vd6tapf (__v8hi, const int) 14890__v8hi __builtin_arc_vmvaw (__v8hi, const int) 14891__v8hi __builtin_arc_vmvw (__v8hi, const int) 14892__v8hi __builtin_arc_vmvzw (__v8hi, const int) 14893@end example 14894 14895The following take two @code{int} arguments, the second of which which 14896must be a 8-bit compile time constant. They return a @code{__v8hi} 14897result: 14898@example 14899__v8hi __builtin_arc_vmovaw (int, const int) 14900__v8hi __builtin_arc_vmovw (int, const int) 14901__v8hi __builtin_arc_vmovzw (int, const int) 14902@end example 14903 14904The following take a single @code{__v8hi} argument and return a 14905@code{__v8hi} result: 14906@example 14907__v8hi __builtin_arc_vabsaw (__v8hi) 14908__v8hi __builtin_arc_vabsw (__v8hi) 14909__v8hi __builtin_arc_vaddsuw (__v8hi) 14910__v8hi __builtin_arc_vexch1 (__v8hi) 14911__v8hi __builtin_arc_vexch2 (__v8hi) 14912__v8hi __builtin_arc_vexch4 (__v8hi) 14913__v8hi __builtin_arc_vsignw (__v8hi) 14914__v8hi __builtin_arc_vupbaw (__v8hi) 14915__v8hi __builtin_arc_vupbw (__v8hi) 14916__v8hi __builtin_arc_vupsbaw (__v8hi) 14917__v8hi __builtin_arc_vupsbw (__v8hi) 14918@end example 14919 14920The following take two @code{int} arguments and return no result: 14921@example 14922void __builtin_arc_vdirun (int, int) 14923void __builtin_arc_vdorun (int, int) 14924@end example 14925 14926The following take two @code{int} arguments and return no result. The 14927first argument must a 3-bit compile time constant indicating one of 14928the DR0-DR7 DMA setup channels: 14929@example 14930void __builtin_arc_vdiwr (const int, int) 14931void __builtin_arc_vdowr (const int, int) 14932@end example 14933 14934The following take an @code{int} argument and return no result: 14935@example 14936void __builtin_arc_vendrec (int) 14937void __builtin_arc_vrec (int) 14938void __builtin_arc_vrecrun (int) 14939void __builtin_arc_vrun (int) 14940@end example 14941 14942The following take a @code{__v8hi} argument and two @code{int} 14943arguments and return a @code{__v8hi} result. The second argument must 14944be a 3-bit compile time constants, indicating one the registers I0-I7, 14945and the third argument must be an 8-bit compile time constant. 14946 14947@emph{Note:} Although the equivalent hardware instructions do not take 14948an SIMD register as an operand, these builtins overwrite the relevant 14949bits of the @code{__v8hi} register provided as the first argument with 14950the value loaded from the @code{[Ib, u8]} location in the SDM. 14951 14952@example 14953__v8hi __builtin_arc_vld32 (__v8hi, const int, const int) 14954__v8hi __builtin_arc_vld32wh (__v8hi, const int, const int) 14955__v8hi __builtin_arc_vld32wl (__v8hi, const int, const int) 14956__v8hi __builtin_arc_vld64 (__v8hi, const int, const int) 14957@end example 14958 14959The following take two @code{int} arguments and return a @code{__v8hi} 14960result. The first argument must be a 3-bit compile time constants, 14961indicating one the registers I0-I7, and the second argument must be an 149628-bit compile time constant. 14963 14964@example 14965__v8hi __builtin_arc_vld128 (const int, const int) 14966__v8hi __builtin_arc_vld64w (const int, const int) 14967@end example 14968 14969The following take a @code{__v8hi} argument and two @code{int} 14970arguments and return no result. The second argument must be a 3-bit 14971compile time constants, indicating one the registers I0-I7, and the 14972third argument must be an 8-bit compile time constant. 14973 14974@example 14975void __builtin_arc_vst128 (__v8hi, const int, const int) 14976void __builtin_arc_vst64 (__v8hi, const int, const int) 14977@end example 14978 14979The following take a @code{__v8hi} argument and three @code{int} 14980arguments and return no result. The second argument must be a 3-bit 14981compile-time constant, identifying the 16-bit sub-register to be 14982stored, the third argument must be a 3-bit compile time constants, 14983indicating one the registers I0-I7, and the fourth argument must be an 149848-bit compile time constant. 14985 14986@example 14987void __builtin_arc_vst16_n (__v8hi, const int, const int, const int) 14988void __builtin_arc_vst32_n (__v8hi, const int, const int, const int) 14989@end example 14990 14991@node ARM iWMMXt Built-in Functions 14992@subsection ARM iWMMXt Built-in Functions 14993 14994These built-in functions are available for the ARM family of 14995processors when the @option{-mcpu=iwmmxt} switch is used: 14996 14997@smallexample 14998typedef int v2si __attribute__ ((vector_size (8))); 14999typedef short v4hi __attribute__ ((vector_size (8))); 15000typedef char v8qi __attribute__ ((vector_size (8))); 15001 15002int __builtin_arm_getwcgr0 (void) 15003void __builtin_arm_setwcgr0 (int) 15004int __builtin_arm_getwcgr1 (void) 15005void __builtin_arm_setwcgr1 (int) 15006int __builtin_arm_getwcgr2 (void) 15007void __builtin_arm_setwcgr2 (int) 15008int __builtin_arm_getwcgr3 (void) 15009void __builtin_arm_setwcgr3 (int) 15010int __builtin_arm_textrmsb (v8qi, int) 15011int __builtin_arm_textrmsh (v4hi, int) 15012int __builtin_arm_textrmsw (v2si, int) 15013int __builtin_arm_textrmub (v8qi, int) 15014int __builtin_arm_textrmuh (v4hi, int) 15015int __builtin_arm_textrmuw (v2si, int) 15016v8qi __builtin_arm_tinsrb (v8qi, int, int) 15017v4hi __builtin_arm_tinsrh (v4hi, int, int) 15018v2si __builtin_arm_tinsrw (v2si, int, int) 15019long long __builtin_arm_tmia (long long, int, int) 15020long long __builtin_arm_tmiabb (long long, int, int) 15021long long __builtin_arm_tmiabt (long long, int, int) 15022long long __builtin_arm_tmiaph (long long, int, int) 15023long long __builtin_arm_tmiatb (long long, int, int) 15024long long __builtin_arm_tmiatt (long long, int, int) 15025int __builtin_arm_tmovmskb (v8qi) 15026int __builtin_arm_tmovmskh (v4hi) 15027int __builtin_arm_tmovmskw (v2si) 15028long long __builtin_arm_waccb (v8qi) 15029long long __builtin_arm_wacch (v4hi) 15030long long __builtin_arm_waccw (v2si) 15031v8qi __builtin_arm_waddb (v8qi, v8qi) 15032v8qi __builtin_arm_waddbss (v8qi, v8qi) 15033v8qi __builtin_arm_waddbus (v8qi, v8qi) 15034v4hi __builtin_arm_waddh (v4hi, v4hi) 15035v4hi __builtin_arm_waddhss (v4hi, v4hi) 15036v4hi __builtin_arm_waddhus (v4hi, v4hi) 15037v2si __builtin_arm_waddw (v2si, v2si) 15038v2si __builtin_arm_waddwss (v2si, v2si) 15039v2si __builtin_arm_waddwus (v2si, v2si) 15040v8qi __builtin_arm_walign (v8qi, v8qi, int) 15041long long __builtin_arm_wand(long long, long long) 15042long long __builtin_arm_wandn (long long, long long) 15043v8qi __builtin_arm_wavg2b (v8qi, v8qi) 15044v8qi __builtin_arm_wavg2br (v8qi, v8qi) 15045v4hi __builtin_arm_wavg2h (v4hi, v4hi) 15046v4hi __builtin_arm_wavg2hr (v4hi, v4hi) 15047v8qi __builtin_arm_wcmpeqb (v8qi, v8qi) 15048v4hi __builtin_arm_wcmpeqh (v4hi, v4hi) 15049v2si __builtin_arm_wcmpeqw (v2si, v2si) 15050v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi) 15051v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi) 15052v2si __builtin_arm_wcmpgtsw (v2si, v2si) 15053v8qi __builtin_arm_wcmpgtub (v8qi, v8qi) 15054v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi) 15055v2si __builtin_arm_wcmpgtuw (v2si, v2si) 15056long long __builtin_arm_wmacs (long long, v4hi, v4hi) 15057long long __builtin_arm_wmacsz (v4hi, v4hi) 15058long long __builtin_arm_wmacu (long long, v4hi, v4hi) 15059long long __builtin_arm_wmacuz (v4hi, v4hi) 15060v4hi __builtin_arm_wmadds (v4hi, v4hi) 15061v4hi __builtin_arm_wmaddu (v4hi, v4hi) 15062v8qi __builtin_arm_wmaxsb (v8qi, v8qi) 15063v4hi __builtin_arm_wmaxsh (v4hi, v4hi) 15064v2si __builtin_arm_wmaxsw (v2si, v2si) 15065v8qi __builtin_arm_wmaxub (v8qi, v8qi) 15066v4hi __builtin_arm_wmaxuh (v4hi, v4hi) 15067v2si __builtin_arm_wmaxuw (v2si, v2si) 15068v8qi __builtin_arm_wminsb (v8qi, v8qi) 15069v4hi __builtin_arm_wminsh (v4hi, v4hi) 15070v2si __builtin_arm_wminsw (v2si, v2si) 15071v8qi __builtin_arm_wminub (v8qi, v8qi) 15072v4hi __builtin_arm_wminuh (v4hi, v4hi) 15073v2si __builtin_arm_wminuw (v2si, v2si) 15074v4hi __builtin_arm_wmulsm (v4hi, v4hi) 15075v4hi __builtin_arm_wmulul (v4hi, v4hi) 15076v4hi __builtin_arm_wmulum (v4hi, v4hi) 15077long long __builtin_arm_wor (long long, long long) 15078v2si __builtin_arm_wpackdss (long long, long long) 15079v2si __builtin_arm_wpackdus (long long, long long) 15080v8qi __builtin_arm_wpackhss (v4hi, v4hi) 15081v8qi __builtin_arm_wpackhus (v4hi, v4hi) 15082v4hi __builtin_arm_wpackwss (v2si, v2si) 15083v4hi __builtin_arm_wpackwus (v2si, v2si) 15084long long __builtin_arm_wrord (long long, long long) 15085long long __builtin_arm_wrordi (long long, int) 15086v4hi __builtin_arm_wrorh (v4hi, long long) 15087v4hi __builtin_arm_wrorhi (v4hi, int) 15088v2si __builtin_arm_wrorw (v2si, long long) 15089v2si __builtin_arm_wrorwi (v2si, int) 15090v2si __builtin_arm_wsadb (v2si, v8qi, v8qi) 15091v2si __builtin_arm_wsadbz (v8qi, v8qi) 15092v2si __builtin_arm_wsadh (v2si, v4hi, v4hi) 15093v2si __builtin_arm_wsadhz (v4hi, v4hi) 15094v4hi __builtin_arm_wshufh (v4hi, int) 15095long long __builtin_arm_wslld (long long, long long) 15096long long __builtin_arm_wslldi (long long, int) 15097v4hi __builtin_arm_wsllh (v4hi, long long) 15098v4hi __builtin_arm_wsllhi (v4hi, int) 15099v2si __builtin_arm_wsllw (v2si, long long) 15100v2si __builtin_arm_wsllwi (v2si, int) 15101long long __builtin_arm_wsrad (long long, long long) 15102long long __builtin_arm_wsradi (long long, int) 15103v4hi __builtin_arm_wsrah (v4hi, long long) 15104v4hi __builtin_arm_wsrahi (v4hi, int) 15105v2si __builtin_arm_wsraw (v2si, long long) 15106v2si __builtin_arm_wsrawi (v2si, int) 15107long long __builtin_arm_wsrld (long long, long long) 15108long long __builtin_arm_wsrldi (long long, int) 15109v4hi __builtin_arm_wsrlh (v4hi, long long) 15110v4hi __builtin_arm_wsrlhi (v4hi, int) 15111v2si __builtin_arm_wsrlw (v2si, long long) 15112v2si __builtin_arm_wsrlwi (v2si, int) 15113v8qi __builtin_arm_wsubb (v8qi, v8qi) 15114v8qi __builtin_arm_wsubbss (v8qi, v8qi) 15115v8qi __builtin_arm_wsubbus (v8qi, v8qi) 15116v4hi __builtin_arm_wsubh (v4hi, v4hi) 15117v4hi __builtin_arm_wsubhss (v4hi, v4hi) 15118v4hi __builtin_arm_wsubhus (v4hi, v4hi) 15119v2si __builtin_arm_wsubw (v2si, v2si) 15120v2si __builtin_arm_wsubwss (v2si, v2si) 15121v2si __builtin_arm_wsubwus (v2si, v2si) 15122v4hi __builtin_arm_wunpckehsb (v8qi) 15123v2si __builtin_arm_wunpckehsh (v4hi) 15124long long __builtin_arm_wunpckehsw (v2si) 15125v4hi __builtin_arm_wunpckehub (v8qi) 15126v2si __builtin_arm_wunpckehuh (v4hi) 15127long long __builtin_arm_wunpckehuw (v2si) 15128v4hi __builtin_arm_wunpckelsb (v8qi) 15129v2si __builtin_arm_wunpckelsh (v4hi) 15130long long __builtin_arm_wunpckelsw (v2si) 15131v4hi __builtin_arm_wunpckelub (v8qi) 15132v2si __builtin_arm_wunpckeluh (v4hi) 15133long long __builtin_arm_wunpckeluw (v2si) 15134v8qi __builtin_arm_wunpckihb (v8qi, v8qi) 15135v4hi __builtin_arm_wunpckihh (v4hi, v4hi) 15136v2si __builtin_arm_wunpckihw (v2si, v2si) 15137v8qi __builtin_arm_wunpckilb (v8qi, v8qi) 15138v4hi __builtin_arm_wunpckilh (v4hi, v4hi) 15139v2si __builtin_arm_wunpckilw (v2si, v2si) 15140long long __builtin_arm_wxor (long long, long long) 15141long long __builtin_arm_wzero () 15142@end smallexample 15143 15144 15145@node ARM C Language Extensions (ACLE) 15146@subsection ARM C Language Extensions (ACLE) 15147 15148GCC implements extensions for C as described in the ARM C Language 15149Extensions (ACLE) specification, which can be found at 15150@uref{https://developer.arm.com/documentation/ihi0053/latest/}. 15151 15152As a part of ACLE, GCC implements extensions for Advanced SIMD as described in 15153the ARM C Language Extensions Specification. The complete list of Advanced SIMD 15154intrinsics can be found at 15155@uref{https://developer.arm.com/documentation/ihi0073/latest/}. 15156The built-in intrinsics for the Advanced SIMD extension are available when 15157NEON is enabled. 15158 15159Currently, ARM and AArch64 back ends do not support ACLE 2.0 fully. Both 15160back ends support CRC32 intrinsics and the ARM back end supports the 15161Coprocessor intrinsics, all from @file{arm_acle.h}. The ARM back end's 16-bit 15162floating-point Advanced SIMD intrinsics currently comply to ACLE v1.1. 15163AArch64's back end does not have support for 16-bit floating point Advanced SIMD 15164intrinsics yet. 15165 15166See @ref{ARM Options} and @ref{AArch64 Options} for more information on the 15167availability of extensions. 15168 15169@node ARM Floating Point Status and Control Intrinsics 15170@subsection ARM Floating Point Status and Control Intrinsics 15171 15172These built-in functions are available for the ARM family of 15173processors with floating-point unit. 15174 15175@smallexample 15176unsigned int __builtin_arm_get_fpscr () 15177void __builtin_arm_set_fpscr (unsigned int) 15178@end smallexample 15179 15180@node ARM ARMv8-M Security Extensions 15181@subsection ARM ARMv8-M Security Extensions 15182 15183GCC implements the ARMv8-M Security Extensions as described in the ARMv8-M 15184Security Extensions: Requirements on Development Tools Engineering 15185Specification, which can be found at 15186@uref{https://developer.arm.com/documentation/ecm0359818/latest/}. 15187 15188As part of the Security Extensions GCC implements two new function attributes: 15189@code{cmse_nonsecure_entry} and @code{cmse_nonsecure_call}. 15190 15191As part of the Security Extensions GCC implements the intrinsics below. FPTR 15192is used here to mean any function pointer type. 15193 15194@smallexample 15195cmse_address_info_t cmse_TT (void *) 15196cmse_address_info_t cmse_TT_fptr (FPTR) 15197cmse_address_info_t cmse_TTT (void *) 15198cmse_address_info_t cmse_TTT_fptr (FPTR) 15199cmse_address_info_t cmse_TTA (void *) 15200cmse_address_info_t cmse_TTA_fptr (FPTR) 15201cmse_address_info_t cmse_TTAT (void *) 15202cmse_address_info_t cmse_TTAT_fptr (FPTR) 15203void * cmse_check_address_range (void *, size_t, int) 15204typeof(p) cmse_nsfptr_create (FPTR p) 15205intptr_t cmse_is_nsfptr (FPTR) 15206int cmse_nonsecure_caller (void) 15207@end smallexample 15208 15209@node AVR Built-in Functions 15210@subsection AVR Built-in Functions 15211 15212For each built-in function for AVR, there is an equally named, 15213uppercase built-in macro defined. That way users can easily query if 15214or if not a specific built-in is implemented or not. For example, if 15215@code{__builtin_avr_nop} is available the macro 15216@code{__BUILTIN_AVR_NOP} is defined to @code{1} and undefined otherwise. 15217 15218@table @code 15219 15220@item void __builtin_avr_nop (void) 15221@itemx void __builtin_avr_sei (void) 15222@itemx void __builtin_avr_cli (void) 15223@itemx void __builtin_avr_sleep (void) 15224@itemx void __builtin_avr_wdr (void) 15225@itemx unsigned char __builtin_avr_swap (unsigned char) 15226@itemx unsigned int __builtin_avr_fmul (unsigned char, unsigned char) 15227@itemx int __builtin_avr_fmuls (char, char) 15228@itemx int __builtin_avr_fmulsu (char, unsigned char) 15229These built-in functions map to the respective machine 15230instruction, i.e.@: @code{nop}, @code{sei}, @code{cli}, @code{sleep}, 15231@code{wdr}, @code{swap}, @code{fmul}, @code{fmuls} 15232resp. @code{fmulsu}. The three @code{fmul*} built-ins are implemented 15233as library call if no hardware multiplier is available. 15234 15235@item void __builtin_avr_delay_cycles (unsigned long ticks) 15236Delay execution for @var{ticks} cycles. Note that this 15237built-in does not take into account the effect of interrupts that 15238might increase delay time. @var{ticks} must be a compile-time 15239integer constant; delays with a variable number of cycles are not supported. 15240 15241@item char __builtin_avr_flash_segment (const __memx void*) 15242This built-in takes a byte address to the 24-bit 15243@ref{AVR Named Address Spaces,address space} @code{__memx} and returns 15244the number of the flash segment (the 64 KiB chunk) where the address 15245points to. Counting starts at @code{0}. 15246If the address does not point to flash memory, return @code{-1}. 15247 15248@item uint8_t __builtin_avr_insert_bits (uint32_t map, uint8_t bits, uint8_t val) 15249Insert bits from @var{bits} into @var{val} and return the resulting 15250value. The nibbles of @var{map} determine how the insertion is 15251performed: Let @var{X} be the @var{n}-th nibble of @var{map} 15252@enumerate 15253@item If @var{X} is @code{0xf}, 15254then the @var{n}-th bit of @var{val} is returned unaltered. 15255 15256@item If X is in the range 0@dots{}7, 15257then the @var{n}-th result bit is set to the @var{X}-th bit of @var{bits} 15258 15259@item If X is in the range 8@dots{}@code{0xe}, 15260then the @var{n}-th result bit is undefined. 15261@end enumerate 15262 15263@noindent 15264One typical use case for this built-in is adjusting input and 15265output values to non-contiguous port layouts. Some examples: 15266 15267@smallexample 15268// same as val, bits is unused 15269__builtin_avr_insert_bits (0xffffffff, bits, val) 15270@end smallexample 15271 15272@smallexample 15273// same as bits, val is unused 15274__builtin_avr_insert_bits (0x76543210, bits, val) 15275@end smallexample 15276 15277@smallexample 15278// same as rotating bits by 4 15279__builtin_avr_insert_bits (0x32107654, bits, 0) 15280@end smallexample 15281 15282@smallexample 15283// high nibble of result is the high nibble of val 15284// low nibble of result is the low nibble of bits 15285__builtin_avr_insert_bits (0xffff3210, bits, val) 15286@end smallexample 15287 15288@smallexample 15289// reverse the bit order of bits 15290__builtin_avr_insert_bits (0x01234567, bits, 0) 15291@end smallexample 15292 15293@item void __builtin_avr_nops (unsigned count) 15294Insert @var{count} @code{NOP} instructions. 15295The number of instructions must be a compile-time integer constant. 15296 15297@end table 15298 15299@noindent 15300There are many more AVR-specific built-in functions that are used to 15301implement the ISO/IEC TR 18037 ``Embedded C'' fixed-point functions of 15302section 7.18a.6. You don't need to use these built-ins directly. 15303Instead, use the declarations as supplied by the @code{stdfix.h} header 15304with GNU-C99: 15305 15306@smallexample 15307#include <stdfix.h> 15308 15309// Re-interpret the bit representation of unsigned 16-bit 15310// integer @var{uval} as Q-format 0.16 value. 15311unsigned fract get_bits (uint_ur_t uval) 15312@{ 15313 return urbits (uval); 15314@} 15315@end smallexample 15316 15317@node Blackfin Built-in Functions 15318@subsection Blackfin Built-in Functions 15319 15320Currently, there are two Blackfin-specific built-in functions. These are 15321used for generating @code{CSYNC} and @code{SSYNC} machine insns without 15322using inline assembly; by using these built-in functions the compiler can 15323automatically add workarounds for hardware errata involving these 15324instructions. These functions are named as follows: 15325 15326@smallexample 15327void __builtin_bfin_csync (void) 15328void __builtin_bfin_ssync (void) 15329@end smallexample 15330 15331@node BPF Built-in Functions 15332@subsection BPF Built-in Functions 15333 15334The following built-in functions are available for eBPF targets. 15335 15336@deftypefn {Built-in Function} unsigned long long __builtin_bpf_load_byte (unsigned long long @var{offset}) 15337Load a byte from the @code{struct sk_buff} packet data pointed by the register @code{%r6} and return it. 15338@end deftypefn 15339 15340@deftypefn {Built-in Function} unsigned long long __builtin_bpf_load_half (unsigned long long @var{offset}) 15341Load 16-bits from the @code{struct sk_buff} packet data pointed by the register @code{%r6} and return it. 15342@end deftypefn 15343 15344@deftypefn {Built-in Function} unsigned long long __builtin_bpf_load_word (unsigned long long @var{offset}) 15345Load 32-bits from the @code{struct sk_buff} packet data pointed by the register @code{%r6} and return it. 15346@end deftypefn 15347 15348@node FR-V Built-in Functions 15349@subsection FR-V Built-in Functions 15350 15351GCC provides many FR-V-specific built-in functions. In general, 15352these functions are intended to be compatible with those described 15353by @cite{FR-V Family, Softune C/C++ Compiler Manual (V6), Fujitsu 15354Semiconductor}. The two exceptions are @code{__MDUNPACKH} and 15355@code{__MBTOHE}, the GCC forms of which pass 128-bit values by 15356pointer rather than by value. 15357 15358Most of the functions are named after specific FR-V instructions. 15359Such functions are said to be ``directly mapped'' and are summarized 15360here in tabular form. 15361 15362@menu 15363* Argument Types:: 15364* Directly-mapped Integer Functions:: 15365* Directly-mapped Media Functions:: 15366* Raw read/write Functions:: 15367* Other Built-in Functions:: 15368@end menu 15369 15370@node Argument Types 15371@subsubsection Argument Types 15372 15373The arguments to the built-in functions can be divided into three groups: 15374register numbers, compile-time constants and run-time values. In order 15375to make this classification clear at a glance, the arguments and return 15376values are given the following pseudo types: 15377 15378@multitable @columnfractions .20 .30 .15 .35 15379@item Pseudo type @tab Real C type @tab Constant? @tab Description 15380@item @code{uh} @tab @code{unsigned short} @tab No @tab an unsigned halfword 15381@item @code{uw1} @tab @code{unsigned int} @tab No @tab an unsigned word 15382@item @code{sw1} @tab @code{int} @tab No @tab a signed word 15383@item @code{uw2} @tab @code{unsigned long long} @tab No 15384@tab an unsigned doubleword 15385@item @code{sw2} @tab @code{long long} @tab No @tab a signed doubleword 15386@item @code{const} @tab @code{int} @tab Yes @tab an integer constant 15387@item @code{acc} @tab @code{int} @tab Yes @tab an ACC register number 15388@item @code{iacc} @tab @code{int} @tab Yes @tab an IACC register number 15389@end multitable 15390 15391These pseudo types are not defined by GCC, they are simply a notational 15392convenience used in this manual. 15393 15394Arguments of type @code{uh}, @code{uw1}, @code{sw1}, @code{uw2} 15395and @code{sw2} are evaluated at run time. They correspond to 15396register operands in the underlying FR-V instructions. 15397 15398@code{const} arguments represent immediate operands in the underlying 15399FR-V instructions. They must be compile-time constants. 15400 15401@code{acc} arguments are evaluated at compile time and specify the number 15402of an accumulator register. For example, an @code{acc} argument of 2 15403selects the ACC2 register. 15404 15405@code{iacc} arguments are similar to @code{acc} arguments but specify the 15406number of an IACC register. See @pxref{Other Built-in Functions} 15407for more details. 15408 15409@node Directly-mapped Integer Functions 15410@subsubsection Directly-Mapped Integer Functions 15411 15412The functions listed below map directly to FR-V I-type instructions. 15413 15414@multitable @columnfractions .45 .32 .23 15415@item Function prototype @tab Example usage @tab Assembly output 15416@item @code{sw1 __ADDSS (sw1, sw1)} 15417@tab @code{@var{c} = __ADDSS (@var{a}, @var{b})} 15418@tab @code{ADDSS @var{a},@var{b},@var{c}} 15419@item @code{sw1 __SCAN (sw1, sw1)} 15420@tab @code{@var{c} = __SCAN (@var{a}, @var{b})} 15421@tab @code{SCAN @var{a},@var{b},@var{c}} 15422@item @code{sw1 __SCUTSS (sw1)} 15423@tab @code{@var{b} = __SCUTSS (@var{a})} 15424@tab @code{SCUTSS @var{a},@var{b}} 15425@item @code{sw1 __SLASS (sw1, sw1)} 15426@tab @code{@var{c} = __SLASS (@var{a}, @var{b})} 15427@tab @code{SLASS @var{a},@var{b},@var{c}} 15428@item @code{void __SMASS (sw1, sw1)} 15429@tab @code{__SMASS (@var{a}, @var{b})} 15430@tab @code{SMASS @var{a},@var{b}} 15431@item @code{void __SMSSS (sw1, sw1)} 15432@tab @code{__SMSSS (@var{a}, @var{b})} 15433@tab @code{SMSSS @var{a},@var{b}} 15434@item @code{void __SMU (sw1, sw1)} 15435@tab @code{__SMU (@var{a}, @var{b})} 15436@tab @code{SMU @var{a},@var{b}} 15437@item @code{sw2 __SMUL (sw1, sw1)} 15438@tab @code{@var{c} = __SMUL (@var{a}, @var{b})} 15439@tab @code{SMUL @var{a},@var{b},@var{c}} 15440@item @code{sw1 __SUBSS (sw1, sw1)} 15441@tab @code{@var{c} = __SUBSS (@var{a}, @var{b})} 15442@tab @code{SUBSS @var{a},@var{b},@var{c}} 15443@item @code{uw2 __UMUL (uw1, uw1)} 15444@tab @code{@var{c} = __UMUL (@var{a}, @var{b})} 15445@tab @code{UMUL @var{a},@var{b},@var{c}} 15446@end multitable 15447 15448@node Directly-mapped Media Functions 15449@subsubsection Directly-Mapped Media Functions 15450 15451The functions listed below map directly to FR-V M-type instructions. 15452 15453@multitable @columnfractions .45 .32 .23 15454@item Function prototype @tab Example usage @tab Assembly output 15455@item @code{uw1 __MABSHS (sw1)} 15456@tab @code{@var{b} = __MABSHS (@var{a})} 15457@tab @code{MABSHS @var{a},@var{b}} 15458@item @code{void __MADDACCS (acc, acc)} 15459@tab @code{__MADDACCS (@var{b}, @var{a})} 15460@tab @code{MADDACCS @var{a},@var{b}} 15461@item @code{sw1 __MADDHSS (sw1, sw1)} 15462@tab @code{@var{c} = __MADDHSS (@var{a}, @var{b})} 15463@tab @code{MADDHSS @var{a},@var{b},@var{c}} 15464@item @code{uw1 __MADDHUS (uw1, uw1)} 15465@tab @code{@var{c} = __MADDHUS (@var{a}, @var{b})} 15466@tab @code{MADDHUS @var{a},@var{b},@var{c}} 15467@item @code{uw1 __MAND (uw1, uw1)} 15468@tab @code{@var{c} = __MAND (@var{a}, @var{b})} 15469@tab @code{MAND @var{a},@var{b},@var{c}} 15470@item @code{void __MASACCS (acc, acc)} 15471@tab @code{__MASACCS (@var{b}, @var{a})} 15472@tab @code{MASACCS @var{a},@var{b}} 15473@item @code{uw1 __MAVEH (uw1, uw1)} 15474@tab @code{@var{c} = __MAVEH (@var{a}, @var{b})} 15475@tab @code{MAVEH @var{a},@var{b},@var{c}} 15476@item @code{uw2 __MBTOH (uw1)} 15477@tab @code{@var{b} = __MBTOH (@var{a})} 15478@tab @code{MBTOH @var{a},@var{b}} 15479@item @code{void __MBTOHE (uw1 *, uw1)} 15480@tab @code{__MBTOHE (&@var{b}, @var{a})} 15481@tab @code{MBTOHE @var{a},@var{b}} 15482@item @code{void __MCLRACC (acc)} 15483@tab @code{__MCLRACC (@var{a})} 15484@tab @code{MCLRACC @var{a}} 15485@item @code{void __MCLRACCA (void)} 15486@tab @code{__MCLRACCA ()} 15487@tab @code{MCLRACCA} 15488@item @code{uw1 __Mcop1 (uw1, uw1)} 15489@tab @code{@var{c} = __Mcop1 (@var{a}, @var{b})} 15490@tab @code{Mcop1 @var{a},@var{b},@var{c}} 15491@item @code{uw1 __Mcop2 (uw1, uw1)} 15492@tab @code{@var{c} = __Mcop2 (@var{a}, @var{b})} 15493@tab @code{Mcop2 @var{a},@var{b},@var{c}} 15494@item @code{uw1 __MCPLHI (uw2, const)} 15495@tab @code{@var{c} = __MCPLHI (@var{a}, @var{b})} 15496@tab @code{MCPLHI @var{a},#@var{b},@var{c}} 15497@item @code{uw1 __MCPLI (uw2, const)} 15498@tab @code{@var{c} = __MCPLI (@var{a}, @var{b})} 15499@tab @code{MCPLI @var{a},#@var{b},@var{c}} 15500@item @code{void __MCPXIS (acc, sw1, sw1)} 15501@tab @code{__MCPXIS (@var{c}, @var{a}, @var{b})} 15502@tab @code{MCPXIS @var{a},@var{b},@var{c}} 15503@item @code{void __MCPXIU (acc, uw1, uw1)} 15504@tab @code{__MCPXIU (@var{c}, @var{a}, @var{b})} 15505@tab @code{MCPXIU @var{a},@var{b},@var{c}} 15506@item @code{void __MCPXRS (acc, sw1, sw1)} 15507@tab @code{__MCPXRS (@var{c}, @var{a}, @var{b})} 15508@tab @code{MCPXRS @var{a},@var{b},@var{c}} 15509@item @code{void __MCPXRU (acc, uw1, uw1)} 15510@tab @code{__MCPXRU (@var{c}, @var{a}, @var{b})} 15511@tab @code{MCPXRU @var{a},@var{b},@var{c}} 15512@item @code{uw1 __MCUT (acc, uw1)} 15513@tab @code{@var{c} = __MCUT (@var{a}, @var{b})} 15514@tab @code{MCUT @var{a},@var{b},@var{c}} 15515@item @code{uw1 __MCUTSS (acc, sw1)} 15516@tab @code{@var{c} = __MCUTSS (@var{a}, @var{b})} 15517@tab @code{MCUTSS @var{a},@var{b},@var{c}} 15518@item @code{void __MDADDACCS (acc, acc)} 15519@tab @code{__MDADDACCS (@var{b}, @var{a})} 15520@tab @code{MDADDACCS @var{a},@var{b}} 15521@item @code{void __MDASACCS (acc, acc)} 15522@tab @code{__MDASACCS (@var{b}, @var{a})} 15523@tab @code{MDASACCS @var{a},@var{b}} 15524@item @code{uw2 __MDCUTSSI (acc, const)} 15525@tab @code{@var{c} = __MDCUTSSI (@var{a}, @var{b})} 15526@tab @code{MDCUTSSI @var{a},#@var{b},@var{c}} 15527@item @code{uw2 __MDPACKH (uw2, uw2)} 15528@tab @code{@var{c} = __MDPACKH (@var{a}, @var{b})} 15529@tab @code{MDPACKH @var{a},@var{b},@var{c}} 15530@item @code{uw2 __MDROTLI (uw2, const)} 15531@tab @code{@var{c} = __MDROTLI (@var{a}, @var{b})} 15532@tab @code{MDROTLI @var{a},#@var{b},@var{c}} 15533@item @code{void __MDSUBACCS (acc, acc)} 15534@tab @code{__MDSUBACCS (@var{b}, @var{a})} 15535@tab @code{MDSUBACCS @var{a},@var{b}} 15536@item @code{void __MDUNPACKH (uw1 *, uw2)} 15537@tab @code{__MDUNPACKH (&@var{b}, @var{a})} 15538@tab @code{MDUNPACKH @var{a},@var{b}} 15539@item @code{uw2 __MEXPDHD (uw1, const)} 15540@tab @code{@var{c} = __MEXPDHD (@var{a}, @var{b})} 15541@tab @code{MEXPDHD @var{a},#@var{b},@var{c}} 15542@item @code{uw1 __MEXPDHW (uw1, const)} 15543@tab @code{@var{c} = __MEXPDHW (@var{a}, @var{b})} 15544@tab @code{MEXPDHW @var{a},#@var{b},@var{c}} 15545@item @code{uw1 __MHDSETH (uw1, const)} 15546@tab @code{@var{c} = __MHDSETH (@var{a}, @var{b})} 15547@tab @code{MHDSETH @var{a},#@var{b},@var{c}} 15548@item @code{sw1 __MHDSETS (const)} 15549@tab @code{@var{b} = __MHDSETS (@var{a})} 15550@tab @code{MHDSETS #@var{a},@var{b}} 15551@item @code{uw1 __MHSETHIH (uw1, const)} 15552@tab @code{@var{b} = __MHSETHIH (@var{b}, @var{a})} 15553@tab @code{MHSETHIH #@var{a},@var{b}} 15554@item @code{sw1 __MHSETHIS (sw1, const)} 15555@tab @code{@var{b} = __MHSETHIS (@var{b}, @var{a})} 15556@tab @code{MHSETHIS #@var{a},@var{b}} 15557@item @code{uw1 __MHSETLOH (uw1, const)} 15558@tab @code{@var{b} = __MHSETLOH (@var{b}, @var{a})} 15559@tab @code{MHSETLOH #@var{a},@var{b}} 15560@item @code{sw1 __MHSETLOS (sw1, const)} 15561@tab @code{@var{b} = __MHSETLOS (@var{b}, @var{a})} 15562@tab @code{MHSETLOS #@var{a},@var{b}} 15563@item @code{uw1 __MHTOB (uw2)} 15564@tab @code{@var{b} = __MHTOB (@var{a})} 15565@tab @code{MHTOB @var{a},@var{b}} 15566@item @code{void __MMACHS (acc, sw1, sw1)} 15567@tab @code{__MMACHS (@var{c}, @var{a}, @var{b})} 15568@tab @code{MMACHS @var{a},@var{b},@var{c}} 15569@item @code{void __MMACHU (acc, uw1, uw1)} 15570@tab @code{__MMACHU (@var{c}, @var{a}, @var{b})} 15571@tab @code{MMACHU @var{a},@var{b},@var{c}} 15572@item @code{void __MMRDHS (acc, sw1, sw1)} 15573@tab @code{__MMRDHS (@var{c}, @var{a}, @var{b})} 15574@tab @code{MMRDHS @var{a},@var{b},@var{c}} 15575@item @code{void __MMRDHU (acc, uw1, uw1)} 15576@tab @code{__MMRDHU (@var{c}, @var{a}, @var{b})} 15577@tab @code{MMRDHU @var{a},@var{b},@var{c}} 15578@item @code{void __MMULHS (acc, sw1, sw1)} 15579@tab @code{__MMULHS (@var{c}, @var{a}, @var{b})} 15580@tab @code{MMULHS @var{a},@var{b},@var{c}} 15581@item @code{void __MMULHU (acc, uw1, uw1)} 15582@tab @code{__MMULHU (@var{c}, @var{a}, @var{b})} 15583@tab @code{MMULHU @var{a},@var{b},@var{c}} 15584@item @code{void __MMULXHS (acc, sw1, sw1)} 15585@tab @code{__MMULXHS (@var{c}, @var{a}, @var{b})} 15586@tab @code{MMULXHS @var{a},@var{b},@var{c}} 15587@item @code{void __MMULXHU (acc, uw1, uw1)} 15588@tab @code{__MMULXHU (@var{c}, @var{a}, @var{b})} 15589@tab @code{MMULXHU @var{a},@var{b},@var{c}} 15590@item @code{uw1 __MNOT (uw1)} 15591@tab @code{@var{b} = __MNOT (@var{a})} 15592@tab @code{MNOT @var{a},@var{b}} 15593@item @code{uw1 __MOR (uw1, uw1)} 15594@tab @code{@var{c} = __MOR (@var{a}, @var{b})} 15595@tab @code{MOR @var{a},@var{b},@var{c}} 15596@item @code{uw1 __MPACKH (uh, uh)} 15597@tab @code{@var{c} = __MPACKH (@var{a}, @var{b})} 15598@tab @code{MPACKH @var{a},@var{b},@var{c}} 15599@item @code{sw2 __MQADDHSS (sw2, sw2)} 15600@tab @code{@var{c} = __MQADDHSS (@var{a}, @var{b})} 15601@tab @code{MQADDHSS @var{a},@var{b},@var{c}} 15602@item @code{uw2 __MQADDHUS (uw2, uw2)} 15603@tab @code{@var{c} = __MQADDHUS (@var{a}, @var{b})} 15604@tab @code{MQADDHUS @var{a},@var{b},@var{c}} 15605@item @code{void __MQCPXIS (acc, sw2, sw2)} 15606@tab @code{__MQCPXIS (@var{c}, @var{a}, @var{b})} 15607@tab @code{MQCPXIS @var{a},@var{b},@var{c}} 15608@item @code{void __MQCPXIU (acc, uw2, uw2)} 15609@tab @code{__MQCPXIU (@var{c}, @var{a}, @var{b})} 15610@tab @code{MQCPXIU @var{a},@var{b},@var{c}} 15611@item @code{void __MQCPXRS (acc, sw2, sw2)} 15612@tab @code{__MQCPXRS (@var{c}, @var{a}, @var{b})} 15613@tab @code{MQCPXRS @var{a},@var{b},@var{c}} 15614@item @code{void __MQCPXRU (acc, uw2, uw2)} 15615@tab @code{__MQCPXRU (@var{c}, @var{a}, @var{b})} 15616@tab @code{MQCPXRU @var{a},@var{b},@var{c}} 15617@item @code{sw2 __MQLCLRHS (sw2, sw2)} 15618@tab @code{@var{c} = __MQLCLRHS (@var{a}, @var{b})} 15619@tab @code{MQLCLRHS @var{a},@var{b},@var{c}} 15620@item @code{sw2 __MQLMTHS (sw2, sw2)} 15621@tab @code{@var{c} = __MQLMTHS (@var{a}, @var{b})} 15622@tab @code{MQLMTHS @var{a},@var{b},@var{c}} 15623@item @code{void __MQMACHS (acc, sw2, sw2)} 15624@tab @code{__MQMACHS (@var{c}, @var{a}, @var{b})} 15625@tab @code{MQMACHS @var{a},@var{b},@var{c}} 15626@item @code{void __MQMACHU (acc, uw2, uw2)} 15627@tab @code{__MQMACHU (@var{c}, @var{a}, @var{b})} 15628@tab @code{MQMACHU @var{a},@var{b},@var{c}} 15629@item @code{void __MQMACXHS (acc, sw2, sw2)} 15630@tab @code{__MQMACXHS (@var{c}, @var{a}, @var{b})} 15631@tab @code{MQMACXHS @var{a},@var{b},@var{c}} 15632@item @code{void __MQMULHS (acc, sw2, sw2)} 15633@tab @code{__MQMULHS (@var{c}, @var{a}, @var{b})} 15634@tab @code{MQMULHS @var{a},@var{b},@var{c}} 15635@item @code{void __MQMULHU (acc, uw2, uw2)} 15636@tab @code{__MQMULHU (@var{c}, @var{a}, @var{b})} 15637@tab @code{MQMULHU @var{a},@var{b},@var{c}} 15638@item @code{void __MQMULXHS (acc, sw2, sw2)} 15639@tab @code{__MQMULXHS (@var{c}, @var{a}, @var{b})} 15640@tab @code{MQMULXHS @var{a},@var{b},@var{c}} 15641@item @code{void __MQMULXHU (acc, uw2, uw2)} 15642@tab @code{__MQMULXHU (@var{c}, @var{a}, @var{b})} 15643@tab @code{MQMULXHU @var{a},@var{b},@var{c}} 15644@item @code{sw2 __MQSATHS (sw2, sw2)} 15645@tab @code{@var{c} = __MQSATHS (@var{a}, @var{b})} 15646@tab @code{MQSATHS @var{a},@var{b},@var{c}} 15647@item @code{uw2 __MQSLLHI (uw2, int)} 15648@tab @code{@var{c} = __MQSLLHI (@var{a}, @var{b})} 15649@tab @code{MQSLLHI @var{a},@var{b},@var{c}} 15650@item @code{sw2 __MQSRAHI (sw2, int)} 15651@tab @code{@var{c} = __MQSRAHI (@var{a}, @var{b})} 15652@tab @code{MQSRAHI @var{a},@var{b},@var{c}} 15653@item @code{sw2 __MQSUBHSS (sw2, sw2)} 15654@tab @code{@var{c} = __MQSUBHSS (@var{a}, @var{b})} 15655@tab @code{MQSUBHSS @var{a},@var{b},@var{c}} 15656@item @code{uw2 __MQSUBHUS (uw2, uw2)} 15657@tab @code{@var{c} = __MQSUBHUS (@var{a}, @var{b})} 15658@tab @code{MQSUBHUS @var{a},@var{b},@var{c}} 15659@item @code{void __MQXMACHS (acc, sw2, sw2)} 15660@tab @code{__MQXMACHS (@var{c}, @var{a}, @var{b})} 15661@tab @code{MQXMACHS @var{a},@var{b},@var{c}} 15662@item @code{void __MQXMACXHS (acc, sw2, sw2)} 15663@tab @code{__MQXMACXHS (@var{c}, @var{a}, @var{b})} 15664@tab @code{MQXMACXHS @var{a},@var{b},@var{c}} 15665@item @code{uw1 __MRDACC (acc)} 15666@tab @code{@var{b} = __MRDACC (@var{a})} 15667@tab @code{MRDACC @var{a},@var{b}} 15668@item @code{uw1 __MRDACCG (acc)} 15669@tab @code{@var{b} = __MRDACCG (@var{a})} 15670@tab @code{MRDACCG @var{a},@var{b}} 15671@item @code{uw1 __MROTLI (uw1, const)} 15672@tab @code{@var{c} = __MROTLI (@var{a}, @var{b})} 15673@tab @code{MROTLI @var{a},#@var{b},@var{c}} 15674@item @code{uw1 __MROTRI (uw1, const)} 15675@tab @code{@var{c} = __MROTRI (@var{a}, @var{b})} 15676@tab @code{MROTRI @var{a},#@var{b},@var{c}} 15677@item @code{sw1 __MSATHS (sw1, sw1)} 15678@tab @code{@var{c} = __MSATHS (@var{a}, @var{b})} 15679@tab @code{MSATHS @var{a},@var{b},@var{c}} 15680@item @code{uw1 __MSATHU (uw1, uw1)} 15681@tab @code{@var{c} = __MSATHU (@var{a}, @var{b})} 15682@tab @code{MSATHU @var{a},@var{b},@var{c}} 15683@item @code{uw1 __MSLLHI (uw1, const)} 15684@tab @code{@var{c} = __MSLLHI (@var{a}, @var{b})} 15685@tab @code{MSLLHI @var{a},#@var{b},@var{c}} 15686@item @code{sw1 __MSRAHI (sw1, const)} 15687@tab @code{@var{c} = __MSRAHI (@var{a}, @var{b})} 15688@tab @code{MSRAHI @var{a},#@var{b},@var{c}} 15689@item @code{uw1 __MSRLHI (uw1, const)} 15690@tab @code{@var{c} = __MSRLHI (@var{a}, @var{b})} 15691@tab @code{MSRLHI @var{a},#@var{b},@var{c}} 15692@item @code{void __MSUBACCS (acc, acc)} 15693@tab @code{__MSUBACCS (@var{b}, @var{a})} 15694@tab @code{MSUBACCS @var{a},@var{b}} 15695@item @code{sw1 __MSUBHSS (sw1, sw1)} 15696@tab @code{@var{c} = __MSUBHSS (@var{a}, @var{b})} 15697@tab @code{MSUBHSS @var{a},@var{b},@var{c}} 15698@item @code{uw1 __MSUBHUS (uw1, uw1)} 15699@tab @code{@var{c} = __MSUBHUS (@var{a}, @var{b})} 15700@tab @code{MSUBHUS @var{a},@var{b},@var{c}} 15701@item @code{void __MTRAP (void)} 15702@tab @code{__MTRAP ()} 15703@tab @code{MTRAP} 15704@item @code{uw2 __MUNPACKH (uw1)} 15705@tab @code{@var{b} = __MUNPACKH (@var{a})} 15706@tab @code{MUNPACKH @var{a},@var{b}} 15707@item @code{uw1 __MWCUT (uw2, uw1)} 15708@tab @code{@var{c} = __MWCUT (@var{a}, @var{b})} 15709@tab @code{MWCUT @var{a},@var{b},@var{c}} 15710@item @code{void __MWTACC (acc, uw1)} 15711@tab @code{__MWTACC (@var{b}, @var{a})} 15712@tab @code{MWTACC @var{a},@var{b}} 15713@item @code{void __MWTACCG (acc, uw1)} 15714@tab @code{__MWTACCG (@var{b}, @var{a})} 15715@tab @code{MWTACCG @var{a},@var{b}} 15716@item @code{uw1 __MXOR (uw1, uw1)} 15717@tab @code{@var{c} = __MXOR (@var{a}, @var{b})} 15718@tab @code{MXOR @var{a},@var{b},@var{c}} 15719@end multitable 15720 15721@node Raw read/write Functions 15722@subsubsection Raw Read/Write Functions 15723 15724This sections describes built-in functions related to read and write 15725instructions to access memory. These functions generate 15726@code{membar} instructions to flush the I/O load and stores where 15727appropriate, as described in Fujitsu's manual described above. 15728 15729@table @code 15730 15731@item unsigned char __builtin_read8 (void *@var{data}) 15732@item unsigned short __builtin_read16 (void *@var{data}) 15733@item unsigned long __builtin_read32 (void *@var{data}) 15734@item unsigned long long __builtin_read64 (void *@var{data}) 15735 15736@item void __builtin_write8 (void *@var{data}, unsigned char @var{datum}) 15737@item void __builtin_write16 (void *@var{data}, unsigned short @var{datum}) 15738@item void __builtin_write32 (void *@var{data}, unsigned long @var{datum}) 15739@item void __builtin_write64 (void *@var{data}, unsigned long long @var{datum}) 15740@end table 15741 15742@node Other Built-in Functions 15743@subsubsection Other Built-in Functions 15744 15745This section describes built-in functions that are not named after 15746a specific FR-V instruction. 15747 15748@table @code 15749@item sw2 __IACCreadll (iacc @var{reg}) 15750Return the full 64-bit value of IACC0@. The @var{reg} argument is reserved 15751for future expansion and must be 0. 15752 15753@item sw1 __IACCreadl (iacc @var{reg}) 15754Return the value of IACC0H if @var{reg} is 0 and IACC0L if @var{reg} is 1. 15755Other values of @var{reg} are rejected as invalid. 15756 15757@item void __IACCsetll (iacc @var{reg}, sw2 @var{x}) 15758Set the full 64-bit value of IACC0 to @var{x}. The @var{reg} argument 15759is reserved for future expansion and must be 0. 15760 15761@item void __IACCsetl (iacc @var{reg}, sw1 @var{x}) 15762Set IACC0H to @var{x} if @var{reg} is 0 and IACC0L to @var{x} if @var{reg} 15763is 1. Other values of @var{reg} are rejected as invalid. 15764 15765@item void __data_prefetch0 (const void *@var{x}) 15766Use the @code{dcpl} instruction to load the contents of address @var{x} 15767into the data cache. 15768 15769@item void __data_prefetch (const void *@var{x}) 15770Use the @code{nldub} instruction to load the contents of address @var{x} 15771into the data cache. The instruction is issued in slot I1@. 15772@end table 15773 15774@node MIPS DSP Built-in Functions 15775@subsection MIPS DSP Built-in Functions 15776 15777The MIPS DSP Application-Specific Extension (ASE) includes new 15778instructions that are designed to improve the performance of DSP and 15779media applications. It provides instructions that operate on packed 157808-bit/16-bit integer data, Q7, Q15 and Q31 fractional data. 15781 15782GCC supports MIPS DSP operations using both the generic 15783vector extensions (@pxref{Vector Extensions}) and a collection of 15784MIPS-specific built-in functions. Both kinds of support are 15785enabled by the @option{-mdsp} command-line option. 15786 15787Revision 2 of the ASE was introduced in the second half of 2006. 15788This revision adds extra instructions to the original ASE, but is 15789otherwise backwards-compatible with it. You can select revision 2 15790using the command-line option @option{-mdspr2}; this option implies 15791@option{-mdsp}. 15792 15793The SCOUNT and POS bits of the DSP control register are global. The 15794WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and 15795POS bits. During optimization, the compiler does not delete these 15796instructions and it does not delete calls to functions containing 15797these instructions. 15798 15799At present, GCC only provides support for operations on 32-bit 15800vectors. The vector type associated with 8-bit integer data is 15801usually called @code{v4i8}, the vector type associated with Q7 15802is usually called @code{v4q7}, the vector type associated with 16-bit 15803integer data is usually called @code{v2i16}, and the vector type 15804associated with Q15 is usually called @code{v2q15}. They can be 15805defined in C as follows: 15806 15807@smallexample 15808typedef signed char v4i8 __attribute__ ((vector_size(4))); 15809typedef signed char v4q7 __attribute__ ((vector_size(4))); 15810typedef short v2i16 __attribute__ ((vector_size(4))); 15811typedef short v2q15 __attribute__ ((vector_size(4))); 15812@end smallexample 15813 15814@code{v4i8}, @code{v4q7}, @code{v2i16} and @code{v2q15} values are 15815initialized in the same way as aggregates. For example: 15816 15817@smallexample 15818v4i8 a = @{1, 2, 3, 4@}; 15819v4i8 b; 15820b = (v4i8) @{5, 6, 7, 8@}; 15821 15822v2q15 c = @{0x0fcb, 0x3a75@}; 15823v2q15 d; 15824d = (v2q15) @{0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15@}; 15825@end smallexample 15826 15827@emph{Note:} The CPU's endianness determines the order in which values 15828are packed. On little-endian targets, the first value is the least 15829significant and the last value is the most significant. The opposite 15830order applies to big-endian targets. For example, the code above 15831sets the lowest byte of @code{a} to @code{1} on little-endian targets 15832and @code{4} on big-endian targets. 15833 15834@emph{Note:} Q7, Q15 and Q31 values must be initialized with their integer 15835representation. As shown in this example, the integer representation 15836of a Q7 value can be obtained by multiplying the fractional value by 15837@code{0x1.0p7}. The equivalent for Q15 values is to multiply by 15838@code{0x1.0p15}. The equivalent for Q31 values is to multiply by 15839@code{0x1.0p31}. 15840 15841The table below lists the @code{v4i8} and @code{v2q15} operations for which 15842hardware support exists. @code{a} and @code{b} are @code{v4i8} values, 15843and @code{c} and @code{d} are @code{v2q15} values. 15844 15845@multitable @columnfractions .50 .50 15846@item C code @tab MIPS instruction 15847@item @code{a + b} @tab @code{addu.qb} 15848@item @code{c + d} @tab @code{addq.ph} 15849@item @code{a - b} @tab @code{subu.qb} 15850@item @code{c - d} @tab @code{subq.ph} 15851@end multitable 15852 15853The table below lists the @code{v2i16} operation for which 15854hardware support exists for the DSP ASE REV 2. @code{e} and @code{f} are 15855@code{v2i16} values. 15856 15857@multitable @columnfractions .50 .50 15858@item C code @tab MIPS instruction 15859@item @code{e * f} @tab @code{mul.ph} 15860@end multitable 15861 15862It is easier to describe the DSP built-in functions if we first define 15863the following types: 15864 15865@smallexample 15866typedef int q31; 15867typedef int i32; 15868typedef unsigned int ui32; 15869typedef long long a64; 15870@end smallexample 15871 15872@code{q31} and @code{i32} are actually the same as @code{int}, but we 15873use @code{q31} to indicate a Q31 fractional value and @code{i32} to 15874indicate a 32-bit integer value. Similarly, @code{a64} is the same as 15875@code{long long}, but we use @code{a64} to indicate values that are 15876placed in one of the four DSP accumulators (@code{$ac0}, 15877@code{$ac1}, @code{$ac2} or @code{$ac3}). 15878 15879Also, some built-in functions prefer or require immediate numbers as 15880parameters, because the corresponding DSP instructions accept both immediate 15881numbers and register operands, or accept immediate numbers only. The 15882immediate parameters are listed as follows. 15883 15884@smallexample 15885imm0_3: 0 to 3. 15886imm0_7: 0 to 7. 15887imm0_15: 0 to 15. 15888imm0_31: 0 to 31. 15889imm0_63: 0 to 63. 15890imm0_255: 0 to 255. 15891imm_n32_31: -32 to 31. 15892imm_n512_511: -512 to 511. 15893@end smallexample 15894 15895The following built-in functions map directly to a particular MIPS DSP 15896instruction. Please refer to the architecture specification 15897for details on what each instruction does. 15898 15899@smallexample 15900v2q15 __builtin_mips_addq_ph (v2q15, v2q15) 15901v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15) 15902q31 __builtin_mips_addq_s_w (q31, q31) 15903v4i8 __builtin_mips_addu_qb (v4i8, v4i8) 15904v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8) 15905v2q15 __builtin_mips_subq_ph (v2q15, v2q15) 15906v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15) 15907q31 __builtin_mips_subq_s_w (q31, q31) 15908v4i8 __builtin_mips_subu_qb (v4i8, v4i8) 15909v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8) 15910i32 __builtin_mips_addsc (i32, i32) 15911i32 __builtin_mips_addwc (i32, i32) 15912i32 __builtin_mips_modsub (i32, i32) 15913i32 __builtin_mips_raddu_w_qb (v4i8) 15914v2q15 __builtin_mips_absq_s_ph (v2q15) 15915q31 __builtin_mips_absq_s_w (q31) 15916v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15) 15917v2q15 __builtin_mips_precrq_ph_w (q31, q31) 15918v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31) 15919v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15) 15920q31 __builtin_mips_preceq_w_phl (v2q15) 15921q31 __builtin_mips_preceq_w_phr (v2q15) 15922v2q15 __builtin_mips_precequ_ph_qbl (v4i8) 15923v2q15 __builtin_mips_precequ_ph_qbr (v4i8) 15924v2q15 __builtin_mips_precequ_ph_qbla (v4i8) 15925v2q15 __builtin_mips_precequ_ph_qbra (v4i8) 15926v2q15 __builtin_mips_preceu_ph_qbl (v4i8) 15927v2q15 __builtin_mips_preceu_ph_qbr (v4i8) 15928v2q15 __builtin_mips_preceu_ph_qbla (v4i8) 15929v2q15 __builtin_mips_preceu_ph_qbra (v4i8) 15930v4i8 __builtin_mips_shll_qb (v4i8, imm0_7) 15931v4i8 __builtin_mips_shll_qb (v4i8, i32) 15932v2q15 __builtin_mips_shll_ph (v2q15, imm0_15) 15933v2q15 __builtin_mips_shll_ph (v2q15, i32) 15934v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15) 15935v2q15 __builtin_mips_shll_s_ph (v2q15, i32) 15936q31 __builtin_mips_shll_s_w (q31, imm0_31) 15937q31 __builtin_mips_shll_s_w (q31, i32) 15938v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7) 15939v4i8 __builtin_mips_shrl_qb (v4i8, i32) 15940v2q15 __builtin_mips_shra_ph (v2q15, imm0_15) 15941v2q15 __builtin_mips_shra_ph (v2q15, i32) 15942v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15) 15943v2q15 __builtin_mips_shra_r_ph (v2q15, i32) 15944q31 __builtin_mips_shra_r_w (q31, imm0_31) 15945q31 __builtin_mips_shra_r_w (q31, i32) 15946v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15) 15947v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15) 15948v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15) 15949q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15) 15950q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15) 15951a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8) 15952a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8) 15953a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8) 15954a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8) 15955a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15) 15956a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31) 15957a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15) 15958a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31) 15959a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15) 15960a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15) 15961a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15) 15962a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15) 15963a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15) 15964i32 __builtin_mips_bitrev (i32) 15965i32 __builtin_mips_insv (i32, i32) 15966v4i8 __builtin_mips_repl_qb (imm0_255) 15967v4i8 __builtin_mips_repl_qb (i32) 15968v2q15 __builtin_mips_repl_ph (imm_n512_511) 15969v2q15 __builtin_mips_repl_ph (i32) 15970void __builtin_mips_cmpu_eq_qb (v4i8, v4i8) 15971void __builtin_mips_cmpu_lt_qb (v4i8, v4i8) 15972void __builtin_mips_cmpu_le_qb (v4i8, v4i8) 15973i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8) 15974i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8) 15975i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8) 15976void __builtin_mips_cmp_eq_ph (v2q15, v2q15) 15977void __builtin_mips_cmp_lt_ph (v2q15, v2q15) 15978void __builtin_mips_cmp_le_ph (v2q15, v2q15) 15979v4i8 __builtin_mips_pick_qb (v4i8, v4i8) 15980v2q15 __builtin_mips_pick_ph (v2q15, v2q15) 15981v2q15 __builtin_mips_packrl_ph (v2q15, v2q15) 15982i32 __builtin_mips_extr_w (a64, imm0_31) 15983i32 __builtin_mips_extr_w (a64, i32) 15984i32 __builtin_mips_extr_r_w (a64, imm0_31) 15985i32 __builtin_mips_extr_s_h (a64, i32) 15986i32 __builtin_mips_extr_rs_w (a64, imm0_31) 15987i32 __builtin_mips_extr_rs_w (a64, i32) 15988i32 __builtin_mips_extr_s_h (a64, imm0_31) 15989i32 __builtin_mips_extr_r_w (a64, i32) 15990i32 __builtin_mips_extp (a64, imm0_31) 15991i32 __builtin_mips_extp (a64, i32) 15992i32 __builtin_mips_extpdp (a64, imm0_31) 15993i32 __builtin_mips_extpdp (a64, i32) 15994a64 __builtin_mips_shilo (a64, imm_n32_31) 15995a64 __builtin_mips_shilo (a64, i32) 15996a64 __builtin_mips_mthlip (a64, i32) 15997void __builtin_mips_wrdsp (i32, imm0_63) 15998i32 __builtin_mips_rddsp (imm0_63) 15999i32 __builtin_mips_lbux (void *, i32) 16000i32 __builtin_mips_lhx (void *, i32) 16001i32 __builtin_mips_lwx (void *, i32) 16002a64 __builtin_mips_ldx (void *, i32) [MIPS64 only] 16003i32 __builtin_mips_bposge32 (void) 16004a64 __builtin_mips_madd (a64, i32, i32); 16005a64 __builtin_mips_maddu (a64, ui32, ui32); 16006a64 __builtin_mips_msub (a64, i32, i32); 16007a64 __builtin_mips_msubu (a64, ui32, ui32); 16008a64 __builtin_mips_mult (i32, i32); 16009a64 __builtin_mips_multu (ui32, ui32); 16010@end smallexample 16011 16012The following built-in functions map directly to a particular MIPS DSP REV 2 16013instruction. Please refer to the architecture specification 16014for details on what each instruction does. 16015 16016@smallexample 16017v4q7 __builtin_mips_absq_s_qb (v4q7); 16018v2i16 __builtin_mips_addu_ph (v2i16, v2i16); 16019v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16); 16020v4i8 __builtin_mips_adduh_qb (v4i8, v4i8); 16021v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8); 16022i32 __builtin_mips_append (i32, i32, imm0_31); 16023i32 __builtin_mips_balign (i32, i32, imm0_3); 16024i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8); 16025i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8); 16026i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8); 16027a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16); 16028a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16); 16029v2i16 __builtin_mips_mul_ph (v2i16, v2i16); 16030v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16); 16031q31 __builtin_mips_mulq_rs_w (q31, q31); 16032v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15); 16033q31 __builtin_mips_mulq_s_w (q31, q31); 16034a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16); 16035v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16); 16036v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31); 16037v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31); 16038i32 __builtin_mips_prepend (i32, i32, imm0_31); 16039v4i8 __builtin_mips_shra_qb (v4i8, imm0_7); 16040v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7); 16041v4i8 __builtin_mips_shra_qb (v4i8, i32); 16042v4i8 __builtin_mips_shra_r_qb (v4i8, i32); 16043v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15); 16044v2i16 __builtin_mips_shrl_ph (v2i16, i32); 16045v2i16 __builtin_mips_subu_ph (v2i16, v2i16); 16046v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16); 16047v4i8 __builtin_mips_subuh_qb (v4i8, v4i8); 16048v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8); 16049v2q15 __builtin_mips_addqh_ph (v2q15, v2q15); 16050v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15); 16051q31 __builtin_mips_addqh_w (q31, q31); 16052q31 __builtin_mips_addqh_r_w (q31, q31); 16053v2q15 __builtin_mips_subqh_ph (v2q15, v2q15); 16054v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15); 16055q31 __builtin_mips_subqh_w (q31, q31); 16056q31 __builtin_mips_subqh_r_w (q31, q31); 16057a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16); 16058a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16); 16059a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15); 16060a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15); 16061a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15); 16062a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15); 16063@end smallexample 16064 16065 16066@node MIPS Paired-Single Support 16067@subsection MIPS Paired-Single Support 16068 16069The MIPS64 architecture includes a number of instructions that 16070operate on pairs of single-precision floating-point values. 16071Each pair is packed into a 64-bit floating-point register, 16072with one element being designated the ``upper half'' and 16073the other being designated the ``lower half''. 16074 16075GCC supports paired-single operations using both the generic 16076vector extensions (@pxref{Vector Extensions}) and a collection of 16077MIPS-specific built-in functions. Both kinds of support are 16078enabled by the @option{-mpaired-single} command-line option. 16079 16080The vector type associated with paired-single values is usually 16081called @code{v2sf}. It can be defined in C as follows: 16082 16083@smallexample 16084typedef float v2sf __attribute__ ((vector_size (8))); 16085@end smallexample 16086 16087@code{v2sf} values are initialized in the same way as aggregates. 16088For example: 16089 16090@smallexample 16091v2sf a = @{1.5, 9.1@}; 16092v2sf b; 16093float e, f; 16094b = (v2sf) @{e, f@}; 16095@end smallexample 16096 16097@emph{Note:} The CPU's endianness determines which value is stored in 16098the upper half of a register and which value is stored in the lower half. 16099On little-endian targets, the first value is the lower one and the second 16100value is the upper one. The opposite order applies to big-endian targets. 16101For example, the code above sets the lower half of @code{a} to 16102@code{1.5} on little-endian targets and @code{9.1} on big-endian targets. 16103 16104@node MIPS Loongson Built-in Functions 16105@subsection MIPS Loongson Built-in Functions 16106 16107GCC provides intrinsics to access the SIMD instructions provided by the 16108ST Microelectronics Loongson-2E and -2F processors. These intrinsics, 16109available after inclusion of the @code{loongson.h} header file, 16110operate on the following 64-bit vector types: 16111 16112@itemize 16113@item @code{uint8x8_t}, a vector of eight unsigned 8-bit integers; 16114@item @code{uint16x4_t}, a vector of four unsigned 16-bit integers; 16115@item @code{uint32x2_t}, a vector of two unsigned 32-bit integers; 16116@item @code{int8x8_t}, a vector of eight signed 8-bit integers; 16117@item @code{int16x4_t}, a vector of four signed 16-bit integers; 16118@item @code{int32x2_t}, a vector of two signed 32-bit integers. 16119@end itemize 16120 16121The intrinsics provided are listed below; each is named after the 16122machine instruction to which it corresponds, with suffixes added as 16123appropriate to distinguish intrinsics that expand to the same machine 16124instruction yet have different argument types. Refer to the architecture 16125documentation for a description of the functionality of each 16126instruction. 16127 16128@smallexample 16129int16x4_t packsswh (int32x2_t s, int32x2_t t); 16130int8x8_t packsshb (int16x4_t s, int16x4_t t); 16131uint8x8_t packushb (uint16x4_t s, uint16x4_t t); 16132uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t); 16133uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t); 16134uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t); 16135int32x2_t paddw_s (int32x2_t s, int32x2_t t); 16136int16x4_t paddh_s (int16x4_t s, int16x4_t t); 16137int8x8_t paddb_s (int8x8_t s, int8x8_t t); 16138uint64_t paddd_u (uint64_t s, uint64_t t); 16139int64_t paddd_s (int64_t s, int64_t t); 16140int16x4_t paddsh (int16x4_t s, int16x4_t t); 16141int8x8_t paddsb (int8x8_t s, int8x8_t t); 16142uint16x4_t paddush (uint16x4_t s, uint16x4_t t); 16143uint8x8_t paddusb (uint8x8_t s, uint8x8_t t); 16144uint64_t pandn_ud (uint64_t s, uint64_t t); 16145uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t); 16146uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t); 16147uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t); 16148int64_t pandn_sd (int64_t s, int64_t t); 16149int32x2_t pandn_sw (int32x2_t s, int32x2_t t); 16150int16x4_t pandn_sh (int16x4_t s, int16x4_t t); 16151int8x8_t pandn_sb (int8x8_t s, int8x8_t t); 16152uint16x4_t pavgh (uint16x4_t s, uint16x4_t t); 16153uint8x8_t pavgb (uint8x8_t s, uint8x8_t t); 16154uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t); 16155uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t); 16156uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t); 16157int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t); 16158int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t); 16159int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t); 16160uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t); 16161uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t); 16162uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t); 16163int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t); 16164int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t); 16165int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t); 16166uint16x4_t pextrh_u (uint16x4_t s, int field); 16167int16x4_t pextrh_s (int16x4_t s, int field); 16168uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t); 16169uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t); 16170uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t); 16171uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t); 16172int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t); 16173int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t); 16174int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t); 16175int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t); 16176int32x2_t pmaddhw (int16x4_t s, int16x4_t t); 16177int16x4_t pmaxsh (int16x4_t s, int16x4_t t); 16178uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t); 16179int16x4_t pminsh (int16x4_t s, int16x4_t t); 16180uint8x8_t pminub (uint8x8_t s, uint8x8_t t); 16181uint8x8_t pmovmskb_u (uint8x8_t s); 16182int8x8_t pmovmskb_s (int8x8_t s); 16183uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t); 16184int16x4_t pmulhh (int16x4_t s, int16x4_t t); 16185int16x4_t pmullh (int16x4_t s, int16x4_t t); 16186int64_t pmuluw (uint32x2_t s, uint32x2_t t); 16187uint8x8_t pasubub (uint8x8_t s, uint8x8_t t); 16188uint16x4_t biadd (uint8x8_t s); 16189uint16x4_t psadbh (uint8x8_t s, uint8x8_t t); 16190uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order); 16191int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order); 16192uint16x4_t psllh_u (uint16x4_t s, uint8_t amount); 16193int16x4_t psllh_s (int16x4_t s, uint8_t amount); 16194uint32x2_t psllw_u (uint32x2_t s, uint8_t amount); 16195int32x2_t psllw_s (int32x2_t s, uint8_t amount); 16196uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount); 16197int16x4_t psrlh_s (int16x4_t s, uint8_t amount); 16198uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount); 16199int32x2_t psrlw_s (int32x2_t s, uint8_t amount); 16200uint16x4_t psrah_u (uint16x4_t s, uint8_t amount); 16201int16x4_t psrah_s (int16x4_t s, uint8_t amount); 16202uint32x2_t psraw_u (uint32x2_t s, uint8_t amount); 16203int32x2_t psraw_s (int32x2_t s, uint8_t amount); 16204uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t); 16205uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t); 16206uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t); 16207int32x2_t psubw_s (int32x2_t s, int32x2_t t); 16208int16x4_t psubh_s (int16x4_t s, int16x4_t t); 16209int8x8_t psubb_s (int8x8_t s, int8x8_t t); 16210uint64_t psubd_u (uint64_t s, uint64_t t); 16211int64_t psubd_s (int64_t s, int64_t t); 16212int16x4_t psubsh (int16x4_t s, int16x4_t t); 16213int8x8_t psubsb (int8x8_t s, int8x8_t t); 16214uint16x4_t psubush (uint16x4_t s, uint16x4_t t); 16215uint8x8_t psubusb (uint8x8_t s, uint8x8_t t); 16216uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t); 16217uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t); 16218uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t); 16219int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t); 16220int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t); 16221int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t); 16222uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t); 16223uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t); 16224uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t); 16225int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t); 16226int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t); 16227int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t); 16228@end smallexample 16229 16230@menu 16231* Paired-Single Arithmetic:: 16232* Paired-Single Built-in Functions:: 16233* MIPS-3D Built-in Functions:: 16234@end menu 16235 16236@node Paired-Single Arithmetic 16237@subsubsection Paired-Single Arithmetic 16238 16239The table below lists the @code{v2sf} operations for which hardware 16240support exists. @code{a}, @code{b} and @code{c} are @code{v2sf} 16241values and @code{x} is an integral value. 16242 16243@multitable @columnfractions .50 .50 16244@item C code @tab MIPS instruction 16245@item @code{a + b} @tab @code{add.ps} 16246@item @code{a - b} @tab @code{sub.ps} 16247@item @code{-a} @tab @code{neg.ps} 16248@item @code{a * b} @tab @code{mul.ps} 16249@item @code{a * b + c} @tab @code{madd.ps} 16250@item @code{a * b - c} @tab @code{msub.ps} 16251@item @code{-(a * b + c)} @tab @code{nmadd.ps} 16252@item @code{-(a * b - c)} @tab @code{nmsub.ps} 16253@item @code{x ? a : b} @tab @code{movn.ps}/@code{movz.ps} 16254@end multitable 16255 16256Note that the multiply-accumulate instructions can be disabled 16257using the command-line option @code{-mno-fused-madd}. 16258 16259@node Paired-Single Built-in Functions 16260@subsubsection Paired-Single Built-in Functions 16261 16262The following paired-single functions map directly to a particular 16263MIPS instruction. Please refer to the architecture specification 16264for details on what each instruction does. 16265 16266@table @code 16267@item v2sf __builtin_mips_pll_ps (v2sf, v2sf) 16268Pair lower lower (@code{pll.ps}). 16269 16270@item v2sf __builtin_mips_pul_ps (v2sf, v2sf) 16271Pair upper lower (@code{pul.ps}). 16272 16273@item v2sf __builtin_mips_plu_ps (v2sf, v2sf) 16274Pair lower upper (@code{plu.ps}). 16275 16276@item v2sf __builtin_mips_puu_ps (v2sf, v2sf) 16277Pair upper upper (@code{puu.ps}). 16278 16279@item v2sf __builtin_mips_cvt_ps_s (float, float) 16280Convert pair to paired single (@code{cvt.ps.s}). 16281 16282@item float __builtin_mips_cvt_s_pl (v2sf) 16283Convert pair lower to single (@code{cvt.s.pl}). 16284 16285@item float __builtin_mips_cvt_s_pu (v2sf) 16286Convert pair upper to single (@code{cvt.s.pu}). 16287 16288@item v2sf __builtin_mips_abs_ps (v2sf) 16289Absolute value (@code{abs.ps}). 16290 16291@item v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int) 16292Align variable (@code{alnv.ps}). 16293 16294@emph{Note:} The value of the third parameter must be 0 or 4 16295modulo 8, otherwise the result is unpredictable. Please read the 16296instruction description for details. 16297@end table 16298 16299The following multi-instruction functions are also available. 16300In each case, @var{cond} can be any of the 16 floating-point conditions: 16301@code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult}, 16302@code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, @code{ngl}, 16303@code{lt}, @code{nge}, @code{le} or @code{ngt}. 16304 16305@table @code 16306@item v2sf __builtin_mips_movt_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16307@itemx v2sf __builtin_mips_movf_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16308Conditional move based on floating-point comparison (@code{c.@var{cond}.ps}, 16309@code{movt.ps}/@code{movf.ps}). 16310 16311The @code{movt} functions return the value @var{x} computed by: 16312 16313@smallexample 16314c.@var{cond}.ps @var{cc},@var{a},@var{b} 16315mov.ps @var{x},@var{c} 16316movt.ps @var{x},@var{d},@var{cc} 16317@end smallexample 16318 16319The @code{movf} functions are similar but use @code{movf.ps} instead 16320of @code{movt.ps}. 16321 16322@item int __builtin_mips_upper_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16323@itemx int __builtin_mips_lower_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16324Comparison of two paired-single values (@code{c.@var{cond}.ps}, 16325@code{bc1t}/@code{bc1f}). 16326 16327These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps} 16328and return either the upper or lower half of the result. For example: 16329 16330@smallexample 16331v2sf a, b; 16332if (__builtin_mips_upper_c_eq_ps (a, b)) 16333 upper_halves_are_equal (); 16334else 16335 upper_halves_are_unequal (); 16336 16337if (__builtin_mips_lower_c_eq_ps (a, b)) 16338 lower_halves_are_equal (); 16339else 16340 lower_halves_are_unequal (); 16341@end smallexample 16342@end table 16343 16344@node MIPS-3D Built-in Functions 16345@subsubsection MIPS-3D Built-in Functions 16346 16347The MIPS-3D Application-Specific Extension (ASE) includes additional 16348paired-single instructions that are designed to improve the performance 16349of 3D graphics operations. Support for these instructions is controlled 16350by the @option{-mips3d} command-line option. 16351 16352The functions listed below map directly to a particular MIPS-3D 16353instruction. Please refer to the architecture specification for 16354more details on what each instruction does. 16355 16356@table @code 16357@item v2sf __builtin_mips_addr_ps (v2sf, v2sf) 16358Reduction add (@code{addr.ps}). 16359 16360@item v2sf __builtin_mips_mulr_ps (v2sf, v2sf) 16361Reduction multiply (@code{mulr.ps}). 16362 16363@item v2sf __builtin_mips_cvt_pw_ps (v2sf) 16364Convert paired single to paired word (@code{cvt.pw.ps}). 16365 16366@item v2sf __builtin_mips_cvt_ps_pw (v2sf) 16367Convert paired word to paired single (@code{cvt.ps.pw}). 16368 16369@item float __builtin_mips_recip1_s (float) 16370@itemx double __builtin_mips_recip1_d (double) 16371@itemx v2sf __builtin_mips_recip1_ps (v2sf) 16372Reduced-precision reciprocal (sequence step 1) (@code{recip1.@var{fmt}}). 16373 16374@item float __builtin_mips_recip2_s (float, float) 16375@itemx double __builtin_mips_recip2_d (double, double) 16376@itemx v2sf __builtin_mips_recip2_ps (v2sf, v2sf) 16377Reduced-precision reciprocal (sequence step 2) (@code{recip2.@var{fmt}}). 16378 16379@item float __builtin_mips_rsqrt1_s (float) 16380@itemx double __builtin_mips_rsqrt1_d (double) 16381@itemx v2sf __builtin_mips_rsqrt1_ps (v2sf) 16382Reduced-precision reciprocal square root (sequence step 1) 16383(@code{rsqrt1.@var{fmt}}). 16384 16385@item float __builtin_mips_rsqrt2_s (float, float) 16386@itemx double __builtin_mips_rsqrt2_d (double, double) 16387@itemx v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf) 16388Reduced-precision reciprocal square root (sequence step 2) 16389(@code{rsqrt2.@var{fmt}}). 16390@end table 16391 16392The following multi-instruction functions are also available. 16393In each case, @var{cond} can be any of the 16 floating-point conditions: 16394@code{f}, @code{un}, @code{eq}, @code{ueq}, @code{olt}, @code{ult}, 16395@code{ole}, @code{ule}, @code{sf}, @code{ngle}, @code{seq}, 16396@code{ngl}, @code{lt}, @code{nge}, @code{le} or @code{ngt}. 16397 16398@table @code 16399@item int __builtin_mips_cabs_@var{cond}_s (float @var{a}, float @var{b}) 16400@itemx int __builtin_mips_cabs_@var{cond}_d (double @var{a}, double @var{b}) 16401Absolute comparison of two scalar values (@code{cabs.@var{cond}.@var{fmt}}, 16402@code{bc1t}/@code{bc1f}). 16403 16404These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.s} 16405or @code{cabs.@var{cond}.d} and return the result as a boolean value. 16406For example: 16407 16408@smallexample 16409float a, b; 16410if (__builtin_mips_cabs_eq_s (a, b)) 16411 true (); 16412else 16413 false (); 16414@end smallexample 16415 16416@item int __builtin_mips_upper_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16417@itemx int __builtin_mips_lower_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16418Absolute comparison of two paired-single values (@code{cabs.@var{cond}.ps}, 16419@code{bc1t}/@code{bc1f}). 16420 16421These functions compare @var{a} and @var{b} using @code{cabs.@var{cond}.ps} 16422and return either the upper or lower half of the result. For example: 16423 16424@smallexample 16425v2sf a, b; 16426if (__builtin_mips_upper_cabs_eq_ps (a, b)) 16427 upper_halves_are_equal (); 16428else 16429 upper_halves_are_unequal (); 16430 16431if (__builtin_mips_lower_cabs_eq_ps (a, b)) 16432 lower_halves_are_equal (); 16433else 16434 lower_halves_are_unequal (); 16435@end smallexample 16436 16437@item v2sf __builtin_mips_movt_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16438@itemx v2sf __builtin_mips_movf_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16439Conditional move based on absolute comparison (@code{cabs.@var{cond}.ps}, 16440@code{movt.ps}/@code{movf.ps}). 16441 16442The @code{movt} functions return the value @var{x} computed by: 16443 16444@smallexample 16445cabs.@var{cond}.ps @var{cc},@var{a},@var{b} 16446mov.ps @var{x},@var{c} 16447movt.ps @var{x},@var{d},@var{cc} 16448@end smallexample 16449 16450The @code{movf} functions are similar but use @code{movf.ps} instead 16451of @code{movt.ps}. 16452 16453@item int __builtin_mips_any_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16454@itemx int __builtin_mips_all_c_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16455@itemx int __builtin_mips_any_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16456@itemx int __builtin_mips_all_cabs_@var{cond}_ps (v2sf @var{a}, v2sf @var{b}) 16457Comparison of two paired-single values 16458(@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps}, 16459@code{bc1any2t}/@code{bc1any2f}). 16460 16461These functions compare @var{a} and @var{b} using @code{c.@var{cond}.ps} 16462or @code{cabs.@var{cond}.ps}. The @code{any} forms return @code{true} if either 16463result is @code{true} and the @code{all} forms return @code{true} if both results are @code{true}. 16464For example: 16465 16466@smallexample 16467v2sf a, b; 16468if (__builtin_mips_any_c_eq_ps (a, b)) 16469 one_is_true (); 16470else 16471 both_are_false (); 16472 16473if (__builtin_mips_all_c_eq_ps (a, b)) 16474 both_are_true (); 16475else 16476 one_is_false (); 16477@end smallexample 16478 16479@item int __builtin_mips_any_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16480@itemx int __builtin_mips_all_c_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16481@itemx int __builtin_mips_any_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16482@itemx int __builtin_mips_all_cabs_@var{cond}_4s (v2sf @var{a}, v2sf @var{b}, v2sf @var{c}, v2sf @var{d}) 16483Comparison of four paired-single values 16484(@code{c.@var{cond}.ps}/@code{cabs.@var{cond}.ps}, 16485@code{bc1any4t}/@code{bc1any4f}). 16486 16487These functions use @code{c.@var{cond}.ps} or @code{cabs.@var{cond}.ps} 16488to compare @var{a} with @var{b} and to compare @var{c} with @var{d}. 16489The @code{any} forms return @code{true} if any of the four results are @code{true} 16490and the @code{all} forms return @code{true} if all four results are @code{true}. 16491For example: 16492 16493@smallexample 16494v2sf a, b, c, d; 16495if (__builtin_mips_any_c_eq_4s (a, b, c, d)) 16496 some_are_true (); 16497else 16498 all_are_false (); 16499 16500if (__builtin_mips_all_c_eq_4s (a, b, c, d)) 16501 all_are_true (); 16502else 16503 some_are_false (); 16504@end smallexample 16505@end table 16506 16507@node MIPS SIMD Architecture (MSA) Support 16508@subsection MIPS SIMD Architecture (MSA) Support 16509 16510@menu 16511* MIPS SIMD Architecture Built-in Functions:: 16512@end menu 16513 16514GCC provides intrinsics to access the SIMD instructions provided by the 16515MSA MIPS SIMD Architecture. The interface is made available by including 16516@code{<msa.h>} and using @option{-mmsa -mhard-float -mfp64 -mnan=2008}. 16517For each @code{__builtin_msa_*}, there is a shortened name of the intrinsic, 16518@code{__msa_*}. 16519 16520MSA implements 128-bit wide vector registers, operating on 8-, 16-, 32- and 1652164-bit integer, 16- and 32-bit fixed-point, or 32- and 64-bit floating point 16522data elements. The following vectors typedefs are included in @code{msa.h}: 16523@itemize 16524@item @code{v16i8}, a vector of sixteen signed 8-bit integers; 16525@item @code{v16u8}, a vector of sixteen unsigned 8-bit integers; 16526@item @code{v8i16}, a vector of eight signed 16-bit integers; 16527@item @code{v8u16}, a vector of eight unsigned 16-bit integers; 16528@item @code{v4i32}, a vector of four signed 32-bit integers; 16529@item @code{v4u32}, a vector of four unsigned 32-bit integers; 16530@item @code{v2i64}, a vector of two signed 64-bit integers; 16531@item @code{v2u64}, a vector of two unsigned 64-bit integers; 16532@item @code{v4f32}, a vector of four 32-bit floats; 16533@item @code{v2f64}, a vector of two 64-bit doubles. 16534@end itemize 16535 16536Instructions and corresponding built-ins may have additional restrictions and/or 16537input/output values manipulated: 16538@itemize 16539@item @code{imm0_1}, an integer literal in range 0 to 1; 16540@item @code{imm0_3}, an integer literal in range 0 to 3; 16541@item @code{imm0_7}, an integer literal in range 0 to 7; 16542@item @code{imm0_15}, an integer literal in range 0 to 15; 16543@item @code{imm0_31}, an integer literal in range 0 to 31; 16544@item @code{imm0_63}, an integer literal in range 0 to 63; 16545@item @code{imm0_255}, an integer literal in range 0 to 255; 16546@item @code{imm_n16_15}, an integer literal in range -16 to 15; 16547@item @code{imm_n512_511}, an integer literal in range -512 to 511; 16548@item @code{imm_n1024_1022}, an integer literal in range -512 to 511 left 16549shifted by 1 bit, i.e., -1024, -1022, @dots{}, 1020, 1022; 16550@item @code{imm_n2048_2044}, an integer literal in range -512 to 511 left 16551shifted by 2 bits, i.e., -2048, -2044, @dots{}, 2040, 2044; 16552@item @code{imm_n4096_4088}, an integer literal in range -512 to 511 left 16553shifted by 3 bits, i.e., -4096, -4088, @dots{}, 4080, 4088; 16554@item @code{imm1_4}, an integer literal in range 1 to 4; 16555@item @code{i32, i64, u32, u64, f32, f64}, defined as follows: 16556@end itemize 16557 16558@smallexample 16559@{ 16560typedef int i32; 16561#if __LONG_MAX__ == __LONG_LONG_MAX__ 16562typedef long i64; 16563#else 16564typedef long long i64; 16565#endif 16566 16567typedef unsigned int u32; 16568#if __LONG_MAX__ == __LONG_LONG_MAX__ 16569typedef unsigned long u64; 16570#else 16571typedef unsigned long long u64; 16572#endif 16573 16574typedef double f64; 16575typedef float f32; 16576@} 16577@end smallexample 16578 16579@node MIPS SIMD Architecture Built-in Functions 16580@subsubsection MIPS SIMD Architecture Built-in Functions 16581 16582The intrinsics provided are listed below; each is named after the 16583machine instruction. 16584 16585@smallexample 16586v16i8 __builtin_msa_add_a_b (v16i8, v16i8); 16587v8i16 __builtin_msa_add_a_h (v8i16, v8i16); 16588v4i32 __builtin_msa_add_a_w (v4i32, v4i32); 16589v2i64 __builtin_msa_add_a_d (v2i64, v2i64); 16590 16591v16i8 __builtin_msa_adds_a_b (v16i8, v16i8); 16592v8i16 __builtin_msa_adds_a_h (v8i16, v8i16); 16593v4i32 __builtin_msa_adds_a_w (v4i32, v4i32); 16594v2i64 __builtin_msa_adds_a_d (v2i64, v2i64); 16595 16596v16i8 __builtin_msa_adds_s_b (v16i8, v16i8); 16597v8i16 __builtin_msa_adds_s_h (v8i16, v8i16); 16598v4i32 __builtin_msa_adds_s_w (v4i32, v4i32); 16599v2i64 __builtin_msa_adds_s_d (v2i64, v2i64); 16600 16601v16u8 __builtin_msa_adds_u_b (v16u8, v16u8); 16602v8u16 __builtin_msa_adds_u_h (v8u16, v8u16); 16603v4u32 __builtin_msa_adds_u_w (v4u32, v4u32); 16604v2u64 __builtin_msa_adds_u_d (v2u64, v2u64); 16605 16606v16i8 __builtin_msa_addv_b (v16i8, v16i8); 16607v8i16 __builtin_msa_addv_h (v8i16, v8i16); 16608v4i32 __builtin_msa_addv_w (v4i32, v4i32); 16609v2i64 __builtin_msa_addv_d (v2i64, v2i64); 16610 16611v16i8 __builtin_msa_addvi_b (v16i8, imm0_31); 16612v8i16 __builtin_msa_addvi_h (v8i16, imm0_31); 16613v4i32 __builtin_msa_addvi_w (v4i32, imm0_31); 16614v2i64 __builtin_msa_addvi_d (v2i64, imm0_31); 16615 16616v16u8 __builtin_msa_and_v (v16u8, v16u8); 16617 16618v16u8 __builtin_msa_andi_b (v16u8, imm0_255); 16619 16620v16i8 __builtin_msa_asub_s_b (v16i8, v16i8); 16621v8i16 __builtin_msa_asub_s_h (v8i16, v8i16); 16622v4i32 __builtin_msa_asub_s_w (v4i32, v4i32); 16623v2i64 __builtin_msa_asub_s_d (v2i64, v2i64); 16624 16625v16u8 __builtin_msa_asub_u_b (v16u8, v16u8); 16626v8u16 __builtin_msa_asub_u_h (v8u16, v8u16); 16627v4u32 __builtin_msa_asub_u_w (v4u32, v4u32); 16628v2u64 __builtin_msa_asub_u_d (v2u64, v2u64); 16629 16630v16i8 __builtin_msa_ave_s_b (v16i8, v16i8); 16631v8i16 __builtin_msa_ave_s_h (v8i16, v8i16); 16632v4i32 __builtin_msa_ave_s_w (v4i32, v4i32); 16633v2i64 __builtin_msa_ave_s_d (v2i64, v2i64); 16634 16635v16u8 __builtin_msa_ave_u_b (v16u8, v16u8); 16636v8u16 __builtin_msa_ave_u_h (v8u16, v8u16); 16637v4u32 __builtin_msa_ave_u_w (v4u32, v4u32); 16638v2u64 __builtin_msa_ave_u_d (v2u64, v2u64); 16639 16640v16i8 __builtin_msa_aver_s_b (v16i8, v16i8); 16641v8i16 __builtin_msa_aver_s_h (v8i16, v8i16); 16642v4i32 __builtin_msa_aver_s_w (v4i32, v4i32); 16643v2i64 __builtin_msa_aver_s_d (v2i64, v2i64); 16644 16645v16u8 __builtin_msa_aver_u_b (v16u8, v16u8); 16646v8u16 __builtin_msa_aver_u_h (v8u16, v8u16); 16647v4u32 __builtin_msa_aver_u_w (v4u32, v4u32); 16648v2u64 __builtin_msa_aver_u_d (v2u64, v2u64); 16649 16650v16u8 __builtin_msa_bclr_b (v16u8, v16u8); 16651v8u16 __builtin_msa_bclr_h (v8u16, v8u16); 16652v4u32 __builtin_msa_bclr_w (v4u32, v4u32); 16653v2u64 __builtin_msa_bclr_d (v2u64, v2u64); 16654 16655v16u8 __builtin_msa_bclri_b (v16u8, imm0_7); 16656v8u16 __builtin_msa_bclri_h (v8u16, imm0_15); 16657v4u32 __builtin_msa_bclri_w (v4u32, imm0_31); 16658v2u64 __builtin_msa_bclri_d (v2u64, imm0_63); 16659 16660v16u8 __builtin_msa_binsl_b (v16u8, v16u8, v16u8); 16661v8u16 __builtin_msa_binsl_h (v8u16, v8u16, v8u16); 16662v4u32 __builtin_msa_binsl_w (v4u32, v4u32, v4u32); 16663v2u64 __builtin_msa_binsl_d (v2u64, v2u64, v2u64); 16664 16665v16u8 __builtin_msa_binsli_b (v16u8, v16u8, imm0_7); 16666v8u16 __builtin_msa_binsli_h (v8u16, v8u16, imm0_15); 16667v4u32 __builtin_msa_binsli_w (v4u32, v4u32, imm0_31); 16668v2u64 __builtin_msa_binsli_d (v2u64, v2u64, imm0_63); 16669 16670v16u8 __builtin_msa_binsr_b (v16u8, v16u8, v16u8); 16671v8u16 __builtin_msa_binsr_h (v8u16, v8u16, v8u16); 16672v4u32 __builtin_msa_binsr_w (v4u32, v4u32, v4u32); 16673v2u64 __builtin_msa_binsr_d (v2u64, v2u64, v2u64); 16674 16675v16u8 __builtin_msa_binsri_b (v16u8, v16u8, imm0_7); 16676v8u16 __builtin_msa_binsri_h (v8u16, v8u16, imm0_15); 16677v4u32 __builtin_msa_binsri_w (v4u32, v4u32, imm0_31); 16678v2u64 __builtin_msa_binsri_d (v2u64, v2u64, imm0_63); 16679 16680v16u8 __builtin_msa_bmnz_v (v16u8, v16u8, v16u8); 16681 16682v16u8 __builtin_msa_bmnzi_b (v16u8, v16u8, imm0_255); 16683 16684v16u8 __builtin_msa_bmz_v (v16u8, v16u8, v16u8); 16685 16686v16u8 __builtin_msa_bmzi_b (v16u8, v16u8, imm0_255); 16687 16688v16u8 __builtin_msa_bneg_b (v16u8, v16u8); 16689v8u16 __builtin_msa_bneg_h (v8u16, v8u16); 16690v4u32 __builtin_msa_bneg_w (v4u32, v4u32); 16691v2u64 __builtin_msa_bneg_d (v2u64, v2u64); 16692 16693v16u8 __builtin_msa_bnegi_b (v16u8, imm0_7); 16694v8u16 __builtin_msa_bnegi_h (v8u16, imm0_15); 16695v4u32 __builtin_msa_bnegi_w (v4u32, imm0_31); 16696v2u64 __builtin_msa_bnegi_d (v2u64, imm0_63); 16697 16698i32 __builtin_msa_bnz_b (v16u8); 16699i32 __builtin_msa_bnz_h (v8u16); 16700i32 __builtin_msa_bnz_w (v4u32); 16701i32 __builtin_msa_bnz_d (v2u64); 16702 16703i32 __builtin_msa_bnz_v (v16u8); 16704 16705v16u8 __builtin_msa_bsel_v (v16u8, v16u8, v16u8); 16706 16707v16u8 __builtin_msa_bseli_b (v16u8, v16u8, imm0_255); 16708 16709v16u8 __builtin_msa_bset_b (v16u8, v16u8); 16710v8u16 __builtin_msa_bset_h (v8u16, v8u16); 16711v4u32 __builtin_msa_bset_w (v4u32, v4u32); 16712v2u64 __builtin_msa_bset_d (v2u64, v2u64); 16713 16714v16u8 __builtin_msa_bseti_b (v16u8, imm0_7); 16715v8u16 __builtin_msa_bseti_h (v8u16, imm0_15); 16716v4u32 __builtin_msa_bseti_w (v4u32, imm0_31); 16717v2u64 __builtin_msa_bseti_d (v2u64, imm0_63); 16718 16719i32 __builtin_msa_bz_b (v16u8); 16720i32 __builtin_msa_bz_h (v8u16); 16721i32 __builtin_msa_bz_w (v4u32); 16722i32 __builtin_msa_bz_d (v2u64); 16723 16724i32 __builtin_msa_bz_v (v16u8); 16725 16726v16i8 __builtin_msa_ceq_b (v16i8, v16i8); 16727v8i16 __builtin_msa_ceq_h (v8i16, v8i16); 16728v4i32 __builtin_msa_ceq_w (v4i32, v4i32); 16729v2i64 __builtin_msa_ceq_d (v2i64, v2i64); 16730 16731v16i8 __builtin_msa_ceqi_b (v16i8, imm_n16_15); 16732v8i16 __builtin_msa_ceqi_h (v8i16, imm_n16_15); 16733v4i32 __builtin_msa_ceqi_w (v4i32, imm_n16_15); 16734v2i64 __builtin_msa_ceqi_d (v2i64, imm_n16_15); 16735 16736i32 __builtin_msa_cfcmsa (imm0_31); 16737 16738v16i8 __builtin_msa_cle_s_b (v16i8, v16i8); 16739v8i16 __builtin_msa_cle_s_h (v8i16, v8i16); 16740v4i32 __builtin_msa_cle_s_w (v4i32, v4i32); 16741v2i64 __builtin_msa_cle_s_d (v2i64, v2i64); 16742 16743v16i8 __builtin_msa_cle_u_b (v16u8, v16u8); 16744v8i16 __builtin_msa_cle_u_h (v8u16, v8u16); 16745v4i32 __builtin_msa_cle_u_w (v4u32, v4u32); 16746v2i64 __builtin_msa_cle_u_d (v2u64, v2u64); 16747 16748v16i8 __builtin_msa_clei_s_b (v16i8, imm_n16_15); 16749v8i16 __builtin_msa_clei_s_h (v8i16, imm_n16_15); 16750v4i32 __builtin_msa_clei_s_w (v4i32, imm_n16_15); 16751v2i64 __builtin_msa_clei_s_d (v2i64, imm_n16_15); 16752 16753v16i8 __builtin_msa_clei_u_b (v16u8, imm0_31); 16754v8i16 __builtin_msa_clei_u_h (v8u16, imm0_31); 16755v4i32 __builtin_msa_clei_u_w (v4u32, imm0_31); 16756v2i64 __builtin_msa_clei_u_d (v2u64, imm0_31); 16757 16758v16i8 __builtin_msa_clt_s_b (v16i8, v16i8); 16759v8i16 __builtin_msa_clt_s_h (v8i16, v8i16); 16760v4i32 __builtin_msa_clt_s_w (v4i32, v4i32); 16761v2i64 __builtin_msa_clt_s_d (v2i64, v2i64); 16762 16763v16i8 __builtin_msa_clt_u_b (v16u8, v16u8); 16764v8i16 __builtin_msa_clt_u_h (v8u16, v8u16); 16765v4i32 __builtin_msa_clt_u_w (v4u32, v4u32); 16766v2i64 __builtin_msa_clt_u_d (v2u64, v2u64); 16767 16768v16i8 __builtin_msa_clti_s_b (v16i8, imm_n16_15); 16769v8i16 __builtin_msa_clti_s_h (v8i16, imm_n16_15); 16770v4i32 __builtin_msa_clti_s_w (v4i32, imm_n16_15); 16771v2i64 __builtin_msa_clti_s_d (v2i64, imm_n16_15); 16772 16773v16i8 __builtin_msa_clti_u_b (v16u8, imm0_31); 16774v8i16 __builtin_msa_clti_u_h (v8u16, imm0_31); 16775v4i32 __builtin_msa_clti_u_w (v4u32, imm0_31); 16776v2i64 __builtin_msa_clti_u_d (v2u64, imm0_31); 16777 16778i32 __builtin_msa_copy_s_b (v16i8, imm0_15); 16779i32 __builtin_msa_copy_s_h (v8i16, imm0_7); 16780i32 __builtin_msa_copy_s_w (v4i32, imm0_3); 16781i64 __builtin_msa_copy_s_d (v2i64, imm0_1); 16782 16783u32 __builtin_msa_copy_u_b (v16i8, imm0_15); 16784u32 __builtin_msa_copy_u_h (v8i16, imm0_7); 16785u32 __builtin_msa_copy_u_w (v4i32, imm0_3); 16786u64 __builtin_msa_copy_u_d (v2i64, imm0_1); 16787 16788void __builtin_msa_ctcmsa (imm0_31, i32); 16789 16790v16i8 __builtin_msa_div_s_b (v16i8, v16i8); 16791v8i16 __builtin_msa_div_s_h (v8i16, v8i16); 16792v4i32 __builtin_msa_div_s_w (v4i32, v4i32); 16793v2i64 __builtin_msa_div_s_d (v2i64, v2i64); 16794 16795v16u8 __builtin_msa_div_u_b (v16u8, v16u8); 16796v8u16 __builtin_msa_div_u_h (v8u16, v8u16); 16797v4u32 __builtin_msa_div_u_w (v4u32, v4u32); 16798v2u64 __builtin_msa_div_u_d (v2u64, v2u64); 16799 16800v8i16 __builtin_msa_dotp_s_h (v16i8, v16i8); 16801v4i32 __builtin_msa_dotp_s_w (v8i16, v8i16); 16802v2i64 __builtin_msa_dotp_s_d (v4i32, v4i32); 16803 16804v8u16 __builtin_msa_dotp_u_h (v16u8, v16u8); 16805v4u32 __builtin_msa_dotp_u_w (v8u16, v8u16); 16806v2u64 __builtin_msa_dotp_u_d (v4u32, v4u32); 16807 16808v8i16 __builtin_msa_dpadd_s_h (v8i16, v16i8, v16i8); 16809v4i32 __builtin_msa_dpadd_s_w (v4i32, v8i16, v8i16); 16810v2i64 __builtin_msa_dpadd_s_d (v2i64, v4i32, v4i32); 16811 16812v8u16 __builtin_msa_dpadd_u_h (v8u16, v16u8, v16u8); 16813v4u32 __builtin_msa_dpadd_u_w (v4u32, v8u16, v8u16); 16814v2u64 __builtin_msa_dpadd_u_d (v2u64, v4u32, v4u32); 16815 16816v8i16 __builtin_msa_dpsub_s_h (v8i16, v16i8, v16i8); 16817v4i32 __builtin_msa_dpsub_s_w (v4i32, v8i16, v8i16); 16818v2i64 __builtin_msa_dpsub_s_d (v2i64, v4i32, v4i32); 16819 16820v8i16 __builtin_msa_dpsub_u_h (v8i16, v16u8, v16u8); 16821v4i32 __builtin_msa_dpsub_u_w (v4i32, v8u16, v8u16); 16822v2i64 __builtin_msa_dpsub_u_d (v2i64, v4u32, v4u32); 16823 16824v4f32 __builtin_msa_fadd_w (v4f32, v4f32); 16825v2f64 __builtin_msa_fadd_d (v2f64, v2f64); 16826 16827v4i32 __builtin_msa_fcaf_w (v4f32, v4f32); 16828v2i64 __builtin_msa_fcaf_d (v2f64, v2f64); 16829 16830v4i32 __builtin_msa_fceq_w (v4f32, v4f32); 16831v2i64 __builtin_msa_fceq_d (v2f64, v2f64); 16832 16833v4i32 __builtin_msa_fclass_w (v4f32); 16834v2i64 __builtin_msa_fclass_d (v2f64); 16835 16836v4i32 __builtin_msa_fcle_w (v4f32, v4f32); 16837v2i64 __builtin_msa_fcle_d (v2f64, v2f64); 16838 16839v4i32 __builtin_msa_fclt_w (v4f32, v4f32); 16840v2i64 __builtin_msa_fclt_d (v2f64, v2f64); 16841 16842v4i32 __builtin_msa_fcne_w (v4f32, v4f32); 16843v2i64 __builtin_msa_fcne_d (v2f64, v2f64); 16844 16845v4i32 __builtin_msa_fcor_w (v4f32, v4f32); 16846v2i64 __builtin_msa_fcor_d (v2f64, v2f64); 16847 16848v4i32 __builtin_msa_fcueq_w (v4f32, v4f32); 16849v2i64 __builtin_msa_fcueq_d (v2f64, v2f64); 16850 16851v4i32 __builtin_msa_fcule_w (v4f32, v4f32); 16852v2i64 __builtin_msa_fcule_d (v2f64, v2f64); 16853 16854v4i32 __builtin_msa_fcult_w (v4f32, v4f32); 16855v2i64 __builtin_msa_fcult_d (v2f64, v2f64); 16856 16857v4i32 __builtin_msa_fcun_w (v4f32, v4f32); 16858v2i64 __builtin_msa_fcun_d (v2f64, v2f64); 16859 16860v4i32 __builtin_msa_fcune_w (v4f32, v4f32); 16861v2i64 __builtin_msa_fcune_d (v2f64, v2f64); 16862 16863v4f32 __builtin_msa_fdiv_w (v4f32, v4f32); 16864v2f64 __builtin_msa_fdiv_d (v2f64, v2f64); 16865 16866v8i16 __builtin_msa_fexdo_h (v4f32, v4f32); 16867v4f32 __builtin_msa_fexdo_w (v2f64, v2f64); 16868 16869v4f32 __builtin_msa_fexp2_w (v4f32, v4i32); 16870v2f64 __builtin_msa_fexp2_d (v2f64, v2i64); 16871 16872v4f32 __builtin_msa_fexupl_w (v8i16); 16873v2f64 __builtin_msa_fexupl_d (v4f32); 16874 16875v4f32 __builtin_msa_fexupr_w (v8i16); 16876v2f64 __builtin_msa_fexupr_d (v4f32); 16877 16878v4f32 __builtin_msa_ffint_s_w (v4i32); 16879v2f64 __builtin_msa_ffint_s_d (v2i64); 16880 16881v4f32 __builtin_msa_ffint_u_w (v4u32); 16882v2f64 __builtin_msa_ffint_u_d (v2u64); 16883 16884v4f32 __builtin_msa_ffql_w (v8i16); 16885v2f64 __builtin_msa_ffql_d (v4i32); 16886 16887v4f32 __builtin_msa_ffqr_w (v8i16); 16888v2f64 __builtin_msa_ffqr_d (v4i32); 16889 16890v16i8 __builtin_msa_fill_b (i32); 16891v8i16 __builtin_msa_fill_h (i32); 16892v4i32 __builtin_msa_fill_w (i32); 16893v2i64 __builtin_msa_fill_d (i64); 16894 16895v4f32 __builtin_msa_flog2_w (v4f32); 16896v2f64 __builtin_msa_flog2_d (v2f64); 16897 16898v4f32 __builtin_msa_fmadd_w (v4f32, v4f32, v4f32); 16899v2f64 __builtin_msa_fmadd_d (v2f64, v2f64, v2f64); 16900 16901v4f32 __builtin_msa_fmax_w (v4f32, v4f32); 16902v2f64 __builtin_msa_fmax_d (v2f64, v2f64); 16903 16904v4f32 __builtin_msa_fmax_a_w (v4f32, v4f32); 16905v2f64 __builtin_msa_fmax_a_d (v2f64, v2f64); 16906 16907v4f32 __builtin_msa_fmin_w (v4f32, v4f32); 16908v2f64 __builtin_msa_fmin_d (v2f64, v2f64); 16909 16910v4f32 __builtin_msa_fmin_a_w (v4f32, v4f32); 16911v2f64 __builtin_msa_fmin_a_d (v2f64, v2f64); 16912 16913v4f32 __builtin_msa_fmsub_w (v4f32, v4f32, v4f32); 16914v2f64 __builtin_msa_fmsub_d (v2f64, v2f64, v2f64); 16915 16916v4f32 __builtin_msa_fmul_w (v4f32, v4f32); 16917v2f64 __builtin_msa_fmul_d (v2f64, v2f64); 16918 16919v4f32 __builtin_msa_frint_w (v4f32); 16920v2f64 __builtin_msa_frint_d (v2f64); 16921 16922v4f32 __builtin_msa_frcp_w (v4f32); 16923v2f64 __builtin_msa_frcp_d (v2f64); 16924 16925v4f32 __builtin_msa_frsqrt_w (v4f32); 16926v2f64 __builtin_msa_frsqrt_d (v2f64); 16927 16928v4i32 __builtin_msa_fsaf_w (v4f32, v4f32); 16929v2i64 __builtin_msa_fsaf_d (v2f64, v2f64); 16930 16931v4i32 __builtin_msa_fseq_w (v4f32, v4f32); 16932v2i64 __builtin_msa_fseq_d (v2f64, v2f64); 16933 16934v4i32 __builtin_msa_fsle_w (v4f32, v4f32); 16935v2i64 __builtin_msa_fsle_d (v2f64, v2f64); 16936 16937v4i32 __builtin_msa_fslt_w (v4f32, v4f32); 16938v2i64 __builtin_msa_fslt_d (v2f64, v2f64); 16939 16940v4i32 __builtin_msa_fsne_w (v4f32, v4f32); 16941v2i64 __builtin_msa_fsne_d (v2f64, v2f64); 16942 16943v4i32 __builtin_msa_fsor_w (v4f32, v4f32); 16944v2i64 __builtin_msa_fsor_d (v2f64, v2f64); 16945 16946v4f32 __builtin_msa_fsqrt_w (v4f32); 16947v2f64 __builtin_msa_fsqrt_d (v2f64); 16948 16949v4f32 __builtin_msa_fsub_w (v4f32, v4f32); 16950v2f64 __builtin_msa_fsub_d (v2f64, v2f64); 16951 16952v4i32 __builtin_msa_fsueq_w (v4f32, v4f32); 16953v2i64 __builtin_msa_fsueq_d (v2f64, v2f64); 16954 16955v4i32 __builtin_msa_fsule_w (v4f32, v4f32); 16956v2i64 __builtin_msa_fsule_d (v2f64, v2f64); 16957 16958v4i32 __builtin_msa_fsult_w (v4f32, v4f32); 16959v2i64 __builtin_msa_fsult_d (v2f64, v2f64); 16960 16961v4i32 __builtin_msa_fsun_w (v4f32, v4f32); 16962v2i64 __builtin_msa_fsun_d (v2f64, v2f64); 16963 16964v4i32 __builtin_msa_fsune_w (v4f32, v4f32); 16965v2i64 __builtin_msa_fsune_d (v2f64, v2f64); 16966 16967v4i32 __builtin_msa_ftint_s_w (v4f32); 16968v2i64 __builtin_msa_ftint_s_d (v2f64); 16969 16970v4u32 __builtin_msa_ftint_u_w (v4f32); 16971v2u64 __builtin_msa_ftint_u_d (v2f64); 16972 16973v8i16 __builtin_msa_ftq_h (v4f32, v4f32); 16974v4i32 __builtin_msa_ftq_w (v2f64, v2f64); 16975 16976v4i32 __builtin_msa_ftrunc_s_w (v4f32); 16977v2i64 __builtin_msa_ftrunc_s_d (v2f64); 16978 16979v4u32 __builtin_msa_ftrunc_u_w (v4f32); 16980v2u64 __builtin_msa_ftrunc_u_d (v2f64); 16981 16982v8i16 __builtin_msa_hadd_s_h (v16i8, v16i8); 16983v4i32 __builtin_msa_hadd_s_w (v8i16, v8i16); 16984v2i64 __builtin_msa_hadd_s_d (v4i32, v4i32); 16985 16986v8u16 __builtin_msa_hadd_u_h (v16u8, v16u8); 16987v4u32 __builtin_msa_hadd_u_w (v8u16, v8u16); 16988v2u64 __builtin_msa_hadd_u_d (v4u32, v4u32); 16989 16990v8i16 __builtin_msa_hsub_s_h (v16i8, v16i8); 16991v4i32 __builtin_msa_hsub_s_w (v8i16, v8i16); 16992v2i64 __builtin_msa_hsub_s_d (v4i32, v4i32); 16993 16994v8i16 __builtin_msa_hsub_u_h (v16u8, v16u8); 16995v4i32 __builtin_msa_hsub_u_w (v8u16, v8u16); 16996v2i64 __builtin_msa_hsub_u_d (v4u32, v4u32); 16997 16998v16i8 __builtin_msa_ilvev_b (v16i8, v16i8); 16999v8i16 __builtin_msa_ilvev_h (v8i16, v8i16); 17000v4i32 __builtin_msa_ilvev_w (v4i32, v4i32); 17001v2i64 __builtin_msa_ilvev_d (v2i64, v2i64); 17002 17003v16i8 __builtin_msa_ilvl_b (v16i8, v16i8); 17004v8i16 __builtin_msa_ilvl_h (v8i16, v8i16); 17005v4i32 __builtin_msa_ilvl_w (v4i32, v4i32); 17006v2i64 __builtin_msa_ilvl_d (v2i64, v2i64); 17007 17008v16i8 __builtin_msa_ilvod_b (v16i8, v16i8); 17009v8i16 __builtin_msa_ilvod_h (v8i16, v8i16); 17010v4i32 __builtin_msa_ilvod_w (v4i32, v4i32); 17011v2i64 __builtin_msa_ilvod_d (v2i64, v2i64); 17012 17013v16i8 __builtin_msa_ilvr_b (v16i8, v16i8); 17014v8i16 __builtin_msa_ilvr_h (v8i16, v8i16); 17015v4i32 __builtin_msa_ilvr_w (v4i32, v4i32); 17016v2i64 __builtin_msa_ilvr_d (v2i64, v2i64); 17017 17018v16i8 __builtin_msa_insert_b (v16i8, imm0_15, i32); 17019v8i16 __builtin_msa_insert_h (v8i16, imm0_7, i32); 17020v4i32 __builtin_msa_insert_w (v4i32, imm0_3, i32); 17021v2i64 __builtin_msa_insert_d (v2i64, imm0_1, i64); 17022 17023v16i8 __builtin_msa_insve_b (v16i8, imm0_15, v16i8); 17024v8i16 __builtin_msa_insve_h (v8i16, imm0_7, v8i16); 17025v4i32 __builtin_msa_insve_w (v4i32, imm0_3, v4i32); 17026v2i64 __builtin_msa_insve_d (v2i64, imm0_1, v2i64); 17027 17028v16i8 __builtin_msa_ld_b (const void *, imm_n512_511); 17029v8i16 __builtin_msa_ld_h (const void *, imm_n1024_1022); 17030v4i32 __builtin_msa_ld_w (const void *, imm_n2048_2044); 17031v2i64 __builtin_msa_ld_d (const void *, imm_n4096_4088); 17032 17033v16i8 __builtin_msa_ldi_b (imm_n512_511); 17034v8i16 __builtin_msa_ldi_h (imm_n512_511); 17035v4i32 __builtin_msa_ldi_w (imm_n512_511); 17036v2i64 __builtin_msa_ldi_d (imm_n512_511); 17037 17038v8i16 __builtin_msa_madd_q_h (v8i16, v8i16, v8i16); 17039v4i32 __builtin_msa_madd_q_w (v4i32, v4i32, v4i32); 17040 17041v8i16 __builtin_msa_maddr_q_h (v8i16, v8i16, v8i16); 17042v4i32 __builtin_msa_maddr_q_w (v4i32, v4i32, v4i32); 17043 17044v16i8 __builtin_msa_maddv_b (v16i8, v16i8, v16i8); 17045v8i16 __builtin_msa_maddv_h (v8i16, v8i16, v8i16); 17046v4i32 __builtin_msa_maddv_w (v4i32, v4i32, v4i32); 17047v2i64 __builtin_msa_maddv_d (v2i64, v2i64, v2i64); 17048 17049v16i8 __builtin_msa_max_a_b (v16i8, v16i8); 17050v8i16 __builtin_msa_max_a_h (v8i16, v8i16); 17051v4i32 __builtin_msa_max_a_w (v4i32, v4i32); 17052v2i64 __builtin_msa_max_a_d (v2i64, v2i64); 17053 17054v16i8 __builtin_msa_max_s_b (v16i8, v16i8); 17055v8i16 __builtin_msa_max_s_h (v8i16, v8i16); 17056v4i32 __builtin_msa_max_s_w (v4i32, v4i32); 17057v2i64 __builtin_msa_max_s_d (v2i64, v2i64); 17058 17059v16u8 __builtin_msa_max_u_b (v16u8, v16u8); 17060v8u16 __builtin_msa_max_u_h (v8u16, v8u16); 17061v4u32 __builtin_msa_max_u_w (v4u32, v4u32); 17062v2u64 __builtin_msa_max_u_d (v2u64, v2u64); 17063 17064v16i8 __builtin_msa_maxi_s_b (v16i8, imm_n16_15); 17065v8i16 __builtin_msa_maxi_s_h (v8i16, imm_n16_15); 17066v4i32 __builtin_msa_maxi_s_w (v4i32, imm_n16_15); 17067v2i64 __builtin_msa_maxi_s_d (v2i64, imm_n16_15); 17068 17069v16u8 __builtin_msa_maxi_u_b (v16u8, imm0_31); 17070v8u16 __builtin_msa_maxi_u_h (v8u16, imm0_31); 17071v4u32 __builtin_msa_maxi_u_w (v4u32, imm0_31); 17072v2u64 __builtin_msa_maxi_u_d (v2u64, imm0_31); 17073 17074v16i8 __builtin_msa_min_a_b (v16i8, v16i8); 17075v8i16 __builtin_msa_min_a_h (v8i16, v8i16); 17076v4i32 __builtin_msa_min_a_w (v4i32, v4i32); 17077v2i64 __builtin_msa_min_a_d (v2i64, v2i64); 17078 17079v16i8 __builtin_msa_min_s_b (v16i8, v16i8); 17080v8i16 __builtin_msa_min_s_h (v8i16, v8i16); 17081v4i32 __builtin_msa_min_s_w (v4i32, v4i32); 17082v2i64 __builtin_msa_min_s_d (v2i64, v2i64); 17083 17084v16u8 __builtin_msa_min_u_b (v16u8, v16u8); 17085v8u16 __builtin_msa_min_u_h (v8u16, v8u16); 17086v4u32 __builtin_msa_min_u_w (v4u32, v4u32); 17087v2u64 __builtin_msa_min_u_d (v2u64, v2u64); 17088 17089v16i8 __builtin_msa_mini_s_b (v16i8, imm_n16_15); 17090v8i16 __builtin_msa_mini_s_h (v8i16, imm_n16_15); 17091v4i32 __builtin_msa_mini_s_w (v4i32, imm_n16_15); 17092v2i64 __builtin_msa_mini_s_d (v2i64, imm_n16_15); 17093 17094v16u8 __builtin_msa_mini_u_b (v16u8, imm0_31); 17095v8u16 __builtin_msa_mini_u_h (v8u16, imm0_31); 17096v4u32 __builtin_msa_mini_u_w (v4u32, imm0_31); 17097v2u64 __builtin_msa_mini_u_d (v2u64, imm0_31); 17098 17099v16i8 __builtin_msa_mod_s_b (v16i8, v16i8); 17100v8i16 __builtin_msa_mod_s_h (v8i16, v8i16); 17101v4i32 __builtin_msa_mod_s_w (v4i32, v4i32); 17102v2i64 __builtin_msa_mod_s_d (v2i64, v2i64); 17103 17104v16u8 __builtin_msa_mod_u_b (v16u8, v16u8); 17105v8u16 __builtin_msa_mod_u_h (v8u16, v8u16); 17106v4u32 __builtin_msa_mod_u_w (v4u32, v4u32); 17107v2u64 __builtin_msa_mod_u_d (v2u64, v2u64); 17108 17109v16i8 __builtin_msa_move_v (v16i8); 17110 17111v8i16 __builtin_msa_msub_q_h (v8i16, v8i16, v8i16); 17112v4i32 __builtin_msa_msub_q_w (v4i32, v4i32, v4i32); 17113 17114v8i16 __builtin_msa_msubr_q_h (v8i16, v8i16, v8i16); 17115v4i32 __builtin_msa_msubr_q_w (v4i32, v4i32, v4i32); 17116 17117v16i8 __builtin_msa_msubv_b (v16i8, v16i8, v16i8); 17118v8i16 __builtin_msa_msubv_h (v8i16, v8i16, v8i16); 17119v4i32 __builtin_msa_msubv_w (v4i32, v4i32, v4i32); 17120v2i64 __builtin_msa_msubv_d (v2i64, v2i64, v2i64); 17121 17122v8i16 __builtin_msa_mul_q_h (v8i16, v8i16); 17123v4i32 __builtin_msa_mul_q_w (v4i32, v4i32); 17124 17125v8i16 __builtin_msa_mulr_q_h (v8i16, v8i16); 17126v4i32 __builtin_msa_mulr_q_w (v4i32, v4i32); 17127 17128v16i8 __builtin_msa_mulv_b (v16i8, v16i8); 17129v8i16 __builtin_msa_mulv_h (v8i16, v8i16); 17130v4i32 __builtin_msa_mulv_w (v4i32, v4i32); 17131v2i64 __builtin_msa_mulv_d (v2i64, v2i64); 17132 17133v16i8 __builtin_msa_nloc_b (v16i8); 17134v8i16 __builtin_msa_nloc_h (v8i16); 17135v4i32 __builtin_msa_nloc_w (v4i32); 17136v2i64 __builtin_msa_nloc_d (v2i64); 17137 17138v16i8 __builtin_msa_nlzc_b (v16i8); 17139v8i16 __builtin_msa_nlzc_h (v8i16); 17140v4i32 __builtin_msa_nlzc_w (v4i32); 17141v2i64 __builtin_msa_nlzc_d (v2i64); 17142 17143v16u8 __builtin_msa_nor_v (v16u8, v16u8); 17144 17145v16u8 __builtin_msa_nori_b (v16u8, imm0_255); 17146 17147v16u8 __builtin_msa_or_v (v16u8, v16u8); 17148 17149v16u8 __builtin_msa_ori_b (v16u8, imm0_255); 17150 17151v16i8 __builtin_msa_pckev_b (v16i8, v16i8); 17152v8i16 __builtin_msa_pckev_h (v8i16, v8i16); 17153v4i32 __builtin_msa_pckev_w (v4i32, v4i32); 17154v2i64 __builtin_msa_pckev_d (v2i64, v2i64); 17155 17156v16i8 __builtin_msa_pckod_b (v16i8, v16i8); 17157v8i16 __builtin_msa_pckod_h (v8i16, v8i16); 17158v4i32 __builtin_msa_pckod_w (v4i32, v4i32); 17159v2i64 __builtin_msa_pckod_d (v2i64, v2i64); 17160 17161v16i8 __builtin_msa_pcnt_b (v16i8); 17162v8i16 __builtin_msa_pcnt_h (v8i16); 17163v4i32 __builtin_msa_pcnt_w (v4i32); 17164v2i64 __builtin_msa_pcnt_d (v2i64); 17165 17166v16i8 __builtin_msa_sat_s_b (v16i8, imm0_7); 17167v8i16 __builtin_msa_sat_s_h (v8i16, imm0_15); 17168v4i32 __builtin_msa_sat_s_w (v4i32, imm0_31); 17169v2i64 __builtin_msa_sat_s_d (v2i64, imm0_63); 17170 17171v16u8 __builtin_msa_sat_u_b (v16u8, imm0_7); 17172v8u16 __builtin_msa_sat_u_h (v8u16, imm0_15); 17173v4u32 __builtin_msa_sat_u_w (v4u32, imm0_31); 17174v2u64 __builtin_msa_sat_u_d (v2u64, imm0_63); 17175 17176v16i8 __builtin_msa_shf_b (v16i8, imm0_255); 17177v8i16 __builtin_msa_shf_h (v8i16, imm0_255); 17178v4i32 __builtin_msa_shf_w (v4i32, imm0_255); 17179 17180v16i8 __builtin_msa_sld_b (v16i8, v16i8, i32); 17181v8i16 __builtin_msa_sld_h (v8i16, v8i16, i32); 17182v4i32 __builtin_msa_sld_w (v4i32, v4i32, i32); 17183v2i64 __builtin_msa_sld_d (v2i64, v2i64, i32); 17184 17185v16i8 __builtin_msa_sldi_b (v16i8, v16i8, imm0_15); 17186v8i16 __builtin_msa_sldi_h (v8i16, v8i16, imm0_7); 17187v4i32 __builtin_msa_sldi_w (v4i32, v4i32, imm0_3); 17188v2i64 __builtin_msa_sldi_d (v2i64, v2i64, imm0_1); 17189 17190v16i8 __builtin_msa_sll_b (v16i8, v16i8); 17191v8i16 __builtin_msa_sll_h (v8i16, v8i16); 17192v4i32 __builtin_msa_sll_w (v4i32, v4i32); 17193v2i64 __builtin_msa_sll_d (v2i64, v2i64); 17194 17195v16i8 __builtin_msa_slli_b (v16i8, imm0_7); 17196v8i16 __builtin_msa_slli_h (v8i16, imm0_15); 17197v4i32 __builtin_msa_slli_w (v4i32, imm0_31); 17198v2i64 __builtin_msa_slli_d (v2i64, imm0_63); 17199 17200v16i8 __builtin_msa_splat_b (v16i8, i32); 17201v8i16 __builtin_msa_splat_h (v8i16, i32); 17202v4i32 __builtin_msa_splat_w (v4i32, i32); 17203v2i64 __builtin_msa_splat_d (v2i64, i32); 17204 17205v16i8 __builtin_msa_splati_b (v16i8, imm0_15); 17206v8i16 __builtin_msa_splati_h (v8i16, imm0_7); 17207v4i32 __builtin_msa_splati_w (v4i32, imm0_3); 17208v2i64 __builtin_msa_splati_d (v2i64, imm0_1); 17209 17210v16i8 __builtin_msa_sra_b (v16i8, v16i8); 17211v8i16 __builtin_msa_sra_h (v8i16, v8i16); 17212v4i32 __builtin_msa_sra_w (v4i32, v4i32); 17213v2i64 __builtin_msa_sra_d (v2i64, v2i64); 17214 17215v16i8 __builtin_msa_srai_b (v16i8, imm0_7); 17216v8i16 __builtin_msa_srai_h (v8i16, imm0_15); 17217v4i32 __builtin_msa_srai_w (v4i32, imm0_31); 17218v2i64 __builtin_msa_srai_d (v2i64, imm0_63); 17219 17220v16i8 __builtin_msa_srar_b (v16i8, v16i8); 17221v8i16 __builtin_msa_srar_h (v8i16, v8i16); 17222v4i32 __builtin_msa_srar_w (v4i32, v4i32); 17223v2i64 __builtin_msa_srar_d (v2i64, v2i64); 17224 17225v16i8 __builtin_msa_srari_b (v16i8, imm0_7); 17226v8i16 __builtin_msa_srari_h (v8i16, imm0_15); 17227v4i32 __builtin_msa_srari_w (v4i32, imm0_31); 17228v2i64 __builtin_msa_srari_d (v2i64, imm0_63); 17229 17230v16i8 __builtin_msa_srl_b (v16i8, v16i8); 17231v8i16 __builtin_msa_srl_h (v8i16, v8i16); 17232v4i32 __builtin_msa_srl_w (v4i32, v4i32); 17233v2i64 __builtin_msa_srl_d (v2i64, v2i64); 17234 17235v16i8 __builtin_msa_srli_b (v16i8, imm0_7); 17236v8i16 __builtin_msa_srli_h (v8i16, imm0_15); 17237v4i32 __builtin_msa_srli_w (v4i32, imm0_31); 17238v2i64 __builtin_msa_srli_d (v2i64, imm0_63); 17239 17240v16i8 __builtin_msa_srlr_b (v16i8, v16i8); 17241v8i16 __builtin_msa_srlr_h (v8i16, v8i16); 17242v4i32 __builtin_msa_srlr_w (v4i32, v4i32); 17243v2i64 __builtin_msa_srlr_d (v2i64, v2i64); 17244 17245v16i8 __builtin_msa_srlri_b (v16i8, imm0_7); 17246v8i16 __builtin_msa_srlri_h (v8i16, imm0_15); 17247v4i32 __builtin_msa_srlri_w (v4i32, imm0_31); 17248v2i64 __builtin_msa_srlri_d (v2i64, imm0_63); 17249 17250void __builtin_msa_st_b (v16i8, void *, imm_n512_511); 17251void __builtin_msa_st_h (v8i16, void *, imm_n1024_1022); 17252void __builtin_msa_st_w (v4i32, void *, imm_n2048_2044); 17253void __builtin_msa_st_d (v2i64, void *, imm_n4096_4088); 17254 17255v16i8 __builtin_msa_subs_s_b (v16i8, v16i8); 17256v8i16 __builtin_msa_subs_s_h (v8i16, v8i16); 17257v4i32 __builtin_msa_subs_s_w (v4i32, v4i32); 17258v2i64 __builtin_msa_subs_s_d (v2i64, v2i64); 17259 17260v16u8 __builtin_msa_subs_u_b (v16u8, v16u8); 17261v8u16 __builtin_msa_subs_u_h (v8u16, v8u16); 17262v4u32 __builtin_msa_subs_u_w (v4u32, v4u32); 17263v2u64 __builtin_msa_subs_u_d (v2u64, v2u64); 17264 17265v16u8 __builtin_msa_subsus_u_b (v16u8, v16i8); 17266v8u16 __builtin_msa_subsus_u_h (v8u16, v8i16); 17267v4u32 __builtin_msa_subsus_u_w (v4u32, v4i32); 17268v2u64 __builtin_msa_subsus_u_d (v2u64, v2i64); 17269 17270v16i8 __builtin_msa_subsuu_s_b (v16u8, v16u8); 17271v8i16 __builtin_msa_subsuu_s_h (v8u16, v8u16); 17272v4i32 __builtin_msa_subsuu_s_w (v4u32, v4u32); 17273v2i64 __builtin_msa_subsuu_s_d (v2u64, v2u64); 17274 17275v16i8 __builtin_msa_subv_b (v16i8, v16i8); 17276v8i16 __builtin_msa_subv_h (v8i16, v8i16); 17277v4i32 __builtin_msa_subv_w (v4i32, v4i32); 17278v2i64 __builtin_msa_subv_d (v2i64, v2i64); 17279 17280v16i8 __builtin_msa_subvi_b (v16i8, imm0_31); 17281v8i16 __builtin_msa_subvi_h (v8i16, imm0_31); 17282v4i32 __builtin_msa_subvi_w (v4i32, imm0_31); 17283v2i64 __builtin_msa_subvi_d (v2i64, imm0_31); 17284 17285v16i8 __builtin_msa_vshf_b (v16i8, v16i8, v16i8); 17286v8i16 __builtin_msa_vshf_h (v8i16, v8i16, v8i16); 17287v4i32 __builtin_msa_vshf_w (v4i32, v4i32, v4i32); 17288v2i64 __builtin_msa_vshf_d (v2i64, v2i64, v2i64); 17289 17290v16u8 __builtin_msa_xor_v (v16u8, v16u8); 17291 17292v16u8 __builtin_msa_xori_b (v16u8, imm0_255); 17293@end smallexample 17294 17295@node Other MIPS Built-in Functions 17296@subsection Other MIPS Built-in Functions 17297 17298GCC provides other MIPS-specific built-in functions: 17299 17300@table @code 17301@item void __builtin_mips_cache (int @var{op}, const volatile void *@var{addr}) 17302Insert a @samp{cache} instruction with operands @var{op} and @var{addr}. 17303GCC defines the preprocessor macro @code{___GCC_HAVE_BUILTIN_MIPS_CACHE} 17304when this function is available. 17305 17306@item unsigned int __builtin_mips_get_fcsr (void) 17307@itemx void __builtin_mips_set_fcsr (unsigned int @var{value}) 17308Get and set the contents of the floating-point control and status register 17309(FPU control register 31). These functions are only available in hard-float 17310code but can be called in both MIPS16 and non-MIPS16 contexts. 17311 17312@code{__builtin_mips_set_fcsr} can be used to change any bit of the 17313register except the condition codes, which GCC assumes are preserved. 17314@end table 17315 17316@node MSP430 Built-in Functions 17317@subsection MSP430 Built-in Functions 17318 17319GCC provides a couple of special builtin functions to aid in the 17320writing of interrupt handlers in C. 17321 17322@table @code 17323@item __bic_SR_register_on_exit (int @var{mask}) 17324This clears the indicated bits in the saved copy of the status register 17325currently residing on the stack. This only works inside interrupt 17326handlers and the changes to the status register will only take affect 17327once the handler returns. 17328 17329@item __bis_SR_register_on_exit (int @var{mask}) 17330This sets the indicated bits in the saved copy of the status register 17331currently residing on the stack. This only works inside interrupt 17332handlers and the changes to the status register will only take affect 17333once the handler returns. 17334 17335@item __delay_cycles (long long @var{cycles}) 17336This inserts an instruction sequence that takes exactly @var{cycles} 17337cycles (between 0 and about 17E9) to complete. The inserted sequence 17338may use jumps, loops, or no-ops, and does not interfere with any other 17339instructions. Note that @var{cycles} must be a compile-time constant 17340integer - that is, you must pass a number, not a variable that may be 17341optimized to a constant later. The number of cycles delayed by this 17342builtin is exact. 17343@end table 17344 17345@node NDS32 Built-in Functions 17346@subsection NDS32 Built-in Functions 17347 17348These built-in functions are available for the NDS32 target: 17349 17350@deftypefn {Built-in Function} void __builtin_nds32_isync (int *@var{addr}) 17351Insert an ISYNC instruction into the instruction stream where 17352@var{addr} is an instruction address for serialization. 17353@end deftypefn 17354 17355@deftypefn {Built-in Function} void __builtin_nds32_isb (void) 17356Insert an ISB instruction into the instruction stream. 17357@end deftypefn 17358 17359@deftypefn {Built-in Function} int __builtin_nds32_mfsr (int @var{sr}) 17360Return the content of a system register which is mapped by @var{sr}. 17361@end deftypefn 17362 17363@deftypefn {Built-in Function} int __builtin_nds32_mfusr (int @var{usr}) 17364Return the content of a user space register which is mapped by @var{usr}. 17365@end deftypefn 17366 17367@deftypefn {Built-in Function} void __builtin_nds32_mtsr (int @var{value}, int @var{sr}) 17368Move the @var{value} to a system register which is mapped by @var{sr}. 17369@end deftypefn 17370 17371@deftypefn {Built-in Function} void __builtin_nds32_mtusr (int @var{value}, int @var{usr}) 17372Move the @var{value} to a user space register which is mapped by @var{usr}. 17373@end deftypefn 17374 17375@deftypefn {Built-in Function} void __builtin_nds32_setgie_en (void) 17376Enable global interrupt. 17377@end deftypefn 17378 17379@deftypefn {Built-in Function} void __builtin_nds32_setgie_dis (void) 17380Disable global interrupt. 17381@end deftypefn 17382 17383@node picoChip Built-in Functions 17384@subsection picoChip Built-in Functions 17385 17386GCC provides an interface to selected machine instructions from the 17387picoChip instruction set. 17388 17389@table @code 17390@item int __builtin_sbc (int @var{value}) 17391Sign bit count. Return the number of consecutive bits in @var{value} 17392that have the same value as the sign bit. The result is the number of 17393leading sign bits minus one, giving the number of redundant sign bits in 17394@var{value}. 17395 17396@item int __builtin_byteswap (int @var{value}) 17397Byte swap. Return the result of swapping the upper and lower bytes of 17398@var{value}. 17399 17400@item int __builtin_brev (int @var{value}) 17401Bit reversal. Return the result of reversing the bits in 17402@var{value}. Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1, 17403and so on. 17404 17405@item int __builtin_adds (int @var{x}, int @var{y}) 17406Saturating addition. Return the result of adding @var{x} and @var{y}, 17407storing the value 32767 if the result overflows. 17408 17409@item int __builtin_subs (int @var{x}, int @var{y}) 17410Saturating subtraction. Return the result of subtracting @var{y} from 17411@var{x}, storing the value @minus{}32768 if the result overflows. 17412 17413@item void __builtin_halt (void) 17414Halt. The processor stops execution. This built-in is useful for 17415implementing assertions. 17416 17417@end table 17418 17419@node Basic PowerPC Built-in Functions 17420@subsection Basic PowerPC Built-in Functions 17421 17422@menu 17423* Basic PowerPC Built-in Functions Available on all Configurations:: 17424* Basic PowerPC Built-in Functions Available on ISA 2.05:: 17425* Basic PowerPC Built-in Functions Available on ISA 2.06:: 17426* Basic PowerPC Built-in Functions Available on ISA 2.07:: 17427* Basic PowerPC Built-in Functions Available on ISA 3.0:: 17428* Basic PowerPC Built-in Functions Available on ISA 3.1:: 17429@end menu 17430 17431This section describes PowerPC built-in functions that do not require 17432the inclusion of any special header files to declare prototypes or 17433provide macro definitions. The sections that follow describe 17434additional PowerPC built-in functions. 17435 17436@node Basic PowerPC Built-in Functions Available on all Configurations 17437@subsubsection Basic PowerPC Built-in Functions Available on all Configurations 17438 17439@deftypefn {Built-in Function} void __builtin_cpu_init (void) 17440This function is a @code{nop} on the PowerPC platform and is included solely 17441to maintain API compatibility with the x86 builtins. 17442@end deftypefn 17443 17444@deftypefn {Built-in Function} int __builtin_cpu_is (const char *@var{cpuname}) 17445This function returns a value of @code{1} if the run-time CPU is of type 17446@var{cpuname} and returns @code{0} otherwise 17447 17448The @code{__builtin_cpu_is} function requires GLIBC 2.23 or newer 17449which exports the hardware capability bits. GCC defines the macro 17450@code{__BUILTIN_CPU_SUPPORTS__} if the @code{__builtin_cpu_supports} 17451built-in function is fully supported. 17452 17453If GCC was configured to use a GLIBC before 2.23, the built-in 17454function @code{__builtin_cpu_is} always returns a 0 and the compiler 17455issues a warning. 17456 17457The following CPU names can be detected: 17458 17459@table @samp 17460@item power10 17461IBM POWER10 Server CPU. 17462@item power9 17463IBM POWER9 Server CPU. 17464@item power8 17465IBM POWER8 Server CPU. 17466@item power7 17467IBM POWER7 Server CPU. 17468@item power6x 17469IBM POWER6 Server CPU (RAW mode). 17470@item power6 17471IBM POWER6 Server CPU (Architected mode). 17472@item power5+ 17473IBM POWER5+ Server CPU. 17474@item power5 17475IBM POWER5 Server CPU. 17476@item ppc970 17477IBM 970 Server CPU (ie, Apple G5). 17478@item power4 17479IBM POWER4 Server CPU. 17480@item ppca2 17481IBM A2 64-bit Embedded CPU 17482@item ppc476 17483IBM PowerPC 476FP 32-bit Embedded CPU. 17484@item ppc464 17485IBM PowerPC 464 32-bit Embedded CPU. 17486@item ppc440 17487PowerPC 440 32-bit Embedded CPU. 17488@item ppc405 17489PowerPC 405 32-bit Embedded CPU. 17490@item ppc-cell-be 17491IBM PowerPC Cell Broadband Engine Architecture CPU. 17492@end table 17493 17494Here is an example: 17495@smallexample 17496#ifdef __BUILTIN_CPU_SUPPORTS__ 17497 if (__builtin_cpu_is ("power8")) 17498 @{ 17499 do_power8 (); // POWER8 specific implementation. 17500 @} 17501 else 17502#endif 17503 @{ 17504 do_generic (); // Generic implementation. 17505 @} 17506@end smallexample 17507@end deftypefn 17508 17509@deftypefn {Built-in Function} int __builtin_cpu_supports (const char *@var{feature}) 17510This function returns a value of @code{1} if the run-time CPU supports the HWCAP 17511feature @var{feature} and returns @code{0} otherwise. 17512 17513The @code{__builtin_cpu_supports} function requires GLIBC 2.23 or 17514newer which exports the hardware capability bits. GCC defines the 17515macro @code{__BUILTIN_CPU_SUPPORTS__} if the 17516@code{__builtin_cpu_supports} built-in function is fully supported. 17517 17518If GCC was configured to use a GLIBC before 2.23, the built-in 17519function @code{__builtin_cpu_suports} always returns a 0 and the 17520compiler issues a warning. 17521 17522The following features can be 17523detected: 17524 17525@table @samp 17526@item 4xxmac 175274xx CPU has a Multiply Accumulator. 17528@item altivec 17529CPU has a SIMD/Vector Unit. 17530@item arch_2_05 17531CPU supports ISA 2.05 (eg, POWER6) 17532@item arch_2_06 17533CPU supports ISA 2.06 (eg, POWER7) 17534@item arch_2_07 17535CPU supports ISA 2.07 (eg, POWER8) 17536@item arch_3_00 17537CPU supports ISA 3.0 (eg, POWER9) 17538@item arch_3_1 17539CPU supports ISA 3.1 (eg, POWER10) 17540@item archpmu 17541CPU supports the set of compatible performance monitoring events. 17542@item booke 17543CPU supports the Embedded ISA category. 17544@item cellbe 17545CPU has a CELL broadband engine. 17546@item darn 17547CPU supports the @code{darn} (deliver a random number) instruction. 17548@item dfp 17549CPU has a decimal floating point unit. 17550@item dscr 17551CPU supports the data stream control register. 17552@item ebb 17553CPU supports event base branching. 17554@item efpdouble 17555CPU has a SPE double precision floating point unit. 17556@item efpsingle 17557CPU has a SPE single precision floating point unit. 17558@item fpu 17559CPU has a floating point unit. 17560@item htm 17561CPU has hardware transaction memory instructions. 17562@item htm-nosc 17563Kernel aborts hardware transactions when a syscall is made. 17564@item htm-no-suspend 17565CPU supports hardware transaction memory but does not support the 17566@code{tsuspend.} instruction. 17567@item ic_snoop 17568CPU supports icache snooping capabilities. 17569@item ieee128 17570CPU supports 128-bit IEEE binary floating point instructions. 17571@item isel 17572CPU supports the integer select instruction. 17573@item mma 17574CPU supports the matrix-multiply assist instructions. 17575@item mmu 17576CPU has a memory management unit. 17577@item notb 17578CPU does not have a timebase (eg, 601 and 403gx). 17579@item pa6t 17580CPU supports the PA Semi 6T CORE ISA. 17581@item power4 17582CPU supports ISA 2.00 (eg, POWER4) 17583@item power5 17584CPU supports ISA 2.02 (eg, POWER5) 17585@item power5+ 17586CPU supports ISA 2.03 (eg, POWER5+) 17587@item power6x 17588CPU supports ISA 2.05 (eg, POWER6) extended opcodes mffgpr and mftgpr. 17589@item ppc32 17590CPU supports 32-bit mode execution. 17591@item ppc601 17592CPU supports the old POWER ISA (eg, 601) 17593@item ppc64 17594CPU supports 64-bit mode execution. 17595@item ppcle 17596CPU supports a little-endian mode that uses address swizzling. 17597@item scv 17598Kernel supports system call vectored. 17599@item smt 17600CPU support simultaneous multi-threading. 17601@item spe 17602CPU has a signal processing extension unit. 17603@item tar 17604CPU supports the target address register. 17605@item true_le 17606CPU supports true little-endian mode. 17607@item ucache 17608CPU has unified I/D cache. 17609@item vcrypto 17610CPU supports the vector cryptography instructions. 17611@item vsx 17612CPU supports the vector-scalar extension. 17613@end table 17614 17615Here is an example: 17616@smallexample 17617#ifdef __BUILTIN_CPU_SUPPORTS__ 17618 if (__builtin_cpu_supports ("fpu")) 17619 @{ 17620 asm("fadd %0,%1,%2" : "=d"(dst) : "d"(src1), "d"(src2)); 17621 @} 17622 else 17623#endif 17624 @{ 17625 dst = __fadd (src1, src2); // Software FP addition function. 17626 @} 17627@end smallexample 17628@end deftypefn 17629 17630The following built-in functions are also available on all PowerPC 17631processors: 17632@smallexample 17633uint64_t __builtin_ppc_get_timebase (); 17634unsigned long __builtin_ppc_mftb (); 17635double __builtin_unpack_ibm128 (__ibm128, int); 17636__ibm128 __builtin_pack_ibm128 (double, double); 17637double __builtin_mffs (void); 17638void __builtin_mtfsf (const int, double); 17639void __builtin_mtfsb0 (const int); 17640void __builtin_mtfsb1 (const int); 17641void __builtin_set_fpscr_rn (int); 17642@end smallexample 17643 17644The @code{__builtin_ppc_get_timebase} and @code{__builtin_ppc_mftb} 17645functions generate instructions to read the Time Base Register. The 17646@code{__builtin_ppc_get_timebase} function may generate multiple 17647instructions and always returns the 64 bits of the Time Base Register. 17648The @code{__builtin_ppc_mftb} function always generates one instruction and 17649returns the Time Base Register value as an unsigned long, throwing away 17650the most significant word on 32-bit environments. The @code{__builtin_mffs} 17651return the value of the FPSCR register. Note, ISA 3.0 supports the 17652@code{__builtin_mffsl()} which permits software to read the control and 17653non-sticky status bits in the FSPCR without the higher latency associated with 17654accessing the sticky status bits. The @code{__builtin_mtfsf} takes a constant 176558-bit integer field mask and a double precision floating point argument 17656and generates the @code{mtfsf} (extended mnemonic) instruction to write new 17657values to selected fields of the FPSCR. The 17658@code{__builtin_mtfsb0} and @code{__builtin_mtfsb1} take the bit to change 17659as an argument. The valid bit range is between 0 and 31. The builtins map to 17660the @code{mtfsb0} and @code{mtfsb1} instructions which take the argument and 17661add 32. Hence these instructions only modify the FPSCR[32:63] bits by 17662changing the specified bit to a zero or one respectively. The 17663@code{__builtin_set_fpscr_rn} builtin allows changing both of the floating 17664point rounding mode bits. The argument is a 2-bit value. The argument can 17665either be a @code{const int} or stored in a variable. The builtin uses 17666the ISA 3.0 17667instruction @code{mffscrn} if available, otherwise it reads the FPSCR, masks 17668the current rounding mode bits out and OR's in the new value. 17669 17670@node Basic PowerPC Built-in Functions Available on ISA 2.05 17671@subsubsection Basic PowerPC Built-in Functions Available on ISA 2.05 17672 17673The basic built-in functions described in this section are 17674available on the PowerPC family of processors starting with ISA 2.05 17675or later. Unless specific options are explicitly disabled on the 17676command line, specifying option @option{-mcpu=power6} has the effect of 17677enabling the @option{-mpowerpc64}, @option{-mpowerpc-gpopt}, 17678@option{-mpowerpc-gfxopt}, @option{-mmfcrf}, @option{-mpopcntb}, 17679@option{-mfprnd}, @option{-mcmpb}, @option{-mhard-dfp}, and 17680@option{-mrecip-precision} options. Specify the 17681@option{-maltivec} option explicitly in 17682combination with the above options if desired. 17683 17684The following functions require option @option{-mcmpb}. 17685@smallexample 17686unsigned long long __builtin_cmpb (unsigned long long int, unsigned long long int); 17687unsigned int __builtin_cmpb (unsigned int, unsigned int); 17688@end smallexample 17689 17690The @code{__builtin_cmpb} function 17691performs a byte-wise compare on the contents of its two arguments, 17692returning the result of the byte-wise comparison as the returned 17693value. For each byte comparison, the corresponding byte of the return 17694value holds 0xff if the input bytes are equal and 0 if the input bytes 17695are not equal. If either of the arguments to this built-in function 17696is wider than 32 bits, the function call expands into the form that 17697expects @code{unsigned long long int} arguments 17698which is only available on 64-bit targets. 17699 17700The following built-in functions are available 17701when hardware decimal floating point 17702(@option{-mhard-dfp}) is available: 17703@smallexample 17704void __builtin_set_fpscr_drn(int); 17705_Decimal64 __builtin_ddedpd (int, _Decimal64); 17706_Decimal128 __builtin_ddedpdq (int, _Decimal128); 17707_Decimal64 __builtin_denbcd (int, _Decimal64); 17708_Decimal128 __builtin_denbcdq (int, _Decimal128); 17709_Decimal64 __builtin_diex (long long, _Decimal64); 17710_Decimal128 _builtin_diexq (long long, _Decimal128); 17711_Decimal64 __builtin_dscli (_Decimal64, int); 17712_Decimal128 __builtin_dscliq (_Decimal128, int); 17713_Decimal64 __builtin_dscri (_Decimal64, int); 17714_Decimal128 __builtin_dscriq (_Decimal128, int); 17715long long __builtin_dxex (_Decimal64); 17716long long __builtin_dxexq (_Decimal128); 17717_Decimal128 __builtin_pack_dec128 (unsigned long long, unsigned long long); 17718unsigned long long __builtin_unpack_dec128 (_Decimal128, int); 17719 17720The @code{__builtin_set_fpscr_drn} builtin allows changing the three decimal 17721floating point rounding mode bits. The argument is a 3-bit value. The 17722argument can either be a @code{const int} or the value can be stored in 17723a variable. 17724The builtin uses the ISA 3.0 instruction @code{mffscdrn} if available. 17725Otherwise the builtin reads the FPSCR, masks the current decimal rounding 17726mode bits out and OR's in the new value. 17727 17728@end smallexample 17729 17730The following functions require @option{-mhard-float}, 17731@option{-mpowerpc-gfxopt}, and @option{-mpopcntb} options. 17732 17733@smallexample 17734double __builtin_recipdiv (double, double); 17735float __builtin_recipdivf (float, float); 17736double __builtin_rsqrt (double); 17737float __builtin_rsqrtf (float); 17738@end smallexample 17739 17740The @code{vec_rsqrt}, @code{__builtin_rsqrt}, and 17741@code{__builtin_rsqrtf} functions generate multiple instructions to 17742implement the reciprocal sqrt functionality using reciprocal sqrt 17743estimate instructions. 17744 17745The @code{__builtin_recipdiv}, and @code{__builtin_recipdivf} 17746functions generate multiple instructions to implement division using 17747the reciprocal estimate instructions. 17748 17749The following functions require @option{-mhard-float} and 17750@option{-mmultiple} options. 17751 17752The @code{__builtin_unpack_longdouble} function takes a 17753@code{long double} argument and a compile time constant of 0 or 1. If 17754the constant is 0, the first @code{double} within the 17755@code{long double} is returned, otherwise the second @code{double} 17756is returned. The @code{__builtin_unpack_longdouble} function is only 17757available if @code{long double} uses the IBM extended double 17758representation. 17759 17760The @code{__builtin_pack_longdouble} function takes two @code{double} 17761arguments and returns a @code{long double} value that combines the two 17762arguments. The @code{__builtin_pack_longdouble} function is only 17763available if @code{long double} uses the IBM extended double 17764representation. 17765 17766The @code{__builtin_unpack_ibm128} function takes a @code{__ibm128} 17767argument and a compile time constant of 0 or 1. If the constant is 0, 17768the first @code{double} within the @code{__ibm128} is returned, 17769otherwise the second @code{double} is returned. 17770 17771The @code{__builtin_pack_ibm128} function takes two @code{double} 17772arguments and returns a @code{__ibm128} value that combines the two 17773arguments. 17774 17775Additional built-in functions are available for the 64-bit PowerPC 17776family of processors, for efficient use of 128-bit floating point 17777(@code{__float128}) values. 17778 17779@node Basic PowerPC Built-in Functions Available on ISA 2.06 17780@subsubsection Basic PowerPC Built-in Functions Available on ISA 2.06 17781 17782The basic built-in functions described in this section are 17783available on the PowerPC family of processors starting with ISA 2.05 17784or later. Unless specific options are explicitly disabled on the 17785command line, specifying option @option{-mcpu=power7} has the effect of 17786enabling all the same options as for @option{-mcpu=power6} in 17787addition to the @option{-maltivec}, @option{-mpopcntd}, and 17788@option{-mvsx} options. 17789 17790The following basic built-in functions require @option{-mpopcntd}: 17791@smallexample 17792unsigned int __builtin_addg6s (unsigned int, unsigned int); 17793long long __builtin_bpermd (long long, long long); 17794unsigned int __builtin_cbcdtd (unsigned int); 17795unsigned int __builtin_cdtbcd (unsigned int); 17796long long __builtin_divde (long long, long long); 17797unsigned long long __builtin_divdeu (unsigned long long, unsigned long long); 17798int __builtin_divwe (int, int); 17799unsigned int __builtin_divweu (unsigned int, unsigned int); 17800vector __int128 __builtin_pack_vector_int128 (long long, long long); 17801void __builtin_rs6000_speculation_barrier (void); 17802long long __builtin_unpack_vector_int128 (vector __int128, signed char); 17803@end smallexample 17804 17805Of these, the @code{__builtin_divde} and @code{__builtin_divdeu} functions 17806require a 64-bit environment. 17807 17808The following basic built-in functions, which are also supported on 17809x86 targets, require @option{-mfloat128}. 17810@smallexample 17811__float128 __builtin_fabsq (__float128); 17812__float128 __builtin_copysignq (__float128, __float128); 17813__float128 __builtin_infq (void); 17814__float128 __builtin_huge_valq (void); 17815__float128 __builtin_nanq (void); 17816__float128 __builtin_nansq (void); 17817 17818__float128 __builtin_sqrtf128 (__float128); 17819__float128 __builtin_fmaf128 (__float128, __float128, __float128); 17820@end smallexample 17821 17822@node Basic PowerPC Built-in Functions Available on ISA 2.07 17823@subsubsection Basic PowerPC Built-in Functions Available on ISA 2.07 17824 17825The basic built-in functions described in this section are 17826available on the PowerPC family of processors starting with ISA 2.07 17827or later. Unless specific options are explicitly disabled on the 17828command line, specifying option @option{-mcpu=power8} has the effect of 17829enabling all the same options as for @option{-mcpu=power7} in 17830addition to the @option{-mpower8-fusion}, @option{-mpower8-vector}, 17831@option{-mcrypto}, @option{-mhtm}, @option{-mquad-memory}, and 17832@option{-mquad-memory-atomic} options. 17833 17834This section intentionally empty. 17835 17836@node Basic PowerPC Built-in Functions Available on ISA 3.0 17837@subsubsection Basic PowerPC Built-in Functions Available on ISA 3.0 17838 17839The basic built-in functions described in this section are 17840available on the PowerPC family of processors starting with ISA 3.0 17841or later. Unless specific options are explicitly disabled on the 17842command line, specifying option @option{-mcpu=power9} has the effect of 17843enabling all the same options as for @option{-mcpu=power8} in 17844addition to the @option{-misel} option. 17845 17846The following built-in functions are available on Linux 64-bit systems 17847that use the ISA 3.0 instruction set (@option{-mcpu=power9}): 17848 17849@table @code 17850@item __float128 __builtin_addf128_round_to_odd (__float128, __float128) 17851Perform a 128-bit IEEE floating point add using round to odd as the 17852rounding mode. 17853@findex __builtin_addf128_round_to_odd 17854 17855@item __float128 __builtin_subf128_round_to_odd (__float128, __float128) 17856Perform a 128-bit IEEE floating point subtract using round to odd as 17857the rounding mode. 17858@findex __builtin_subf128_round_to_odd 17859 17860@item __float128 __builtin_mulf128_round_to_odd (__float128, __float128) 17861Perform a 128-bit IEEE floating point multiply using round to odd as 17862the rounding mode. 17863@findex __builtin_mulf128_round_to_odd 17864 17865@item __float128 __builtin_divf128_round_to_odd (__float128, __float128) 17866Perform a 128-bit IEEE floating point divide using round to odd as 17867the rounding mode. 17868@findex __builtin_divf128_round_to_odd 17869 17870@item __float128 __builtin_sqrtf128_round_to_odd (__float128) 17871Perform a 128-bit IEEE floating point square root using round to odd 17872as the rounding mode. 17873@findex __builtin_sqrtf128_round_to_odd 17874 17875@item __float128 __builtin_fmaf128_round_to_odd (__float128, __float128, __float128) 17876Perform a 128-bit IEEE floating point fused multiply and add operation 17877using round to odd as the rounding mode. 17878@findex __builtin_fmaf128_round_to_odd 17879 17880@item double __builtin_truncf128_round_to_odd (__float128) 17881Convert a 128-bit IEEE floating point value to @code{double} using 17882round to odd as the rounding mode. 17883@findex __builtin_truncf128_round_to_odd 17884@end table 17885 17886The following additional built-in functions are also available for the 17887PowerPC family of processors, starting with ISA 3.0 or later: 17888@smallexample 17889long long __builtin_darn (void); 17890long long __builtin_darn_raw (void); 17891int __builtin_darn_32 (void); 17892@end smallexample 17893 17894The @code{__builtin_darn} and @code{__builtin_darn_raw} 17895functions require a 1789664-bit environment supporting ISA 3.0 or later. 17897The @code{__builtin_darn} function provides a 64-bit conditioned 17898random number. The @code{__builtin_darn_raw} function provides a 1789964-bit raw random number. The @code{__builtin_darn_32} function 17900provides a 32-bit conditioned random number. 17901 17902The following additional built-in functions are also available for the 17903PowerPC family of processors, starting with ISA 3.0 or later: 17904 17905@smallexample 17906int __builtin_byte_in_set (unsigned char u, unsigned long long set); 17907int __builtin_byte_in_range (unsigned char u, unsigned int range); 17908int __builtin_byte_in_either_range (unsigned char u, unsigned int ranges); 17909 17910int __builtin_dfp_dtstsfi_lt (unsigned int comparison, _Decimal64 value); 17911int __builtin_dfp_dtstsfi_lt (unsigned int comparison, _Decimal128 value); 17912int __builtin_dfp_dtstsfi_lt_dd (unsigned int comparison, _Decimal64 value); 17913int __builtin_dfp_dtstsfi_lt_td (unsigned int comparison, _Decimal128 value); 17914 17915int __builtin_dfp_dtstsfi_gt (unsigned int comparison, _Decimal64 value); 17916int __builtin_dfp_dtstsfi_gt (unsigned int comparison, _Decimal128 value); 17917int __builtin_dfp_dtstsfi_gt_dd (unsigned int comparison, _Decimal64 value); 17918int __builtin_dfp_dtstsfi_gt_td (unsigned int comparison, _Decimal128 value); 17919 17920int __builtin_dfp_dtstsfi_eq (unsigned int comparison, _Decimal64 value); 17921int __builtin_dfp_dtstsfi_eq (unsigned int comparison, _Decimal128 value); 17922int __builtin_dfp_dtstsfi_eq_dd (unsigned int comparison, _Decimal64 value); 17923int __builtin_dfp_dtstsfi_eq_td (unsigned int comparison, _Decimal128 value); 17924 17925int __builtin_dfp_dtstsfi_ov (unsigned int comparison, _Decimal64 value); 17926int __builtin_dfp_dtstsfi_ov (unsigned int comparison, _Decimal128 value); 17927int __builtin_dfp_dtstsfi_ov_dd (unsigned int comparison, _Decimal64 value); 17928int __builtin_dfp_dtstsfi_ov_td (unsigned int comparison, _Decimal128 value); 17929 17930double __builtin_mffsl(void); 17931 17932@end smallexample 17933The @code{__builtin_byte_in_set} function requires a 1793464-bit environment supporting ISA 3.0 or later. This function returns 17935a non-zero value if and only if its @code{u} argument exactly equals one of 17936the eight bytes contained within its 64-bit @code{set} argument. 17937 17938The @code{__builtin_byte_in_range} and 17939@code{__builtin_byte_in_either_range} require an environment 17940supporting ISA 3.0 or later. For these two functions, the 17941@code{range} argument is encoded as 4 bytes, organized as 17942@code{hi_1:lo_1:hi_2:lo_2}. 17943The @code{__builtin_byte_in_range} function returns a 17944non-zero value if and only if its @code{u} argument is within the 17945range bounded between @code{lo_2} and @code{hi_2} inclusive. 17946The @code{__builtin_byte_in_either_range} function returns non-zero if 17947and only if its @code{u} argument is within either the range bounded 17948between @code{lo_1} and @code{hi_1} inclusive or the range bounded 17949between @code{lo_2} and @code{hi_2} inclusive. 17950 17951The @code{__builtin_dfp_dtstsfi_lt} function returns a non-zero value 17952if and only if the number of signficant digits of its @code{value} argument 17953is less than its @code{comparison} argument. The 17954@code{__builtin_dfp_dtstsfi_lt_dd} and 17955@code{__builtin_dfp_dtstsfi_lt_td} functions behave similarly, but 17956require that the type of the @code{value} argument be 17957@code{__Decimal64} and @code{__Decimal128} respectively. 17958 17959The @code{__builtin_dfp_dtstsfi_gt} function returns a non-zero value 17960if and only if the number of signficant digits of its @code{value} argument 17961is greater than its @code{comparison} argument. The 17962@code{__builtin_dfp_dtstsfi_gt_dd} and 17963@code{__builtin_dfp_dtstsfi_gt_td} functions behave similarly, but 17964require that the type of the @code{value} argument be 17965@code{__Decimal64} and @code{__Decimal128} respectively. 17966 17967The @code{__builtin_dfp_dtstsfi_eq} function returns a non-zero value 17968if and only if the number of signficant digits of its @code{value} argument 17969equals its @code{comparison} argument. The 17970@code{__builtin_dfp_dtstsfi_eq_dd} and 17971@code{__builtin_dfp_dtstsfi_eq_td} functions behave similarly, but 17972require that the type of the @code{value} argument be 17973@code{__Decimal64} and @code{__Decimal128} respectively. 17974 17975The @code{__builtin_dfp_dtstsfi_ov} function returns a non-zero value 17976if and only if its @code{value} argument has an undefined number of 17977significant digits, such as when @code{value} is an encoding of @code{NaN}. 17978The @code{__builtin_dfp_dtstsfi_ov_dd} and 17979@code{__builtin_dfp_dtstsfi_ov_td} functions behave similarly, but 17980require that the type of the @code{value} argument be 17981@code{__Decimal64} and @code{__Decimal128} respectively. 17982 17983The @code{__builtin_mffsl} uses the ISA 3.0 @code{mffsl} instruction to read 17984the FPSCR. The instruction is a lower latency version of the @code{mffs} 17985instruction. If the @code{mffsl} instruction is not available, then the 17986builtin uses the older @code{mffs} instruction to read the FPSCR. 17987 17988@node Basic PowerPC Built-in Functions Available on ISA 3.1 17989@subsubsection Basic PowerPC Built-in Functions Available on ISA 3.1 17990 17991The basic built-in functions described in this section are 17992available on the PowerPC family of processors starting with ISA 3.1. 17993Unless specific options are explicitly disabled on the 17994command line, specifying option @option{-mcpu=power10} has the effect of 17995enabling all the same options as for @option{-mcpu=power9}. 17996 17997The following built-in functions are available on Linux 64-bit systems 17998that use a future architecture instruction set (@option{-mcpu=power10}): 17999 18000@smallexample 18001@exdent unsigned long long int 18002@exdent __builtin_cfuged (unsigned long long int, unsigned long long int) 18003@end smallexample 18004Perform a 64-bit centrifuge operation, as if implemented by the 18005@code{cfuged} instruction. 18006@findex __builtin_cfuged 18007 18008@smallexample 18009@exdent unsigned long long int 18010@exdent __builtin_cntlzdm (unsigned long long int, unsigned long long int) 18011@end smallexample 18012Perform a 64-bit count leading zeros operation under mask, as if 18013implemented by the @code{cntlzdm} instruction. 18014@findex __builtin_cntlzdm 18015 18016@smallexample 18017@exdent unsigned long long int 18018@exdent __builtin_cnttzdm (unsigned long long int, unsigned long long int) 18019@end smallexample 18020Perform a 64-bit count trailing zeros operation under mask, as if 18021implemented by the @code{cnttzdm} instruction. 18022@findex __builtin_cnttzdm 18023 18024@smallexample 18025@exdent unsigned long long int 18026@exdent __builtin_pdepd (unsigned long long int, unsigned long long int) 18027@end smallexample 18028Perform a 64-bit parallel bits deposit operation, as if implemented by the 18029@code{pdepd} instruction. 18030@findex __builtin_pdepd 18031 18032@smallexample 18033@exdent unsigned long long int 18034@exdent __builtin_pextd (unsigned long long int, unsigned long long int) 18035@end smallexample 18036Perform a 64-bit parallel bits extract operation, as if implemented by the 18037@code{pextd} instruction. 18038@findex __builtin_pextd 18039 18040@smallexample 18041@exdent vector signed __int128 vsx_xl_sext (signed long long, signed char *); 18042@exdent vector signed __int128 vsx_xl_sext (signed long long, signed short *); 18043@exdent vector signed __int128 vsx_xl_sext (signed long long, signed int *); 18044@exdent vector signed __int128 vsx_xl_sext (signed long long, signed long long *); 18045@exdent vector unsigned __int128 vsx_xl_zext (signed long long, unsigned char *); 18046@exdent vector unsigned __int128 vsx_xl_zext (signed long long, unsigned short *); 18047@exdent vector unsigned __int128 vsx_xl_zext (signed long long, unsigned int *); 18048@exdent vector unsigned __int128 vsx_xl_zext (signed long long, unsigned long long *); 18049@end smallexample 18050 18051Load (and sign extend) to an __int128 vector, as if implemented by the ISA 3.1 18052@code{lxvrbx} @code{lxvrhx} @code{lxvrwx} @code{lxvrdx} instructions. 18053@findex vsx_xl_sext 18054@findex vsx_xl_zext 18055 18056@smallexample 18057@exdent void vec_xst_trunc (vector signed __int128, signed long long, signed char *); 18058@exdent void vec_xst_trunc (vector signed __int128, signed long long, signed short *); 18059@exdent void vec_xst_trunc (vector signed __int128, signed long long, signed int *); 18060@exdent void vec_xst_trunc (vector signed __int128, signed long long, signed long long *); 18061@exdent void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned char *); 18062@exdent void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned short *); 18063@exdent void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned int *); 18064@exdent void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned long long *); 18065@end smallexample 18066 18067Truncate and store the rightmost element of a vector, as if implemented by the 18068ISA 3.1 @code{stxvrbx} @code{stxvrhx} @code{stxvrwx} @code{stxvrdx} instructions. 18069@findex vec_xst_trunc 18070 18071@node PowerPC AltiVec/VSX Built-in Functions 18072@subsection PowerPC AltiVec/VSX Built-in Functions 18073 18074GCC provides an interface for the PowerPC family of processors to access 18075the AltiVec operations described in Motorola's AltiVec Programming 18076Interface Manual. The interface is made available by including 18077@code{<altivec.h>} and using @option{-maltivec} and 18078@option{-mabi=altivec}. The interface supports the following vector 18079types. 18080 18081@smallexample 18082vector unsigned char 18083vector signed char 18084vector bool char 18085 18086vector unsigned short 18087vector signed short 18088vector bool short 18089vector pixel 18090 18091vector unsigned int 18092vector signed int 18093vector bool int 18094vector float 18095@end smallexample 18096 18097GCC's implementation of the high-level language interface available from 18098C and C++ code differs from Motorola's documentation in several ways. 18099 18100@itemize @bullet 18101 18102@item 18103A vector constant is a list of constant expressions within curly braces. 18104 18105@item 18106A vector initializer requires no cast if the vector constant is of the 18107same type as the variable it is initializing. 18108 18109@item 18110If @code{signed} or @code{unsigned} is omitted, the signedness of the 18111vector type is the default signedness of the base type. The default 18112varies depending on the operating system, so a portable program should 18113always specify the signedness. 18114 18115@item 18116Compiling with @option{-maltivec} adds keywords @code{__vector}, 18117@code{vector}, @code{__pixel}, @code{pixel}, @code{__bool} and 18118@code{bool}. When compiling ISO C, the context-sensitive substitution 18119of the keywords @code{vector}, @code{pixel} and @code{bool} is 18120disabled. To use them, you must include @code{<altivec.h>} instead. 18121 18122@item 18123GCC allows using a @code{typedef} name as the type specifier for a 18124vector type, but only under the following circumstances: 18125 18126@itemize @bullet 18127 18128@item 18129When using @code{__vector} instead of @code{vector}; for example, 18130 18131@smallexample 18132typedef signed short int16; 18133__vector int16 data; 18134@end smallexample 18135 18136@item 18137When using @code{vector} in keyword-and-predefine mode; for example, 18138 18139@smallexample 18140typedef signed short int16; 18141vector int16 data; 18142@end smallexample 18143 18144Note that keyword-and-predefine mode is enabled by disabling GNU 18145extensions (e.g., by using @code{-std=c11}) and including 18146@code{<altivec.h>}. 18147@end itemize 18148 18149@item 18150For C, overloaded functions are implemented with macros so the following 18151does not work: 18152 18153@smallexample 18154 vec_add ((vector signed int)@{1, 2, 3, 4@}, foo); 18155@end smallexample 18156 18157@noindent 18158Since @code{vec_add} is a macro, the vector constant in the example 18159is treated as four separate arguments. Wrap the entire argument in 18160parentheses for this to work. 18161@end itemize 18162 18163@emph{Note:} Only the @code{<altivec.h>} interface is supported. 18164Internally, GCC uses built-in functions to achieve the functionality in 18165the aforementioned header file, but they are not supported and are 18166subject to change without notice. 18167 18168GCC complies with the Power Vector Intrinsic Programming Reference (PVIPR), 18169which may be found at 18170@uref{https://openpowerfoundation.org/?resource_lib=power-vector-intrinsic-programming-reference}. 18171Chapter 4 of this document fully documents the vector API interfaces 18172that must be 18173provided by compliant compilers. Programmers should preferentially use 18174the interfaces described therein. However, historically GCC has provided 18175additional interfaces for access to vector instructions. These are 18176briefly described below. Where the PVIPR provides a portable interface, 18177other functions in GCC that provide the same capabilities should be 18178considered deprecated. 18179 18180The PVIPR documents the following overloaded functions: 18181 18182@multitable @columnfractions 0.33 0.33 0.33 18183 18184@item @code{vec_abs} 18185@tab @code{vec_absd} 18186@tab @code{vec_abss} 18187@item @code{vec_add} 18188@tab @code{vec_addc} 18189@tab @code{vec_adde} 18190@item @code{vec_addec} 18191@tab @code{vec_adds} 18192@tab @code{vec_all_eq} 18193@item @code{vec_all_ge} 18194@tab @code{vec_all_gt} 18195@tab @code{vec_all_in} 18196@item @code{vec_all_le} 18197@tab @code{vec_all_lt} 18198@tab @code{vec_all_nan} 18199@item @code{vec_all_ne} 18200@tab @code{vec_all_nge} 18201@tab @code{vec_all_ngt} 18202@item @code{vec_all_nle} 18203@tab @code{vec_all_nlt} 18204@tab @code{vec_all_numeric} 18205@item @code{vec_and} 18206@tab @code{vec_andc} 18207@tab @code{vec_any_eq} 18208@item @code{vec_any_ge} 18209@tab @code{vec_any_gt} 18210@tab @code{vec_any_le} 18211@item @code{vec_any_lt} 18212@tab @code{vec_any_nan} 18213@tab @code{vec_any_ne} 18214@item @code{vec_any_nge} 18215@tab @code{vec_any_ngt} 18216@tab @code{vec_any_nle} 18217@item @code{vec_any_nlt} 18218@tab @code{vec_any_numeric} 18219@tab @code{vec_any_out} 18220@item @code{vec_avg} 18221@tab @code{vec_bperm} 18222@tab @code{vec_ceil} 18223@item @code{vec_cipher_be} 18224@tab @code{vec_cipherlast_be} 18225@tab @code{vec_cmpb} 18226@item @code{vec_cmpeq} 18227@tab @code{vec_cmpge} 18228@tab @code{vec_cmpgt} 18229@item @code{vec_cmple} 18230@tab @code{vec_cmplt} 18231@tab @code{vec_cmpne} 18232@item @code{vec_cmpnez} 18233@tab @code{vec_cntlz} 18234@tab @code{vec_cntlz_lsbb} 18235@item @code{vec_cnttz} 18236@tab @code{vec_cnttz_lsbb} 18237@tab @code{vec_cpsgn} 18238@item @code{vec_ctf} 18239@tab @code{vec_cts} 18240@tab @code{vec_ctu} 18241@item @code{vec_div} 18242@tab @code{vec_double} 18243@tab @code{vec_doublee} 18244@item @code{vec_doubleh} 18245@tab @code{vec_doublel} 18246@tab @code{vec_doubleo} 18247@item @code{vec_eqv} 18248@tab @code{vec_expte} 18249@tab @code{vec_extract} 18250@item @code{vec_extract_exp} 18251@tab @code{vec_extract_fp32_from_shorth} 18252@tab @code{vec_extract_fp32_from_shortl} 18253@item @code{vec_extract_sig} 18254@tab @code{vec_extract_4b} 18255@tab @code{vec_first_match_index} 18256@item @code{vec_first_match_or_eos_index} 18257@tab @code{vec_first_mismatch_index} 18258@tab @code{vec_first_mismatch_or_eos_index} 18259@item @code{vec_float} 18260@tab @code{vec_float2} 18261@tab @code{vec_floate} 18262@item @code{vec_floato} 18263@tab @code{vec_floor} 18264@tab @code{vec_gb} 18265@item @code{vec_insert} 18266@tab @code{vec_insert_exp} 18267@tab @code{vec_insert4b} 18268@item @code{vec_ld} 18269@tab @code{vec_lde} 18270@tab @code{vec_ldl} 18271@item @code{vec_loge} 18272@tab @code{vec_madd} 18273@tab @code{vec_madds} 18274@item @code{vec_max} 18275@tab @code{vec_mergee} 18276@tab @code{vec_mergeh} 18277@item @code{vec_mergel} 18278@tab @code{vec_mergeo} 18279@tab @code{vec_mfvscr} 18280@item @code{vec_min} 18281@tab @code{vec_mradds} 18282@tab @code{vec_msub} 18283@item @code{vec_msum} 18284@tab @code{vec_msums} 18285@tab @code{vec_mtvscr} 18286@item @code{vec_mul} 18287@tab @code{vec_mule} 18288@tab @code{vec_mulo} 18289@item @code{vec_nabs} 18290@tab @code{vec_nand} 18291@tab @code{vec_ncipher_be} 18292@item @code{vec_ncipherlast_be} 18293@tab @code{vec_nearbyint} 18294@tab @code{vec_neg} 18295@item @code{vec_nmadd} 18296@tab @code{vec_nmsub} 18297@tab @code{vec_nor} 18298@item @code{vec_or} 18299@tab @code{vec_orc} 18300@tab @code{vec_pack} 18301@item @code{vec_pack_to_short_fp32} 18302@tab @code{vec_packpx} 18303@tab @code{vec_packs} 18304@item @code{vec_packsu} 18305@tab @code{vec_parity_lsbb} 18306@tab @code{vec_perm} 18307@item @code{vec_permxor} 18308@tab @code{vec_pmsum_be} 18309@tab @code{vec_popcnt} 18310@item @code{vec_re} 18311@tab @code{vec_recipdiv} 18312@tab @code{vec_revb} 18313@item @code{vec_reve} 18314@tab @code{vec_rint} 18315@tab @code{vec_rl} 18316@item @code{vec_rlmi} 18317@tab @code{vec_rlnm} 18318@tab @code{vec_round} 18319@item @code{vec_rsqrt} 18320@tab @code{vec_rsqrte} 18321@tab @code{vec_sbox_be} 18322@item @code{vec_sel} 18323@tab @code{vec_shasigma_be} 18324@tab @code{vec_signed} 18325@item @code{vec_signed2} 18326@tab @code{vec_signede} 18327@tab @code{vec_signedo} 18328@item @code{vec_sl} 18329@tab @code{vec_sld} 18330@tab @code{vec_sldw} 18331@item @code{vec_sll} 18332@tab @code{vec_slo} 18333@tab @code{vec_slv} 18334@item @code{vec_splat} 18335@tab @code{vec_splat_s8} 18336@tab @code{vec_splat_s16} 18337@item @code{vec_splat_s32} 18338@tab @code{vec_splat_u8} 18339@tab @code{vec_splat_u16} 18340@item @code{vec_splat_u32} 18341@tab @code{vec_splats} 18342@tab @code{vec_sqrt} 18343@item @code{vec_sr} 18344@tab @code{vec_sra} 18345@tab @code{vec_srl} 18346@item @code{vec_sro} 18347@tab @code{vec_srv} 18348@tab @code{vec_st} 18349@item @code{vec_ste} 18350@tab @code{vec_stl} 18351@tab @code{vec_sub} 18352@item @code{vec_subc} 18353@tab @code{vec_sube} 18354@tab @code{vec_subec} 18355@item @code{vec_subs} 18356@tab @code{vec_sum2s} 18357@tab @code{vec_sum4s} 18358@item @code{vec_sums} 18359@tab @code{vec_test_data_class} 18360@tab @code{vec_trunc} 18361@item @code{vec_unpackh} 18362@tab @code{vec_unpackl} 18363@tab @code{vec_unsigned} 18364@item @code{vec_unsigned2} 18365@tab @code{vec_unsignede} 18366@tab @code{vec_unsignedo} 18367@item @code{vec_xl} 18368@tab @code{vec_xl_be} 18369@tab @code{vec_xl_len} 18370@item @code{vec_xl_len_r} 18371@tab @code{vec_xor} 18372@tab @code{vec_xst} 18373@item @code{vec_xst_be} 18374@tab @code{vec_xst_len} 18375@tab @code{vec_xst_len_r} 18376 18377@end multitable 18378 18379@menu 18380* PowerPC AltiVec Built-in Functions on ISA 2.05:: 18381* PowerPC AltiVec Built-in Functions Available on ISA 2.06:: 18382* PowerPC AltiVec Built-in Functions Available on ISA 2.07:: 18383* PowerPC AltiVec Built-in Functions Available on ISA 3.0:: 18384* PowerPC AltiVec Built-in Functions Available on ISA 3.1:: 18385@end menu 18386 18387@node PowerPC AltiVec Built-in Functions on ISA 2.05 18388@subsubsection PowerPC AltiVec Built-in Functions on ISA 2.05 18389 18390The following interfaces are supported for the generic and specific 18391AltiVec operations and the AltiVec predicates. In cases where there 18392is a direct mapping between generic and specific operations, only the 18393generic names are shown here, although the specific operations can also 18394be used. 18395 18396Arguments that are documented as @code{const int} require literal 18397integral values within the range required for that operation. 18398 18399Only functions excluded from the PVIPR are listed here. 18400 18401@smallexample 18402void vec_dss (const int); 18403 18404void vec_dssall (void); 18405 18406void vec_dst (const vector unsigned char *, int, const int); 18407void vec_dst (const vector signed char *, int, const int); 18408void vec_dst (const vector bool char *, int, const int); 18409void vec_dst (const vector unsigned short *, int, const int); 18410void vec_dst (const vector signed short *, int, const int); 18411void vec_dst (const vector bool short *, int, const int); 18412void vec_dst (const vector pixel *, int, const int); 18413void vec_dst (const vector unsigned int *, int, const int); 18414void vec_dst (const vector signed int *, int, const int); 18415void vec_dst (const vector bool int *, int, const int); 18416void vec_dst (const vector float *, int, const int); 18417void vec_dst (const unsigned char *, int, const int); 18418void vec_dst (const signed char *, int, const int); 18419void vec_dst (const unsigned short *, int, const int); 18420void vec_dst (const short *, int, const int); 18421void vec_dst (const unsigned int *, int, const int); 18422void vec_dst (const int *, int, const int); 18423void vec_dst (const float *, int, const int); 18424 18425void vec_dstst (const vector unsigned char *, int, const int); 18426void vec_dstst (const vector signed char *, int, const int); 18427void vec_dstst (const vector bool char *, int, const int); 18428void vec_dstst (const vector unsigned short *, int, const int); 18429void vec_dstst (const vector signed short *, int, const int); 18430void vec_dstst (const vector bool short *, int, const int); 18431void vec_dstst (const vector pixel *, int, const int); 18432void vec_dstst (const vector unsigned int *, int, const int); 18433void vec_dstst (const vector signed int *, int, const int); 18434void vec_dstst (const vector bool int *, int, const int); 18435void vec_dstst (const vector float *, int, const int); 18436void vec_dstst (const unsigned char *, int, const int); 18437void vec_dstst (const signed char *, int, const int); 18438void vec_dstst (const unsigned short *, int, const int); 18439void vec_dstst (const short *, int, const int); 18440void vec_dstst (const unsigned int *, int, const int); 18441void vec_dstst (const int *, int, const int); 18442void vec_dstst (const unsigned long *, int, const int); 18443void vec_dstst (const long *, int, const int); 18444void vec_dstst (const float *, int, const int); 18445 18446void vec_dststt (const vector unsigned char *, int, const int); 18447void vec_dststt (const vector signed char *, int, const int); 18448void vec_dststt (const vector bool char *, int, const int); 18449void vec_dststt (const vector unsigned short *, int, const int); 18450void vec_dststt (const vector signed short *, int, const int); 18451void vec_dststt (const vector bool short *, int, const int); 18452void vec_dststt (const vector pixel *, int, const int); 18453void vec_dststt (const vector unsigned int *, int, const int); 18454void vec_dststt (const vector signed int *, int, const int); 18455void vec_dststt (const vector bool int *, int, const int); 18456void vec_dststt (const vector float *, int, const int); 18457void vec_dststt (const unsigned char *, int, const int); 18458void vec_dststt (const signed char *, int, const int); 18459void vec_dststt (const unsigned short *, int, const int); 18460void vec_dststt (const short *, int, const int); 18461void vec_dststt (const unsigned int *, int, const int); 18462void vec_dststt (const int *, int, const int); 18463void vec_dststt (const float *, int, const int); 18464 18465void vec_dstt (const vector unsigned char *, int, const int); 18466void vec_dstt (const vector signed char *, int, const int); 18467void vec_dstt (const vector bool char *, int, const int); 18468void vec_dstt (const vector unsigned short *, int, const int); 18469void vec_dstt (const vector signed short *, int, const int); 18470void vec_dstt (const vector bool short *, int, const int); 18471void vec_dstt (const vector pixel *, int, const int); 18472void vec_dstt (const vector unsigned int *, int, const int); 18473void vec_dstt (const vector signed int *, int, const int); 18474void vec_dstt (const vector bool int *, int, const int); 18475void vec_dstt (const vector float *, int, const int); 18476void vec_dstt (const unsigned char *, int, const int); 18477void vec_dstt (const signed char *, int, const int); 18478void vec_dstt (const unsigned short *, int, const int); 18479void vec_dstt (const short *, int, const int); 18480void vec_dstt (const unsigned int *, int, const int); 18481void vec_dstt (const int *, int, const int); 18482void vec_dstt (const float *, int, const int); 18483 18484vector signed char vec_lvebx (int, char *); 18485vector unsigned char vec_lvebx (int, unsigned char *); 18486 18487vector signed short vec_lvehx (int, short *); 18488vector unsigned short vec_lvehx (int, unsigned short *); 18489 18490vector float vec_lvewx (int, float *); 18491vector signed int vec_lvewx (int, int *); 18492vector unsigned int vec_lvewx (int, unsigned int *); 18493 18494vector unsigned char vec_lvsl (int, const unsigned char *); 18495vector unsigned char vec_lvsl (int, const signed char *); 18496vector unsigned char vec_lvsl (int, const unsigned short *); 18497vector unsigned char vec_lvsl (int, const short *); 18498vector unsigned char vec_lvsl (int, const unsigned int *); 18499vector unsigned char vec_lvsl (int, const int *); 18500vector unsigned char vec_lvsl (int, const float *); 18501 18502vector unsigned char vec_lvsr (int, const unsigned char *); 18503vector unsigned char vec_lvsr (int, const signed char *); 18504vector unsigned char vec_lvsr (int, const unsigned short *); 18505vector unsigned char vec_lvsr (int, const short *); 18506vector unsigned char vec_lvsr (int, const unsigned int *); 18507vector unsigned char vec_lvsr (int, const int *); 18508vector unsigned char vec_lvsr (int, const float *); 18509 18510void vec_stvebx (vector signed char, int, signed char *); 18511void vec_stvebx (vector unsigned char, int, unsigned char *); 18512void vec_stvebx (vector bool char, int, signed char *); 18513void vec_stvebx (vector bool char, int, unsigned char *); 18514 18515void vec_stvehx (vector signed short, int, short *); 18516void vec_stvehx (vector unsigned short, int, unsigned short *); 18517void vec_stvehx (vector bool short, int, short *); 18518void vec_stvehx (vector bool short, int, unsigned short *); 18519 18520void vec_stvewx (vector float, int, float *); 18521void vec_stvewx (vector signed int, int, int *); 18522void vec_stvewx (vector unsigned int, int, unsigned int *); 18523void vec_stvewx (vector bool int, int, int *); 18524void vec_stvewx (vector bool int, int, unsigned int *); 18525 18526vector float vec_vaddfp (vector float, vector float); 18527 18528vector signed char vec_vaddsbs (vector bool char, vector signed char); 18529vector signed char vec_vaddsbs (vector signed char, vector bool char); 18530vector signed char vec_vaddsbs (vector signed char, vector signed char); 18531 18532vector signed short vec_vaddshs (vector bool short, vector signed short); 18533vector signed short vec_vaddshs (vector signed short, vector bool short); 18534vector signed short vec_vaddshs (vector signed short, vector signed short); 18535 18536vector signed int vec_vaddsws (vector bool int, vector signed int); 18537vector signed int vec_vaddsws (vector signed int, vector bool int); 18538vector signed int vec_vaddsws (vector signed int, vector signed int); 18539 18540vector signed char vec_vaddubm (vector bool char, vector signed char); 18541vector signed char vec_vaddubm (vector signed char, vector bool char); 18542vector signed char vec_vaddubm (vector signed char, vector signed char); 18543vector unsigned char vec_vaddubm (vector bool char, vector unsigned char); 18544vector unsigned char vec_vaddubm (vector unsigned char, vector bool char); 18545vector unsigned char vec_vaddubm (vector unsigned char, vector unsigned char); 18546 18547vector unsigned char vec_vaddubs (vector bool char, vector unsigned char); 18548vector unsigned char vec_vaddubs (vector unsigned char, vector bool char); 18549vector unsigned char vec_vaddubs (vector unsigned char, vector unsigned char); 18550 18551vector signed short vec_vadduhm (vector bool short, vector signed short); 18552vector signed short vec_vadduhm (vector signed short, vector bool short); 18553vector signed short vec_vadduhm (vector signed short, vector signed short); 18554vector unsigned short vec_vadduhm (vector bool short, vector unsigned short); 18555vector unsigned short vec_vadduhm (vector unsigned short, vector bool short); 18556vector unsigned short vec_vadduhm (vector unsigned short, vector unsigned short); 18557 18558vector unsigned short vec_vadduhs (vector bool short, vector unsigned short); 18559vector unsigned short vec_vadduhs (vector unsigned short, vector bool short); 18560vector unsigned short vec_vadduhs (vector unsigned short, vector unsigned short); 18561 18562vector signed int vec_vadduwm (vector bool int, vector signed int); 18563vector signed int vec_vadduwm (vector signed int, vector bool int); 18564vector signed int vec_vadduwm (vector signed int, vector signed int); 18565vector unsigned int vec_vadduwm (vector bool int, vector unsigned int); 18566vector unsigned int vec_vadduwm (vector unsigned int, vector bool int); 18567vector unsigned int vec_vadduwm (vector unsigned int, vector unsigned int); 18568 18569vector unsigned int vec_vadduws (vector bool int, vector unsigned int); 18570vector unsigned int vec_vadduws (vector unsigned int, vector bool int); 18571vector unsigned int vec_vadduws (vector unsigned int, vector unsigned int); 18572 18573vector signed char vec_vavgsb (vector signed char, vector signed char); 18574 18575vector signed short vec_vavgsh (vector signed short, vector signed short); 18576 18577vector signed int vec_vavgsw (vector signed int, vector signed int); 18578 18579vector unsigned char vec_vavgub (vector unsigned char, vector unsigned char); 18580 18581vector unsigned short vec_vavguh (vector unsigned short, vector unsigned short); 18582 18583vector unsigned int vec_vavguw (vector unsigned int, vector unsigned int); 18584 18585vector float vec_vcfsx (vector signed int, const int); 18586 18587vector float vec_vcfux (vector unsigned int, const int); 18588 18589vector bool int vec_vcmpeqfp (vector float, vector float); 18590 18591vector bool char vec_vcmpequb (vector signed char, vector signed char); 18592vector bool char vec_vcmpequb (vector unsigned char, vector unsigned char); 18593 18594vector bool short vec_vcmpequh (vector signed short, vector signed short); 18595vector bool short vec_vcmpequh (vector unsigned short, vector unsigned short); 18596 18597vector bool int vec_vcmpequw (vector signed int, vector signed int); 18598vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int); 18599 18600vector bool int vec_vcmpgtfp (vector float, vector float); 18601 18602vector bool char vec_vcmpgtsb (vector signed char, vector signed char); 18603 18604vector bool short vec_vcmpgtsh (vector signed short, vector signed short); 18605 18606vector bool int vec_vcmpgtsw (vector signed int, vector signed int); 18607 18608vector bool char vec_vcmpgtub (vector unsigned char, vector unsigned char); 18609 18610vector bool short vec_vcmpgtuh (vector unsigned short, vector unsigned short); 18611 18612vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int); 18613 18614vector float vec_vmaxfp (vector float, vector float); 18615 18616vector signed char vec_vmaxsb (vector bool char, vector signed char); 18617vector signed char vec_vmaxsb (vector signed char, vector bool char); 18618vector signed char vec_vmaxsb (vector signed char, vector signed char); 18619 18620vector signed short vec_vmaxsh (vector bool short, vector signed short); 18621vector signed short vec_vmaxsh (vector signed short, vector bool short); 18622vector signed short vec_vmaxsh (vector signed short, vector signed short); 18623 18624vector signed int vec_vmaxsw (vector bool int, vector signed int); 18625vector signed int vec_vmaxsw (vector signed int, vector bool int); 18626vector signed int vec_vmaxsw (vector signed int, vector signed int); 18627 18628vector unsigned char vec_vmaxub (vector bool char, vector unsigned char); 18629vector unsigned char vec_vmaxub (vector unsigned char, vector bool char); 18630vector unsigned char vec_vmaxub (vector unsigned char, vector unsigned char); 18631 18632vector unsigned short vec_vmaxuh (vector bool short, vector unsigned short); 18633vector unsigned short vec_vmaxuh (vector unsigned short, vector bool short); 18634vector unsigned short vec_vmaxuh (vector unsigned short, vector unsigned short); 18635 18636vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int); 18637vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int); 18638vector unsigned int vec_vmaxuw (vector unsigned int, vector unsigned int); 18639 18640vector float vec_vminfp (vector float, vector float); 18641 18642vector signed char vec_vminsb (vector bool char, vector signed char); 18643vector signed char vec_vminsb (vector signed char, vector bool char); 18644vector signed char vec_vminsb (vector signed char, vector signed char); 18645 18646vector signed short vec_vminsh (vector bool short, vector signed short); 18647vector signed short vec_vminsh (vector signed short, vector bool short); 18648vector signed short vec_vminsh (vector signed short, vector signed short); 18649 18650vector signed int vec_vminsw (vector bool int, vector signed int); 18651vector signed int vec_vminsw (vector signed int, vector bool int); 18652vector signed int vec_vminsw (vector signed int, vector signed int); 18653 18654vector unsigned char vec_vminub (vector bool char, vector unsigned char); 18655vector unsigned char vec_vminub (vector unsigned char, vector bool char); 18656vector unsigned char vec_vminub (vector unsigned char, vector unsigned char); 18657 18658vector unsigned short vec_vminuh (vector bool short, vector unsigned short); 18659vector unsigned short vec_vminuh (vector unsigned short, vector bool short); 18660vector unsigned short vec_vminuh (vector unsigned short, vector unsigned short); 18661 18662vector unsigned int vec_vminuw (vector bool int, vector unsigned int); 18663vector unsigned int vec_vminuw (vector unsigned int, vector bool int); 18664vector unsigned int vec_vminuw (vector unsigned int, vector unsigned int); 18665 18666vector bool char vec_vmrghb (vector bool char, vector bool char); 18667vector signed char vec_vmrghb (vector signed char, vector signed char); 18668vector unsigned char vec_vmrghb (vector unsigned char, vector unsigned char); 18669 18670vector bool short vec_vmrghh (vector bool short, vector bool short); 18671vector signed short vec_vmrghh (vector signed short, vector signed short); 18672vector unsigned short vec_vmrghh (vector unsigned short, vector unsigned short); 18673vector pixel vec_vmrghh (vector pixel, vector pixel); 18674 18675vector float vec_vmrghw (vector float, vector float); 18676vector bool int vec_vmrghw (vector bool int, vector bool int); 18677vector signed int vec_vmrghw (vector signed int, vector signed int); 18678vector unsigned int vec_vmrghw (vector unsigned int, vector unsigned int); 18679 18680vector bool char vec_vmrglb (vector bool char, vector bool char); 18681vector signed char vec_vmrglb (vector signed char, vector signed char); 18682vector unsigned char vec_vmrglb (vector unsigned char, vector unsigned char); 18683 18684vector bool short vec_vmrglh (vector bool short, vector bool short); 18685vector signed short vec_vmrglh (vector signed short, vector signed short); 18686vector unsigned short vec_vmrglh (vector unsigned short, vector unsigned short); 18687vector pixel vec_vmrglh (vector pixel, vector pixel); 18688 18689vector float vec_vmrglw (vector float, vector float); 18690vector signed int vec_vmrglw (vector signed int, vector signed int); 18691vector unsigned int vec_vmrglw (vector unsigned int, vector unsigned int); 18692vector bool int vec_vmrglw (vector bool int, vector bool int); 18693 18694vector signed int vec_vmsummbm (vector signed char, vector unsigned char, 18695 vector signed int); 18696 18697vector signed int vec_vmsumshm (vector signed short, vector signed short, 18698 vector signed int); 18699 18700vector signed int vec_vmsumshs (vector signed short, vector signed short, 18701 vector signed int); 18702 18703vector unsigned int vec_vmsumubm (vector unsigned char, vector unsigned char, 18704 vector unsigned int); 18705 18706vector unsigned int vec_vmsumuhm (vector unsigned short, vector unsigned short, 18707 vector unsigned int); 18708 18709vector unsigned int vec_vmsumuhs (vector unsigned short, vector unsigned short, 18710 vector unsigned int); 18711 18712vector signed short vec_vmulesb (vector signed char, vector signed char); 18713 18714vector signed int vec_vmulesh (vector signed short, vector signed short); 18715 18716vector unsigned short vec_vmuleub (vector unsigned char, vector unsigned char); 18717 18718vector unsigned int vec_vmuleuh (vector unsigned short, vector unsigned short); 18719 18720vector signed short vec_vmulosb (vector signed char, vector signed char); 18721 18722vector signed int vec_vmulosh (vector signed short, vector signed short); 18723 18724vector unsigned short vec_vmuloub (vector unsigned char, vector unsigned char); 18725 18726vector unsigned int vec_vmulouh (vector unsigned short, vector unsigned short); 18727 18728vector signed char vec_vpkshss (vector signed short, vector signed short); 18729 18730vector unsigned char vec_vpkshus (vector signed short, vector signed short); 18731 18732vector signed short vec_vpkswss (vector signed int, vector signed int); 18733 18734vector unsigned short vec_vpkswus (vector signed int, vector signed int); 18735 18736vector bool char vec_vpkuhum (vector bool short, vector bool short); 18737vector signed char vec_vpkuhum (vector signed short, vector signed short); 18738vector unsigned char vec_vpkuhum (vector unsigned short, vector unsigned short); 18739 18740vector unsigned char vec_vpkuhus (vector unsigned short, vector unsigned short); 18741 18742vector bool short vec_vpkuwum (vector bool int, vector bool int); 18743vector signed short vec_vpkuwum (vector signed int, vector signed int); 18744vector unsigned short vec_vpkuwum (vector unsigned int, vector unsigned int); 18745 18746vector unsigned short vec_vpkuwus (vector unsigned int, vector unsigned int); 18747 18748vector signed char vec_vrlb (vector signed char, vector unsigned char); 18749vector unsigned char vec_vrlb (vector unsigned char, vector unsigned char); 18750 18751vector signed short vec_vrlh (vector signed short, vector unsigned short); 18752vector unsigned short vec_vrlh (vector unsigned short, vector unsigned short); 18753 18754vector signed int vec_vrlw (vector signed int, vector unsigned int); 18755vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int); 18756 18757vector signed char vec_vslb (vector signed char, vector unsigned char); 18758vector unsigned char vec_vslb (vector unsigned char, vector unsigned char); 18759 18760vector signed short vec_vslh (vector signed short, vector unsigned short); 18761vector unsigned short vec_vslh (vector unsigned short, vector unsigned short); 18762 18763vector signed int vec_vslw (vector signed int, vector unsigned int); 18764vector unsigned int vec_vslw (vector unsigned int, vector unsigned int); 18765 18766vector signed char vec_vspltb (vector signed char, const int); 18767vector unsigned char vec_vspltb (vector unsigned char, const int); 18768vector bool char vec_vspltb (vector bool char, const int); 18769 18770vector bool short vec_vsplth (vector bool short, const int); 18771vector signed short vec_vsplth (vector signed short, const int); 18772vector unsigned short vec_vsplth (vector unsigned short, const int); 18773vector pixel vec_vsplth (vector pixel, const int); 18774 18775vector float vec_vspltw (vector float, const int); 18776vector signed int vec_vspltw (vector signed int, const int); 18777vector unsigned int vec_vspltw (vector unsigned int, const int); 18778vector bool int vec_vspltw (vector bool int, const int); 18779 18780vector signed char vec_vsrab (vector signed char, vector unsigned char); 18781vector unsigned char vec_vsrab (vector unsigned char, vector unsigned char); 18782 18783vector signed short vec_vsrah (vector signed short, vector unsigned short); 18784vector unsigned short vec_vsrah (vector unsigned short, vector unsigned short); 18785 18786vector signed int vec_vsraw (vector signed int, vector unsigned int); 18787vector unsigned int vec_vsraw (vector unsigned int, vector unsigned int); 18788 18789vector signed char vec_vsrb (vector signed char, vector unsigned char); 18790vector unsigned char vec_vsrb (vector unsigned char, vector unsigned char); 18791 18792vector signed short vec_vsrh (vector signed short, vector unsigned short); 18793vector unsigned short vec_vsrh (vector unsigned short, vector unsigned short); 18794 18795vector signed int vec_vsrw (vector signed int, vector unsigned int); 18796vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int); 18797 18798vector float vec_vsubfp (vector float, vector float); 18799 18800vector signed char vec_vsubsbs (vector bool char, vector signed char); 18801vector signed char vec_vsubsbs (vector signed char, vector bool char); 18802vector signed char vec_vsubsbs (vector signed char, vector signed char); 18803 18804vector signed short vec_vsubshs (vector bool short, vector signed short); 18805vector signed short vec_vsubshs (vector signed short, vector bool short); 18806vector signed short vec_vsubshs (vector signed short, vector signed short); 18807 18808vector signed int vec_vsubsws (vector bool int, vector signed int); 18809vector signed int vec_vsubsws (vector signed int, vector bool int); 18810vector signed int vec_vsubsws (vector signed int, vector signed int); 18811 18812vector signed char vec_vsububm (vector bool char, vector signed char); 18813vector signed char vec_vsububm (vector signed char, vector bool char); 18814vector signed char vec_vsububm (vector signed char, vector signed char); 18815vector unsigned char vec_vsububm (vector bool char, vector unsigned char); 18816vector unsigned char vec_vsububm (vector unsigned char, vector bool char); 18817vector unsigned char vec_vsububm (vector unsigned char, vector unsigned char); 18818 18819vector unsigned char vec_vsububs (vector bool char, vector unsigned char); 18820vector unsigned char vec_vsububs (vector unsigned char, vector bool char); 18821vector unsigned char vec_vsububs (vector unsigned char, vector unsigned char); 18822 18823vector signed short vec_vsubuhm (vector bool short, vector signed short); 18824vector signed short vec_vsubuhm (vector signed short, vector bool short); 18825vector signed short vec_vsubuhm (vector signed short, vector signed short); 18826vector unsigned short vec_vsubuhm (vector bool short, vector unsigned short); 18827vector unsigned short vec_vsubuhm (vector unsigned short, vector bool short); 18828vector unsigned short vec_vsubuhm (vector unsigned short, vector unsigned short); 18829 18830vector unsigned short vec_vsubuhs (vector bool short, vector unsigned short); 18831vector unsigned short vec_vsubuhs (vector unsigned short, vector bool short); 18832vector unsigned short vec_vsubuhs (vector unsigned short, vector unsigned short); 18833 18834vector signed int vec_vsubuwm (vector bool int, vector signed int); 18835vector signed int vec_vsubuwm (vector signed int, vector bool int); 18836vector signed int vec_vsubuwm (vector signed int, vector signed int); 18837vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int); 18838vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int); 18839vector unsigned int vec_vsubuwm (vector unsigned int, vector unsigned int); 18840 18841vector unsigned int vec_vsubuws (vector bool int, vector unsigned int); 18842vector unsigned int vec_vsubuws (vector unsigned int, vector bool int); 18843vector unsigned int vec_vsubuws (vector unsigned int, vector unsigned int); 18844 18845vector signed int vec_vsum4sbs (vector signed char, vector signed int); 18846 18847vector signed int vec_vsum4shs (vector signed short, vector signed int); 18848 18849vector unsigned int vec_vsum4ubs (vector unsigned char, vector unsigned int); 18850 18851vector unsigned int vec_vupkhpx (vector pixel); 18852 18853vector bool short vec_vupkhsb (vector bool char); 18854vector signed short vec_vupkhsb (vector signed char); 18855 18856vector bool int vec_vupkhsh (vector bool short); 18857vector signed int vec_vupkhsh (vector signed short); 18858 18859vector unsigned int vec_vupklpx (vector pixel); 18860 18861vector bool short vec_vupklsb (vector bool char); 18862vector signed short vec_vupklsb (vector signed char); 18863 18864vector bool int vec_vupklsh (vector bool short); 18865vector signed int vec_vupklsh (vector signed short); 18866@end smallexample 18867 18868@node PowerPC AltiVec Built-in Functions Available on ISA 2.06 18869@subsubsection PowerPC AltiVec Built-in Functions Available on ISA 2.06 18870 18871The AltiVec built-in functions described in this section are 18872available on the PowerPC family of processors starting with ISA 2.06 18873or later. These are normally enabled by adding @option{-mvsx} to the 18874command line. 18875 18876When @option{-mvsx} is used, the following additional vector types are 18877implemented. 18878 18879@smallexample 18880vector unsigned __int128 18881vector signed __int128 18882vector unsigned long long int 18883vector signed long long int 18884vector double 18885@end smallexample 18886 18887The long long types are only implemented for 64-bit code generation. 18888 18889Only functions excluded from the PVIPR are listed here. 18890 18891@smallexample 18892void vec_dst (const unsigned long *, int, const int); 18893void vec_dst (const long *, int, const int); 18894 18895void vec_dststt (const unsigned long *, int, const int); 18896void vec_dststt (const long *, int, const int); 18897 18898void vec_dstt (const unsigned long *, int, const int); 18899void vec_dstt (const long *, int, const int); 18900 18901vector unsigned char vec_lvsl (int, const unsigned long *); 18902vector unsigned char vec_lvsl (int, const long *); 18903 18904vector unsigned char vec_lvsr (int, const unsigned long *); 18905vector unsigned char vec_lvsr (int, const long *); 18906 18907vector unsigned char vec_lvsl (int, const double *); 18908vector unsigned char vec_lvsr (int, const double *); 18909 18910vector double vec_vsx_ld (int, const vector double *); 18911vector double vec_vsx_ld (int, const double *); 18912vector float vec_vsx_ld (int, const vector float *); 18913vector float vec_vsx_ld (int, const float *); 18914vector bool int vec_vsx_ld (int, const vector bool int *); 18915vector signed int vec_vsx_ld (int, const vector signed int *); 18916vector signed int vec_vsx_ld (int, const int *); 18917vector signed int vec_vsx_ld (int, const long *); 18918vector unsigned int vec_vsx_ld (int, const vector unsigned int *); 18919vector unsigned int vec_vsx_ld (int, const unsigned int *); 18920vector unsigned int vec_vsx_ld (int, const unsigned long *); 18921vector bool short vec_vsx_ld (int, const vector bool short *); 18922vector pixel vec_vsx_ld (int, const vector pixel *); 18923vector signed short vec_vsx_ld (int, const vector signed short *); 18924vector signed short vec_vsx_ld (int, const short *); 18925vector unsigned short vec_vsx_ld (int, const vector unsigned short *); 18926vector unsigned short vec_vsx_ld (int, const unsigned short *); 18927vector bool char vec_vsx_ld (int, const vector bool char *); 18928vector signed char vec_vsx_ld (int, const vector signed char *); 18929vector signed char vec_vsx_ld (int, const signed char *); 18930vector unsigned char vec_vsx_ld (int, const vector unsigned char *); 18931vector unsigned char vec_vsx_ld (int, const unsigned char *); 18932 18933void vec_vsx_st (vector double, int, vector double *); 18934void vec_vsx_st (vector double, int, double *); 18935void vec_vsx_st (vector float, int, vector float *); 18936void vec_vsx_st (vector float, int, float *); 18937void vec_vsx_st (vector signed int, int, vector signed int *); 18938void vec_vsx_st (vector signed int, int, int *); 18939void vec_vsx_st (vector unsigned int, int, vector unsigned int *); 18940void vec_vsx_st (vector unsigned int, int, unsigned int *); 18941void vec_vsx_st (vector bool int, int, vector bool int *); 18942void vec_vsx_st (vector bool int, int, unsigned int *); 18943void vec_vsx_st (vector bool int, int, int *); 18944void vec_vsx_st (vector signed short, int, vector signed short *); 18945void vec_vsx_st (vector signed short, int, short *); 18946void vec_vsx_st (vector unsigned short, int, vector unsigned short *); 18947void vec_vsx_st (vector unsigned short, int, unsigned short *); 18948void vec_vsx_st (vector bool short, int, vector bool short *); 18949void vec_vsx_st (vector bool short, int, unsigned short *); 18950void vec_vsx_st (vector pixel, int, vector pixel *); 18951void vec_vsx_st (vector pixel, int, unsigned short *); 18952void vec_vsx_st (vector pixel, int, short *); 18953void vec_vsx_st (vector bool short, int, short *); 18954void vec_vsx_st (vector signed char, int, vector signed char *); 18955void vec_vsx_st (vector signed char, int, signed char *); 18956void vec_vsx_st (vector unsigned char, int, vector unsigned char *); 18957void vec_vsx_st (vector unsigned char, int, unsigned char *); 18958void vec_vsx_st (vector bool char, int, vector bool char *); 18959void vec_vsx_st (vector bool char, int, unsigned char *); 18960void vec_vsx_st (vector bool char, int, signed char *); 18961 18962vector double vec_xxpermdi (vector double, vector double, const int); 18963vector float vec_xxpermdi (vector float, vector float, const int); 18964vector long long vec_xxpermdi (vector long long, vector long long, const int); 18965vector unsigned long long vec_xxpermdi (vector unsigned long long, 18966 vector unsigned long long, const int); 18967vector int vec_xxpermdi (vector int, vector int, const int); 18968vector unsigned int vec_xxpermdi (vector unsigned int, 18969 vector unsigned int, const int); 18970vector short vec_xxpermdi (vector short, vector short, const int); 18971vector unsigned short vec_xxpermdi (vector unsigned short, 18972 vector unsigned short, const int); 18973vector signed char vec_xxpermdi (vector signed char, vector signed char, 18974 const int); 18975vector unsigned char vec_xxpermdi (vector unsigned char, 18976 vector unsigned char, const int); 18977 18978vector double vec_xxsldi (vector double, vector double, int); 18979vector float vec_xxsldi (vector float, vector float, int); 18980vector long long vec_xxsldi (vector long long, vector long long, int); 18981vector unsigned long long vec_xxsldi (vector unsigned long long, 18982 vector unsigned long long, int); 18983vector int vec_xxsldi (vector int, vector int, int); 18984vector unsigned int vec_xxsldi (vector unsigned int, vector unsigned int, int); 18985vector short vec_xxsldi (vector short, vector short, int); 18986vector unsigned short vec_xxsldi (vector unsigned short, 18987 vector unsigned short, int); 18988vector signed char vec_xxsldi (vector signed char, vector signed char, int); 18989vector unsigned char vec_xxsldi (vector unsigned char, 18990 vector unsigned char, int); 18991@end smallexample 18992 18993Note that the @samp{vec_ld} and @samp{vec_st} built-in functions always 18994generate the AltiVec @samp{LVX} and @samp{STVX} instructions even 18995if the VSX instruction set is available. The @samp{vec_vsx_ld} and 18996@samp{vec_vsx_st} built-in functions always generate the VSX @samp{LXVD2X}, 18997@samp{LXVW4X}, @samp{STXVD2X}, and @samp{STXVW4X} instructions. 18998 18999@node PowerPC AltiVec Built-in Functions Available on ISA 2.07 19000@subsubsection PowerPC AltiVec Built-in Functions Available on ISA 2.07 19001 19002If the ISA 2.07 additions to the vector/scalar (power8-vector) 19003instruction set are available, the following additional functions are 19004available for both 32-bit and 64-bit targets. For 64-bit targets, you 19005can use @var{vector long} instead of @var{vector long long}, 19006@var{vector bool long} instead of @var{vector bool long long}, and 19007@var{vector unsigned long} instead of @var{vector unsigned long long}. 19008 19009Only functions excluded from the PVIPR are listed here. 19010 19011@smallexample 19012vector long long vec_vaddudm (vector long long, vector long long); 19013vector long long vec_vaddudm (vector bool long long, vector long long); 19014vector long long vec_vaddudm (vector long long, vector bool long long); 19015vector unsigned long long vec_vaddudm (vector unsigned long long, 19016 vector unsigned long long); 19017vector unsigned long long vec_vaddudm (vector bool unsigned long long, 19018 vector unsigned long long); 19019vector unsigned long long vec_vaddudm (vector unsigned long long, 19020 vector bool unsigned long long); 19021 19022vector long long vec_vclz (vector long long); 19023vector unsigned long long vec_vclz (vector unsigned long long); 19024vector int vec_vclz (vector int); 19025vector unsigned int vec_vclz (vector int); 19026vector short vec_vclz (vector short); 19027vector unsigned short vec_vclz (vector unsigned short); 19028vector signed char vec_vclz (vector signed char); 19029vector unsigned char vec_vclz (vector unsigned char); 19030 19031vector signed char vec_vclzb (vector signed char); 19032vector unsigned char vec_vclzb (vector unsigned char); 19033 19034vector long long vec_vclzd (vector long long); 19035vector unsigned long long vec_vclzd (vector unsigned long long); 19036 19037vector short vec_vclzh (vector short); 19038vector unsigned short vec_vclzh (vector unsigned short); 19039 19040vector int vec_vclzw (vector int); 19041vector unsigned int vec_vclzw (vector int); 19042 19043vector signed char vec_vgbbd (vector signed char); 19044vector unsigned char vec_vgbbd (vector unsigned char); 19045 19046vector long long vec_vmaxsd (vector long long, vector long long); 19047 19048vector unsigned long long vec_vmaxud (vector unsigned long long, 19049 unsigned vector long long); 19050 19051vector long long vec_vminsd (vector long long, vector long long); 19052 19053vector unsigned long long vec_vminud (vector long long, vector long long); 19054 19055vector int vec_vpksdss (vector long long, vector long long); 19056vector unsigned int vec_vpksdss (vector long long, vector long long); 19057 19058vector unsigned int vec_vpkudus (vector unsigned long long, 19059 vector unsigned long long); 19060 19061vector int vec_vpkudum (vector long long, vector long long); 19062vector unsigned int vec_vpkudum (vector unsigned long long, 19063 vector unsigned long long); 19064vector bool int vec_vpkudum (vector bool long long, vector bool long long); 19065 19066vector long long vec_vpopcnt (vector long long); 19067vector unsigned long long vec_vpopcnt (vector unsigned long long); 19068vector int vec_vpopcnt (vector int); 19069vector unsigned int vec_vpopcnt (vector int); 19070vector short vec_vpopcnt (vector short); 19071vector unsigned short vec_vpopcnt (vector unsigned short); 19072vector signed char vec_vpopcnt (vector signed char); 19073vector unsigned char vec_vpopcnt (vector unsigned char); 19074 19075vector signed char vec_vpopcntb (vector signed char); 19076vector unsigned char vec_vpopcntb (vector unsigned char); 19077 19078vector long long vec_vpopcntd (vector long long); 19079vector unsigned long long vec_vpopcntd (vector unsigned long long); 19080 19081vector short vec_vpopcnth (vector short); 19082vector unsigned short vec_vpopcnth (vector unsigned short); 19083 19084vector int vec_vpopcntw (vector int); 19085vector unsigned int vec_vpopcntw (vector int); 19086 19087vector long long vec_vrld (vector long long, vector unsigned long long); 19088vector unsigned long long vec_vrld (vector unsigned long long, 19089 vector unsigned long long); 19090 19091vector long long vec_vsld (vector long long, vector unsigned long long); 19092vector long long vec_vsld (vector unsigned long long, 19093 vector unsigned long long); 19094 19095vector long long vec_vsrad (vector long long, vector unsigned long long); 19096vector unsigned long long vec_vsrad (vector unsigned long long, 19097 vector unsigned long long); 19098 19099vector long long vec_vsrd (vector long long, vector unsigned long long); 19100vector unsigned long long char vec_vsrd (vector unsigned long long, 19101 vector unsigned long long); 19102 19103vector long long vec_vsubudm (vector long long, vector long long); 19104vector long long vec_vsubudm (vector bool long long, vector long long); 19105vector long long vec_vsubudm (vector long long, vector bool long long); 19106vector unsigned long long vec_vsubudm (vector unsigned long long, 19107 vector unsigned long long); 19108vector unsigned long long vec_vsubudm (vector bool long long, 19109 vector unsigned long long); 19110vector unsigned long long vec_vsubudm (vector unsigned long long, 19111 vector bool long long); 19112 19113vector long long vec_vupkhsw (vector int); 19114vector unsigned long long vec_vupkhsw (vector unsigned int); 19115 19116vector long long vec_vupklsw (vector int); 19117vector unsigned long long vec_vupklsw (vector int); 19118@end smallexample 19119 19120If the ISA 2.07 additions to the vector/scalar (power8-vector) 19121instruction set are available, the following additional functions are 19122available for 64-bit targets. New vector types 19123(@var{vector __int128} and @var{vector __uint128}) are available 19124to hold the @var{__int128} and @var{__uint128} types to use these 19125builtins. 19126 19127The normal vector extract, and set operations work on 19128@var{vector __int128} and @var{vector __uint128} types, 19129but the index value must be 0. 19130 19131Only functions excluded from the PVIPR are listed here. 19132 19133@smallexample 19134vector __int128 vec_vaddcuq (vector __int128, vector __int128); 19135vector __uint128 vec_vaddcuq (vector __uint128, vector __uint128); 19136 19137vector __int128 vec_vadduqm (vector __int128, vector __int128); 19138vector __uint128 vec_vadduqm (vector __uint128, vector __uint128); 19139 19140vector __int128 vec_vaddecuq (vector __int128, vector __int128, 19141 vector __int128); 19142vector __uint128 vec_vaddecuq (vector __uint128, vector __uint128, 19143 vector __uint128); 19144 19145vector __int128 vec_vaddeuqm (vector __int128, vector __int128, 19146 vector __int128); 19147vector __uint128 vec_vaddeuqm (vector __uint128, vector __uint128, 19148 vector __uint128); 19149 19150vector __int128 vec_vsubecuq (vector __int128, vector __int128, 19151 vector __int128); 19152vector __uint128 vec_vsubecuq (vector __uint128, vector __uint128, 19153 vector __uint128); 19154 19155vector __int128 vec_vsubeuqm (vector __int128, vector __int128, 19156 vector __int128); 19157vector __uint128 vec_vsubeuqm (vector __uint128, vector __uint128, 19158 vector __uint128); 19159 19160vector __int128 vec_vsubcuq (vector __int128, vector __int128); 19161vector __uint128 vec_vsubcuq (vector __uint128, vector __uint128); 19162 19163__int128 vec_vsubuqm (__int128, __int128); 19164__uint128 vec_vsubuqm (__uint128, __uint128); 19165 19166vector __int128 __builtin_bcdadd (vector __int128, vector __int128, const int); 19167vector unsigned char __builtin_bcdadd (vector unsigned char, vector unsigned char, 19168 const int); 19169int __builtin_bcdadd_lt (vector __int128, vector __int128, const int); 19170int __builtin_bcdadd_lt (vector unsigned char, vector unsigned char, const int); 19171int __builtin_bcdadd_eq (vector __int128, vector __int128, const int); 19172int __builtin_bcdadd_eq (vector unsigned char, vector unsigned char, const int); 19173int __builtin_bcdadd_gt (vector __int128, vector __int128, const int); 19174int __builtin_bcdadd_gt (vector unsigned char, vector unsigned char, const int); 19175int __builtin_bcdadd_ov (vector __int128, vector __int128, const int); 19176int __builtin_bcdadd_ov (vector unsigned char, vector unsigned char, const int); 19177 19178vector __int128 __builtin_bcdsub (vector __int128, vector __int128, const int); 19179vector unsigned char __builtin_bcdsub (vector unsigned char, vector unsigned char, 19180 const int); 19181int __builtin_bcdsub_lt (vector __int128, vector __int128, const int); 19182int __builtin_bcdsub_lt (vector unsigned char, vector unsigned char, const int); 19183int __builtin_bcdsub_eq (vector __int128, vector __int128, const int); 19184int __builtin_bcdsub_eq (vector unsigned char, vector unsigned char, const int); 19185int __builtin_bcdsub_gt (vector __int128, vector __int128, const int); 19186int __builtin_bcdsub_gt (vector unsigned char, vector unsigned char, const int); 19187int __builtin_bcdsub_ov (vector __int128, vector __int128, const int); 19188int __builtin_bcdsub_ov (vector unsigned char, vector unsigned char, const int); 19189@end smallexample 19190 19191@node PowerPC AltiVec Built-in Functions Available on ISA 3.0 19192@subsubsection PowerPC AltiVec Built-in Functions Available on ISA 3.0 19193 19194The following additional built-in functions are also available for the 19195PowerPC family of processors, starting with ISA 3.0 19196(@option{-mcpu=power9}) or later. 19197 19198Only instructions excluded from the PVIPR are listed here. 19199 19200@smallexample 19201unsigned int scalar_extract_exp (double source); 19202unsigned long long int scalar_extract_exp (__ieee128 source); 19203 19204unsigned long long int scalar_extract_sig (double source); 19205unsigned __int128 scalar_extract_sig (__ieee128 source); 19206 19207double scalar_insert_exp (unsigned long long int significand, 19208 unsigned long long int exponent); 19209double scalar_insert_exp (double significand, unsigned long long int exponent); 19210 19211ieee_128 scalar_insert_exp (unsigned __int128 significand, 19212 unsigned long long int exponent); 19213ieee_128 scalar_insert_exp (ieee_128 significand, unsigned long long int exponent); 19214 19215int scalar_cmp_exp_gt (double arg1, double arg2); 19216int scalar_cmp_exp_lt (double arg1, double arg2); 19217int scalar_cmp_exp_eq (double arg1, double arg2); 19218int scalar_cmp_exp_unordered (double arg1, double arg2); 19219 19220bool scalar_test_data_class (float source, const int condition); 19221bool scalar_test_data_class (double source, const int condition); 19222bool scalar_test_data_class (__ieee128 source, const int condition); 19223 19224bool scalar_test_neg (float source); 19225bool scalar_test_neg (double source); 19226bool scalar_test_neg (__ieee128 source); 19227@end smallexample 19228 19229The @code{scalar_extract_exp} and @code{scalar_extract_sig} 19230functions require a 64-bit environment supporting ISA 3.0 or later. 19231The @code{scalar_extract_exp} and @code{scalar_extract_sig} built-in 19232functions return the significand and the biased exponent value 19233respectively of their @code{source} arguments. 19234When supplied with a 64-bit @code{source} argument, the 19235result returned by @code{scalar_extract_sig} has 19236the @code{0x0010000000000000} bit set if the 19237function's @code{source} argument is in normalized form. 19238Otherwise, this bit is set to 0. 19239When supplied with a 128-bit @code{source} argument, the 19240@code{0x00010000000000000000000000000000} bit of the result is 19241treated similarly. 19242Note that the sign of the significand is not represented in the result 19243returned from the @code{scalar_extract_sig} function. Use the 19244@code{scalar_test_neg} function to test the sign of its @code{double} 19245argument. 19246 19247The @code{scalar_insert_exp} 19248functions require a 64-bit environment supporting ISA 3.0 or later. 19249When supplied with a 64-bit first argument, the 19250@code{scalar_insert_exp} built-in function returns a double-precision 19251floating point value that is constructed by assembling the values of its 19252@code{significand} and @code{exponent} arguments. The sign of the 19253result is copied from the most significant bit of the 19254@code{significand} argument. The significand and exponent components 19255of the result are composed of the least significant 11 bits of the 19256@code{exponent} argument and the least significant 52 bits of the 19257@code{significand} argument respectively. 19258 19259When supplied with a 128-bit first argument, the 19260@code{scalar_insert_exp} built-in function returns a quad-precision 19261ieee floating point value. The sign bit of the result is copied from 19262the most significant bit of the @code{significand} argument. 19263The significand and exponent components of the result are composed of 19264the least significant 15 bits of the @code{exponent} argument and the 19265least significant 112 bits of the @code{significand} argument respectively. 19266 19267The @code{scalar_cmp_exp_gt}, @code{scalar_cmp_exp_lt}, 19268@code{scalar_cmp_exp_eq}, and @code{scalar_cmp_exp_unordered} built-in 19269functions return a non-zero value if @code{arg1} is greater than, less 19270than, equal to, or not comparable to @code{arg2} respectively. The 19271arguments are not comparable if one or the other equals NaN (not a 19272number). 19273 19274The @code{scalar_test_data_class} built-in function returns 1 19275if any of the condition tests enabled by the value of the 19276@code{condition} variable are true, and 0 otherwise. The 19277@code{condition} argument must be a compile-time constant integer with 19278value not exceeding 127. The 19279@code{condition} argument is encoded as a bitmask with each bit 19280enabling the testing of a different condition, as characterized by the 19281following: 19282@smallexample 192830x40 Test for NaN 192840x20 Test for +Infinity 192850x10 Test for -Infinity 192860x08 Test for +Zero 192870x04 Test for -Zero 192880x02 Test for +Denormal 192890x01 Test for -Denormal 19290@end smallexample 19291 19292The @code{scalar_test_neg} built-in function returns 1 if its 19293@code{source} argument holds a negative value, 0 otherwise. 19294 19295The following built-in functions are also available for the PowerPC family 19296of processors, starting with ISA 3.0 or later 19297(@option{-mcpu=power9}). These string functions are described 19298separately in order to group the descriptions closer to the function 19299prototypes. 19300 19301Only functions excluded from the PVIPR are listed here. 19302 19303@smallexample 19304int vec_all_nez (vector signed char, vector signed char); 19305int vec_all_nez (vector unsigned char, vector unsigned char); 19306int vec_all_nez (vector signed short, vector signed short); 19307int vec_all_nez (vector unsigned short, vector unsigned short); 19308int vec_all_nez (vector signed int, vector signed int); 19309int vec_all_nez (vector unsigned int, vector unsigned int); 19310 19311int vec_any_eqz (vector signed char, vector signed char); 19312int vec_any_eqz (vector unsigned char, vector unsigned char); 19313int vec_any_eqz (vector signed short, vector signed short); 19314int vec_any_eqz (vector unsigned short, vector unsigned short); 19315int vec_any_eqz (vector signed int, vector signed int); 19316int vec_any_eqz (vector unsigned int, vector unsigned int); 19317 19318signed char vec_xlx (unsigned int index, vector signed char data); 19319unsigned char vec_xlx (unsigned int index, vector unsigned char data); 19320signed short vec_xlx (unsigned int index, vector signed short data); 19321unsigned short vec_xlx (unsigned int index, vector unsigned short data); 19322signed int vec_xlx (unsigned int index, vector signed int data); 19323unsigned int vec_xlx (unsigned int index, vector unsigned int data); 19324float vec_xlx (unsigned int index, vector float data); 19325 19326signed char vec_xrx (unsigned int index, vector signed char data); 19327unsigned char vec_xrx (unsigned int index, vector unsigned char data); 19328signed short vec_xrx (unsigned int index, vector signed short data); 19329unsigned short vec_xrx (unsigned int index, vector unsigned short data); 19330signed int vec_xrx (unsigned int index, vector signed int data); 19331unsigned int vec_xrx (unsigned int index, vector unsigned int data); 19332float vec_xrx (unsigned int index, vector float data); 19333@end smallexample 19334 19335The @code{vec_all_nez}, @code{vec_any_eqz}, and @code{vec_cmpnez} 19336perform pairwise comparisons between the elements at the same 19337positions within their two vector arguments. 19338The @code{vec_all_nez} function returns a 19339non-zero value if and only if all pairwise comparisons are not 19340equal and no element of either vector argument contains a zero. 19341The @code{vec_any_eqz} function returns a 19342non-zero value if and only if at least one pairwise comparison is equal 19343or if at least one element of either vector argument contains a zero. 19344The @code{vec_cmpnez} function returns a vector of the same type as 19345its two arguments, within which each element consists of all ones to 19346denote that either the corresponding elements of the incoming arguments are 19347not equal or that at least one of the corresponding elements contains 19348zero. Otherwise, the element of the returned vector contains all zeros. 19349 19350The @code{vec_xlx} and @code{vec_xrx} functions extract the single 19351element selected by the @code{index} argument from the vector 19352represented by the @code{data} argument. The @code{index} argument 19353always specifies a byte offset, regardless of the size of the vector 19354element. With @code{vec_xlx}, @code{index} is the offset of the first 19355byte of the element to be extracted. With @code{vec_xrx}, @code{index} 19356represents the last byte of the element to be extracted, measured 19357from the right end of the vector. In other words, the last byte of 19358the element to be extracted is found at position @code{(15 - index)}. 19359There is no requirement that @code{index} be a multiple of the vector 19360element size. However, if the size of the vector element added to 19361@code{index} is greater than 15, the content of the returned value is 19362undefined. 19363 19364The following functions are also available if the ISA 3.0 instruction 19365set additions (@option{-mcpu=power9}) are available. 19366 19367Only functions excluded from the PVIPR are listed here. 19368 19369@smallexample 19370vector long long vec_vctz (vector long long); 19371vector unsigned long long vec_vctz (vector unsigned long long); 19372vector int vec_vctz (vector int); 19373vector unsigned int vec_vctz (vector int); 19374vector short vec_vctz (vector short); 19375vector unsigned short vec_vctz (vector unsigned short); 19376vector signed char vec_vctz (vector signed char); 19377vector unsigned char vec_vctz (vector unsigned char); 19378 19379vector signed char vec_vctzb (vector signed char); 19380vector unsigned char vec_vctzb (vector unsigned char); 19381 19382vector long long vec_vctzd (vector long long); 19383vector unsigned long long vec_vctzd (vector unsigned long long); 19384 19385vector short vec_vctzh (vector short); 19386vector unsigned short vec_vctzh (vector unsigned short); 19387 19388vector int vec_vctzw (vector int); 19389vector unsigned int vec_vctzw (vector int); 19390 19391vector int vec_vprtyb (vector int); 19392vector unsigned int vec_vprtyb (vector unsigned int); 19393vector long long vec_vprtyb (vector long long); 19394vector unsigned long long vec_vprtyb (vector unsigned long long); 19395 19396vector int vec_vprtybw (vector int); 19397vector unsigned int vec_vprtybw (vector unsigned int); 19398 19399vector long long vec_vprtybd (vector long long); 19400vector unsigned long long vec_vprtybd (vector unsigned long long); 19401@end smallexample 19402 19403On 64-bit targets, if the ISA 3.0 additions (@option{-mcpu=power9}) 19404are available: 19405 19406@smallexample 19407vector long vec_vprtyb (vector long); 19408vector unsigned long vec_vprtyb (vector unsigned long); 19409vector __int128 vec_vprtyb (vector __int128); 19410vector __uint128 vec_vprtyb (vector __uint128); 19411 19412vector long vec_vprtybd (vector long); 19413vector unsigned long vec_vprtybd (vector unsigned long); 19414 19415vector __int128 vec_vprtybq (vector __int128); 19416vector __uint128 vec_vprtybd (vector __uint128); 19417@end smallexample 19418 19419The following built-in functions are available for the PowerPC family 19420of processors, starting with ISA 3.0 or later (@option{-mcpu=power9}). 19421 19422Only functions excluded from the PVIPR are listed here. 19423 19424@smallexample 19425__vector unsigned char 19426vec_absdb (__vector unsigned char arg1, __vector unsigned char arg2); 19427__vector unsigned short 19428vec_absdh (__vector unsigned short arg1, __vector unsigned short arg2); 19429__vector unsigned int 19430vec_absdw (__vector unsigned int arg1, __vector unsigned int arg2); 19431@end smallexample 19432 19433The @code{vec_absd}, @code{vec_absdb}, @code{vec_absdh}, and 19434@code{vec_absdw} built-in functions each computes the absolute 19435differences of the pairs of vector elements supplied in its two vector 19436arguments, placing the absolute differences into the corresponding 19437elements of the vector result. 19438 19439The following built-in functions are available for the PowerPC family 19440of processors, starting with ISA 3.0 or later (@option{-mcpu=power9}): 19441@smallexample 19442vector unsigned int vec_vrlnm (vector unsigned int, vector unsigned int); 19443vector unsigned long long vec_vrlnm (vector unsigned long long, 19444 vector unsigned long long); 19445@end smallexample 19446 19447The result of @code{vec_vrlnm} is obtained by rotating each element 19448of the first argument vector left and ANDing it with a mask. The 19449second argument vector contains the mask beginning in bits 11:15, 19450the mask end in bits 19:23, and the shift count in bits 27:31, 19451of each element. 19452 19453If the cryptographic instructions are enabled (@option{-mcrypto} or 19454@option{-mcpu=power8}), the following builtins are enabled. 19455 19456Only functions excluded from the PVIPR are listed here. 19457 19458@smallexample 19459vector unsigned long long __builtin_crypto_vsbox (vector unsigned long long); 19460 19461vector unsigned long long __builtin_crypto_vcipher (vector unsigned long long, 19462 vector unsigned long long); 19463 19464vector unsigned long long __builtin_crypto_vcipherlast 19465 (vector unsigned long long, 19466 vector unsigned long long); 19467 19468vector unsigned long long __builtin_crypto_vncipher (vector unsigned long long, 19469 vector unsigned long long); 19470 19471vector unsigned long long __builtin_crypto_vncipherlast (vector unsigned long long, 19472 vector unsigned long long); 19473 19474vector unsigned char __builtin_crypto_vpermxor (vector unsigned char, 19475 vector unsigned char, 19476 vector unsigned char); 19477 19478vector unsigned short __builtin_crypto_vpermxor (vector unsigned short, 19479 vector unsigned short, 19480 vector unsigned short); 19481 19482vector unsigned int __builtin_crypto_vpermxor (vector unsigned int, 19483 vector unsigned int, 19484 vector unsigned int); 19485 19486vector unsigned long long __builtin_crypto_vpermxor (vector unsigned long long, 19487 vector unsigned long long, 19488 vector unsigned long long); 19489 19490vector unsigned char __builtin_crypto_vpmsumb (vector unsigned char, 19491 vector unsigned char); 19492 19493vector unsigned short __builtin_crypto_vpmsumh (vector unsigned short, 19494 vector unsigned short); 19495 19496vector unsigned int __builtin_crypto_vpmsumw (vector unsigned int, 19497 vector unsigned int); 19498 19499vector unsigned long long __builtin_crypto_vpmsumd (vector unsigned long long, 19500 vector unsigned long long); 19501 19502vector unsigned long long __builtin_crypto_vshasigmad (vector unsigned long long, 19503 int, int); 19504 19505vector unsigned int __builtin_crypto_vshasigmaw (vector unsigned int, int, int); 19506@end smallexample 19507 19508The second argument to @var{__builtin_crypto_vshasigmad} and 19509@var{__builtin_crypto_vshasigmaw} must be a constant 19510integer that is 0 or 1. The third argument to these built-in functions 19511must be a constant integer in the range of 0 to 15. 19512 19513The following sign extension builtins are provided: 19514 19515@smallexample 19516vector signed int vec_signexti (vector signed char a) 19517vector signed long long vec_signextll (vector signed char a) 19518vector signed int vec_signexti (vector signed short a) 19519vector signed long long vec_signextll (vector signed short a) 19520vector signed long long vec_signextll (vector signed int a) 19521vector signed long long vec_signextq (vector signed long long a) 19522@end smallexample 19523 19524Each element of the result is produced by sign-extending the element of the 19525input vector that would fall in the least significant portion of the result 19526element. For example, a sign-extension of a vector signed char to a vector 19527signed long long will sign extend the rightmost byte of each doubleword. 19528 19529@node PowerPC AltiVec Built-in Functions Available on ISA 3.1 19530@subsubsection PowerPC AltiVec Built-in Functions Available on ISA 3.1 19531 19532The following additional built-in functions are also available for the 19533PowerPC family of processors, starting with ISA 3.1 (@option{-mcpu=power10}): 19534 19535 19536@smallexample 19537@exdent vector unsigned long long int 19538@exdent vec_cfuge (vector unsigned long long int, vector unsigned long long int) 19539@end smallexample 19540Perform a vector centrifuge operation, as if implemented by the 19541@code{vcfuged} instruction. 19542@findex vec_cfuge 19543 19544@smallexample 19545@exdent vector unsigned long long int 19546@exdent vec_cntlzm (vector unsigned long long int, vector unsigned long long int) 19547@end smallexample 19548Perform a vector count leading zeros under bit mask operation, as if 19549implemented by the @code{vclzdm} instruction. 19550@findex vec_cntlzm 19551 19552@smallexample 19553@exdent vector unsigned long long int 19554@exdent vec_cnttzm (vector unsigned long long int, vector unsigned long long int) 19555@end smallexample 19556Perform a vector count trailing zeros under bit mask operation, as if 19557implemented by the @code{vctzdm} instruction. 19558@findex vec_cnttzm 19559 19560@smallexample 19561@exdent vector signed char 19562@exdent vec_clrl (vector signed char a, unsigned int n) 19563@exdent vector unsigned char 19564@exdent vec_clrl (vector unsigned char a, unsigned int n) 19565@end smallexample 19566Clear the left-most @code{(16 - n)} bytes of vector argument @code{a}, as if 19567implemented by the @code{vclrlb} instruction on a big-endian target 19568and by the @code{vclrrb} instruction on a little-endian target. A 19569value of @code{n} that is greater than 16 is treated as if it equaled 16. 19570@findex vec_clrl 19571 19572@smallexample 19573@exdent vector signed char 19574@exdent vec_clrr (vector signed char a, unsigned int n) 19575@exdent vector unsigned char 19576@exdent vec_clrr (vector unsigned char a, unsigned int n) 19577@end smallexample 19578Clear the right-most @code{(16 - n)} bytes of vector argument @code{a}, as if 19579implemented by the @code{vclrrb} instruction on a big-endian target 19580and by the @code{vclrlb} instruction on a little-endian target. A 19581value of @code{n} that is greater than 16 is treated as if it equaled 16. 19582@findex vec_clrr 19583 19584@smallexample 19585@exdent vector unsigned long long int 19586@exdent vec_gnb (vector unsigned __int128, const unsigned char) 19587@end smallexample 19588Perform a 128-bit vector gather operation, as if implemented by the 19589@code{vgnb} instruction. The second argument must be a literal 19590integer value between 2 and 7 inclusive. 19591@findex vec_gnb 19592 19593 19594Vector Extract 19595 19596@smallexample 19597@exdent vector unsigned long long int 19598@exdent vec_extractl (vector unsigned char, vector unsigned char, unsigned int) 19599@exdent vector unsigned long long int 19600@exdent vec_extractl (vector unsigned short, vector unsigned short, unsigned int) 19601@exdent vector unsigned long long int 19602@exdent vec_extractl (vector unsigned int, vector unsigned int, unsigned int) 19603@exdent vector unsigned long long int 19604@exdent vec_extractl (vector unsigned long long, vector unsigned long long, unsigned int) 19605@end smallexample 19606Extract an element from two concatenated vectors starting at the given byte index 19607in natural-endian order, and place it zero-extended in doubleword 1 of the result 19608according to natural element order. If the byte index is out of range for the 19609data type, the intrinsic will be rejected. 19610For little-endian, this output will match the placement by the hardware 19611instruction, i.e., dword[0] in RTL notation. For big-endian, an additional 19612instruction is needed to move it from the "left" doubleword to the "right" one. 19613For little-endian, semantics matching the @code{vextdubvrx}, 19614@code{vextduhvrx}, @code{vextduwvrx} instruction will be generated, while for 19615big-endian, semantics matching the @code{vextdubvlx}, @code{vextduhvlx}, 19616@code{vextduwvlx} instructions 19617will be generated. Note that some fairly anomalous results can be generated if 19618the byte index is not aligned on an element boundary for the element being 19619extracted. This is a limitation of the bi-endian vector programming model is 19620consistent with the limitation on @code{vec_perm}. 19621@findex vec_extractl 19622 19623@smallexample 19624@exdent vector unsigned long long int 19625@exdent vec_extracth (vector unsigned char, vector unsigned char, unsigned int) 19626@exdent vector unsigned long long int 19627@exdent vec_extracth (vector unsigned short, vector unsigned short, 19628unsigned int) 19629@exdent vector unsigned long long int 19630@exdent vec_extracth (vector unsigned int, vector unsigned int, unsigned int) 19631@exdent vector unsigned long long int 19632@exdent vec_extracth (vector unsigned long long, vector unsigned long long, 19633unsigned int) 19634@end smallexample 19635Extract an element from two concatenated vectors starting at the given byte 19636index. The index is based on big endian order for a little endian system. 19637Similarly, the index is based on little endian order for a big endian system. 19638The extraced elements are zero-extended and put in doubleword 1 19639according to natural element order. If the byte index is out of range for the 19640data type, the intrinsic will be rejected. For little-endian, this output 19641will match the placement by the hardware instruction (vextdubvrx, vextduhvrx, 19642vextduwvrx, vextddvrx) i.e., dword[0] in RTL 19643notation. For big-endian, an additional instruction is needed to move it 19644from the "left" doubleword to the "right" one. For little-endian, semantics 19645matching the @code{vextdubvlx}, @code{vextduhvlx}, @code{vextduwvlx} 19646instructions will be generated, while for big-endian, semantics matching the 19647@code{vextdubvrx}, @code{vextduhvrx}, @code{vextduwvrx} instructions will 19648be generated. Note that some fairly anomalous 19649results can be generated if the byte index is not aligned on the 19650element boundary for the element being extracted. This is a 19651limitation of the bi-endian vector programming model consistent with the 19652limitation on @code{vec_perm}. 19653@findex vec_extracth 19654@smallexample 19655@exdent vector unsigned long long int 19656@exdent vec_pdep (vector unsigned long long int, vector unsigned long long int) 19657@end smallexample 19658Perform a vector parallel bits deposit operation, as if implemented by 19659the @code{vpdepd} instruction. 19660@findex vec_pdep 19661 19662Vector Insert 19663 19664@smallexample 19665@exdent vector unsigned char 19666@exdent vec_insertl (unsigned char, vector unsigned char, unsigned int); 19667@exdent vector unsigned short 19668@exdent vec_insertl (unsigned short, vector unsigned short, unsigned int); 19669@exdent vector unsigned int 19670@exdent vec_insertl (unsigned int, vector unsigned int, unsigned int); 19671@exdent vector unsigned long long 19672@exdent vec_insertl (unsigned long long, vector unsigned long long, 19673unsigned int); 19674@exdent vector unsigned char 19675@exdent vec_insertl (vector unsigned char, vector unsigned char, unsigned int; 19676@exdent vector unsigned short 19677@exdent vec_insertl (vector unsigned short, vector unsigned short, 19678unsigned int); 19679@exdent vector unsigned int 19680@exdent vec_insertl (vector unsigned int, vector unsigned int, unsigned int); 19681@end smallexample 19682 19683Let src be the first argument, when the first argument is a scalar, or the 19684rightmost element of the left doubleword of the first argument, when the first 19685argument is a vector. Insert the source into the destination at the position 19686given by the third argument, using natural element order in the second 19687argument. The rest of the second argument is unchanged. If the byte 19688index is greater than 14 for halfwords, greater than 12 for words, or 19689greater than 8 for doublewords the result is undefined. For little-endian, 19690the generated code will be semantically equivalent to @code{vins[bhwd]rx} 19691instructions. Similarly for big-endian it will be semantically equivalent 19692to @code{vins[bhwd]lx}. Note that some fairly anomalous results can be 19693generated if the byte index is not aligned on an element boundary for the 19694type of element being inserted. 19695@findex vec_insertl 19696 19697@smallexample 19698@exdent vector unsigned char 19699@exdent vec_inserth (unsigned char, vector unsigned char, unsigned int); 19700@exdent vector unsigned short 19701@exdent vec_inserth (unsigned short, vector unsigned short, unsigned int); 19702@exdent vector unsigned int 19703@exdent vec_inserth (unsigned int, vector unsigned int, unsigned int); 19704@exdent vector unsigned long long 19705@exdent vec_inserth (unsigned long long, vector unsigned long long, 19706unsigned int); 19707@exdent vector unsigned char 19708@exdent vec_inserth (vector unsigned char, vector unsigned char, unsigned int); 19709@exdent vector unsigned short 19710@exdent vec_inserth (vector unsigned short, vector unsigned short, 19711unsigned int); 19712@exdent vector unsigned int 19713@exdent vec_inserth (vector unsigned int, vector unsigned int, unsigned int); 19714@end smallexample 19715 19716Let src be the first argument, when the first argument is a scalar, or the 19717rightmost element of the first argument, when the first argument is a vector. 19718Insert src into the second argument at the position identified by the third 19719argument, using opposite element order in the second argument, and leaving the 19720rest of the second argument unchanged. If the byte index is greater than 14 19721for halfwords, 12 for words, or 8 for doublewords, the intrinsic will be 19722rejected. Note that the underlying hardware instruction uses the same register 19723for the second argument and the result. 19724For little-endian, the code generation will be semantically equivalent to 19725@code{vins[bhwd]lx}, while for big-endian it will be semantically equivalent to 19726@code{vins[bhwd]rx}. 19727Note that some fairly anomalous results can be generated if the byte index is 19728not aligned on an element boundary for the sort of element being inserted. 19729@findex vec_inserth 19730 19731Vector Replace Element 19732@smallexample 19733@exdent vector signed int vec_replace_elt (vector signed int, signed int, 19734const int); 19735@exdent vector unsigned int vec_replace_elt (vector unsigned int, 19736unsigned int, const int); 19737@exdent vector float vec_replace_elt (vector float, float, const int); 19738@exdent vector signed long long vec_replace_elt (vector signed long long, 19739signed long long, const int); 19740@exdent vector unsigned long long vec_replace_elt (vector unsigned long long, 19741unsigned long long, const int); 19742@exdent vector double rec_replace_elt (vector double, double, const int); 19743@end smallexample 19744The third argument (constrained to [0,3]) identifies the natural-endian 19745element number of the first argument that will be replaced by the second 19746argument to produce the result. The other elements of the first argument will 19747remain unchanged in the result. 19748 19749If it's desirable to insert a word at an unaligned position, use 19750vec_replace_unaligned instead. 19751 19752@findex vec_replace_element 19753 19754Vector Replace Unaligned 19755@smallexample 19756@exdent vector unsigned char vec_replace_unaligned (vector unsigned char, 19757signed int, const int); 19758@exdent vector unsigned char vec_replace_unaligned (vector unsigned char, 19759unsigned int, const int); 19760@exdent vector unsigned char vec_replace_unaligned (vector unsigned char, 19761float, const int); 19762@exdent vector unsigned char vec_replace_unaligned (vector unsigned char, 19763signed long long, const int); 19764@exdent vector unsigned char vec_replace_unaligned (vector unsigned char, 19765unsigned long long, const int); 19766@exdent vector unsigned char vec_replace_unaligned (vector unsigned char, 19767double, const int); 19768@end smallexample 19769 19770The second argument replaces a portion of the first argument to produce the 19771result, with the rest of the first argument unchanged in the result. The 19772third argument identifies the byte index (using left-to-right, or big-endian 19773order) where the high-order byte of the second argument will be placed, with 19774the remaining bytes of the second argument placed naturally "to the right" 19775of the high-order byte. 19776 19777The programmer is responsible for understanding the endianness issues involved 19778with the first argument and the result. 19779@findex vec_replace_unaligned 19780 19781Vector Shift Left Double Bit Immediate 19782@smallexample 19783@exdent vector signed char vec_sldb (vector signed char, vector signed char, 19784const unsigned int); 19785@exdent vector unsigned char vec_sldb (vector unsigned char, 19786vector unsigned char, const unsigned int); 19787@exdent vector signed short vec_sldb (vector signed short, vector signed short, 19788const unsigned int); 19789@exdent vector unsigned short vec_sldb (vector unsigned short, 19790vector unsigned short, const unsigned int); 19791@exdent vector signed int vec_sldb (vector signed int, vector signed int, 19792const unsigned int); 19793@exdent vector unsigned int vec_sldb (vector unsigned int, vector unsigned int, 19794const unsigned int); 19795@exdent vector signed long long vec_sldb (vector signed long long, 19796vector signed long long, const unsigned int); 19797@exdent vector unsigned long long vec_sldb (vector unsigned long long, 19798vector unsigned long long, const unsigned int); 19799@end smallexample 19800 19801Shift the combined input vectors left by the amount specified by the low-order 19802three bits of the third argument, and return the leftmost remaining 128 bits. 19803Code using this instruction must be endian-aware. 19804 19805@findex vec_sldb 19806 19807Vector Shift Right Double Bit Immediate 19808 19809@smallexample 19810@exdent vector signed char vec_srdb (vector signed char, vector signed char, 19811const unsigned int); 19812@exdent vector unsigned char vec_srdb (vector unsigned char, vector unsigned char, 19813const unsigned int); 19814@exdent vector signed short vec_srdb (vector signed short, vector signed short, 19815const unsigned int); 19816@exdent vector unsigned short vec_srdb (vector unsigned short, vector unsigned short, 19817const unsigned int); 19818@exdent vector signed int vec_srdb (vector signed int, vector signed int, 19819const unsigned int); 19820@exdent vector unsigned int vec_srdb (vector unsigned int, vector unsigned int, 19821const unsigned int); 19822@exdent vector signed long long vec_srdb (vector signed long long, 19823vector signed long long, const unsigned int); 19824@exdent vector unsigned long long vec_srdb (vector unsigned long long, 19825vector unsigned long long, const unsigned int); 19826@end smallexample 19827 19828Shift the combined input vectors right by the amount specified by the low-order 19829three bits of the third argument, and return the remaining 128 bits. Code 19830using this built-in must be endian-aware. 19831 19832@findex vec_srdb 19833 19834Vector Splat 19835 19836@smallexample 19837@exdent vector signed int vec_splati (const signed int); 19838@exdent vector float vec_splati (const float); 19839@end smallexample 19840 19841Splat a 32-bit immediate into a vector of words. 19842 19843@findex vec_splati 19844 19845@smallexample 19846@exdent vector double vec_splatid (const float); 19847@end smallexample 19848 19849Convert a single precision floating-point value to double-precision and splat 19850the result to a vector of double-precision floats. 19851 19852@findex vec_splatid 19853 19854@smallexample 19855@exdent vector signed int vec_splati_ins (vector signed int, 19856const unsigned int, const signed int); 19857@exdent vector unsigned int vec_splati_ins (vector unsigned int, 19858const unsigned int, const unsigned int); 19859@exdent vector float vec_splati_ins (vector float, const unsigned int, 19860const float); 19861@end smallexample 19862 19863Argument 2 must be either 0 or 1. Splat the value of argument 3 into the word 19864identified by argument 2 of each doubleword of argument 1 and return the 19865result. The other words of argument 1 are unchanged. 19866 19867@findex vec_splati_ins 19868 19869Vector Blend Variable 19870 19871@smallexample 19872@exdent vector signed char vec_blendv (vector signed char, vector signed char, 19873vector unsigned char); 19874@exdent vector unsigned char vec_blendv (vector unsigned char, 19875vector unsigned char, vector unsigned char); 19876@exdent vector signed short vec_blendv (vector signed short, 19877vector signed short, vector unsigned short); 19878@exdent vector unsigned short vec_blendv (vector unsigned short, 19879vector unsigned short, vector unsigned short); 19880@exdent vector signed int vec_blendv (vector signed int, vector signed int, 19881vector unsigned int); 19882@exdent vector unsigned int vec_blendv (vector unsigned int, 19883vector unsigned int, vector unsigned int); 19884@exdent vector signed long long vec_blendv (vector signed long long, 19885vector signed long long, vector unsigned long long); 19886@exdent vector unsigned long long vec_blendv (vector unsigned long long, 19887vector unsigned long long, vector unsigned long long); 19888@exdent vector float vec_blendv (vector float, vector float, 19889vector unsigned int); 19890@exdent vector double vec_blendv (vector double, vector double, 19891vector unsigned long long); 19892@end smallexample 19893 19894Blend the first and second argument vectors according to the sign bits of the 19895corresponding elements of the third argument vector. This is similar to the 19896@code{vsel} and @code{xxsel} instructions but for bigger elements. 19897 19898@findex vec_blendv 19899 19900Vector Permute Extended 19901 19902@smallexample 19903@exdent vector signed char vec_permx (vector signed char, vector signed char, 19904vector unsigned char, const int); 19905@exdent vector unsigned char vec_permx (vector unsigned char, 19906vector unsigned char, vector unsigned char, const int); 19907@exdent vector signed short vec_permx (vector signed short, 19908vector signed short, vector unsigned char, const int); 19909@exdent vector unsigned short vec_permx (vector unsigned short, 19910vector unsigned short, vector unsigned char, const int); 19911@exdent vector signed int vec_permx (vector signed int, vector signed int, 19912vector unsigned char, const int); 19913@exdent vector unsigned int vec_permx (vector unsigned int, 19914vector unsigned int, vector unsigned char, const int); 19915@exdent vector signed long long vec_permx (vector signed long long, 19916vector signed long long, vector unsigned char, const int); 19917@exdent vector unsigned long long vec_permx (vector unsigned long long, 19918vector unsigned long long, vector unsigned char, const int); 19919@exdent vector float (vector float, vector float, vector unsigned char, 19920const int); 19921@exdent vector double (vector double, vector double, vector unsigned char, 19922const int); 19923@end smallexample 19924 19925Perform a partial permute of the first two arguments, which form a 32-byte 19926section of an emulated vector up to 256 bytes wide, using the partial permute 19927control vector in the third argument. The fourth argument (constrained to 19928values of 0-7) identifies which 32-byte section of the emulated vector is 19929contained in the first two arguments. 19930@findex vec_permx 19931 19932@smallexample 19933@exdent vector unsigned long long int 19934@exdent vec_pext (vector unsigned long long int, vector unsigned long long int) 19935@end smallexample 19936Perform a vector parallel bit extract operation, as if implemented by 19937the @code{vpextd} instruction. 19938@findex vec_pext 19939 19940@smallexample 19941@exdent vector unsigned char vec_stril (vector unsigned char) 19942@exdent vector signed char vec_stril (vector signed char) 19943@exdent vector unsigned short vec_stril (vector unsigned short) 19944@exdent vector signed short vec_stril (vector signed short) 19945@end smallexample 19946Isolate the left-most non-zero elements of the incoming vector argument, 19947replacing all elements to the right of the left-most zero element 19948found within the argument with zero. The typical implementation uses 19949the @code{vstribl} or @code{vstrihl} instruction on big-endian targets 19950and uses the @code{vstribr} or @code{vstrihr} instruction on 19951little-endian targets. 19952@findex vec_stril 19953 19954@smallexample 19955@exdent int vec_stril_p (vector unsigned char) 19956@exdent int vec_stril_p (vector signed char) 19957@exdent int short vec_stril_p (vector unsigned short) 19958@exdent int vec_stril_p (vector signed short) 19959@end smallexample 19960Return a non-zero value if and only if the argument contains a zero 19961element. The typical implementation uses 19962the @code{vstribl.} or @code{vstrihl.} instruction on big-endian targets 19963and uses the @code{vstribr.} or @code{vstrihr.} instruction on 19964little-endian targets. Choose this built-in to check for presence of 19965zero element if the same argument is also passed to @code{vec_stril}. 19966@findex vec_stril_p 19967 19968@smallexample 19969@exdent vector unsigned char vec_strir (vector unsigned char) 19970@exdent vector signed char vec_strir (vector signed char) 19971@exdent vector unsigned short vec_strir (vector unsigned short) 19972@exdent vector signed short vec_strir (vector signed short) 19973@end smallexample 19974Isolate the right-most non-zero elements of the incoming vector argument, 19975replacing all elements to the left of the right-most zero element 19976found within the argument with zero. The typical implementation uses 19977the @code{vstribr} or @code{vstrihr} instruction on big-endian targets 19978and uses the @code{vstribl} or @code{vstrihl} instruction on 19979little-endian targets. 19980@findex vec_strir 19981 19982@smallexample 19983@exdent int vec_strir_p (vector unsigned char) 19984@exdent int vec_strir_p (vector signed char) 19985@exdent int short vec_strir_p (vector unsigned short) 19986@exdent int vec_strir_p (vector signed short) 19987@end smallexample 19988Return a non-zero value if and only if the argument contains a zero 19989element. The typical implementation uses 19990the @code{vstribr.} or @code{vstrihr.} instruction on big-endian targets 19991and uses the @code{vstribl.} or @code{vstrihl.} instruction on 19992little-endian targets. Choose this built-in to check for presence of 19993zero element if the same argument is also passed to @code{vec_strir}. 19994@findex vec_strir_p 19995 19996@smallexample 19997@exdent vector unsigned char 19998@exdent vec_ternarylogic (vector unsigned char, vector unsigned char, 19999 vector unsigned char, const unsigned int) 20000@exdent vector unsigned short 20001@exdent vec_ternarylogic (vector unsigned short, vector unsigned short, 20002 vector unsigned short, const unsigned int) 20003@exdent vector unsigned int 20004@exdent vec_ternarylogic (vector unsigned int, vector unsigned int, 20005 vector unsigned int, const unsigned int) 20006@exdent vector unsigned long long int 20007@exdent vec_ternarylogic (vector unsigned long long int, vector unsigned long long int, 20008 vector unsigned long long int, const unsigned int) 20009@exdent vector unsigned __int128 20010@exdent vec_ternarylogic (vector unsigned __int128, vector unsigned __int128, 20011 vector unsigned __int128, const unsigned int) 20012@end smallexample 20013Perform a 128-bit vector evaluate operation, as if implemented by the 20014@code{xxeval} instruction. The fourth argument must be a literal 20015integer value between 0 and 255 inclusive. 20016@findex vec_ternarylogic 20017 20018@smallexample 20019@exdent vector unsigned char vec_genpcvm (vector unsigned char, const int) 20020@exdent vector unsigned short vec_genpcvm (vector unsigned short, const int) 20021@exdent vector unsigned int vec_genpcvm (vector unsigned int, const int) 20022@exdent vector unsigned int vec_genpcvm (vector unsigned long long int, 20023 const int) 20024@end smallexample 20025 20026Vector Integer Multiply/Divide/Modulo 20027 20028@smallexample 20029@exdent vector signed int 20030@exdent vec_mulh (vector signed int a, vector signed int b) 20031@exdent vector unsigned int 20032@exdent vec_mulh (vector unsigned int a, vector unsigned int b) 20033@end smallexample 20034 20035For each integer value @code{i} from 0 to 3, do the following. The integer 20036value in word element @code{i} of a is multiplied by the integer value in word 20037element @code{i} of b. The high-order 32 bits of the 64-bit product are placed 20038into word element @code{i} of the vector returned. 20039 20040@smallexample 20041@exdent vector signed long long 20042@exdent vec_mulh (vector signed long long a, vector signed long long b) 20043@exdent vector unsigned long long 20044@exdent vec_mulh (vector unsigned long long a, vector unsigned long long b) 20045@end smallexample 20046 20047For each integer value @code{i} from 0 to 1, do the following. The integer 20048value in doubleword element @code{i} of a is multiplied by the integer value in 20049doubleword element @code{i} of b. The high-order 64 bits of the 128-bit product 20050are placed into doubleword element @code{i} of the vector returned. 20051 20052@smallexample 20053@exdent vector unsigned long long 20054@exdent vec_mul (vector unsigned long long a, vector unsigned long long b) 20055@exdent vector signed long long 20056@exdent vec_mul (vector signed long long a, vector signed long long b) 20057@end smallexample 20058 20059For each integer value @code{i} from 0 to 1, do the following. The integer 20060value in doubleword element @code{i} of a is multiplied by the integer value in 20061doubleword element @code{i} of b. The low-order 64 bits of the 128-bit product 20062are placed into doubleword element @code{i} of the vector returned. 20063 20064@smallexample 20065@exdent vector signed int 20066@exdent vec_div (vector signed int a, vector signed int b) 20067@exdent vector unsigned int 20068@exdent vec_div (vector unsigned int a, vector unsigned int b) 20069@end smallexample 20070 20071For each integer value @code{i} from 0 to 3, do the following. The integer in 20072word element @code{i} of a is divided by the integer in word element @code{i} 20073of b. The unique integer quotient is placed into the word element @code{i} of 20074the vector returned. If an attempt is made to perform any of the divisions 20075<anything> ÷ 0 then the quotient is undefined. 20076 20077@smallexample 20078@exdent vector signed long long 20079@exdent vec_div (vector signed long long a, vector signed long long b) 20080@exdent vector unsigned long long 20081@exdent vec_div (vector unsigned long long a, vector unsigned long long b) 20082@end smallexample 20083 20084For each integer value @code{i} from 0 to 1, do the following. The integer in 20085doubleword element @code{i} of a is divided by the integer in doubleword 20086element @code{i} of b. The unique integer quotient is placed into the 20087doubleword element @code{i} of the vector returned. If an attempt is made to 20088perform any of the divisions 0x8000_0000_0000_0000 ÷ -1 or <anything> ÷ 0 then 20089the quotient is undefined. 20090 20091@smallexample 20092@exdent vector signed int 20093@exdent vec_dive (vector signed int a, vector signed int b) 20094@exdent vector unsigned int 20095@exdent vec_dive (vector unsigned int a, vector unsigned int b) 20096@end smallexample 20097 20098For each integer value @code{i} from 0 to 3, do the following. The integer in 20099word element @code{i} of a is shifted left by 32 bits, then divided by the 20100integer in word element @code{i} of b. The unique integer quotient is placed 20101into the word element @code{i} of the vector returned. If the quotient cannot 20102be represented in 32 bits, or if an attempt is made to perform any of the 20103divisions <anything> ÷ 0 then the quotient is undefined. 20104 20105@smallexample 20106@exdent vector signed long long 20107@exdent vec_dive (vector signed long long a, vector signed long long b) 20108@exdent vector unsigned long long 20109@exdent vec_dive (vector unsigned long long a, vector unsigned long long b) 20110@end smallexample 20111 20112For each integer value @code{i} from 0 to 1, do the following. The integer in 20113doubleword element @code{i} of a is shifted left by 64 bits, then divided by 20114the integer in doubleword element @code{i} of b. The unique integer quotient is 20115placed into the doubleword element @code{i} of the vector returned. If the 20116quotient cannot be represented in 64 bits, or if an attempt is made to perform 20117<anything> ÷ 0 then the quotient is undefined. 20118 20119@smallexample 20120@exdent vector signed int 20121@exdent vec_mod (vector signed int a, vector signed int b) 20122@exdent vector unsigned int 20123@exdent vec_mod (vector unsigned int a, vector unsigned int b) 20124@end smallexample 20125 20126For each integer value @code{i} from 0 to 3, do the following. The integer in 20127word element @code{i} of a is divided by the integer in word element @code{i} 20128of b. The unique integer remainder is placed into the word element @code{i} of 20129the vector returned. If an attempt is made to perform any of the divisions 201300x8000_0000 ÷ -1 or <anything> ÷ 0 then the remainder is undefined. 20131 20132@smallexample 20133@exdent vector signed long long 20134@exdent vec_mod (vector signed long long a, vector signed long long b) 20135@exdent vector unsigned long long 20136@exdent vec_mod (vector unsigned long long a, vector unsigned long long b) 20137@end smallexample 20138 20139For each integer value @code{i} from 0 to 1, do the following. The integer in 20140doubleword element @code{i} of a is divided by the integer in doubleword 20141element @code{i} of b. The unique integer remainder is placed into the 20142doubleword element @code{i} of the vector returned. If an attempt is made to 20143perform <anything> ÷ 0 then the remainder is undefined. 20144 20145Generate PCV from specified Mask size, as if implemented by the 20146@code{xxgenpcvbm}, @code{xxgenpcvhm}, @code{xxgenpcvwm} instructions, where 20147immediate value is either 0, 1, 2 or 3. 20148@findex vec_genpcvm 20149 20150@smallexample 20151@exdent vector unsigned __int128 vec_rl (vector unsigned __int128 A, 20152 vector unsigned __int128 B); 20153@exdent vector signed __int128 vec_rl (vector signed __int128 A, 20154 vector unsigned __int128 B); 20155@end smallexample 20156 20157Result value: Each element of R is obtained by rotating the corresponding element 20158of A left by the number of bits specified by the corresponding element of B. 20159 20160 20161@smallexample 20162@exdent vector unsigned __int128 vec_rlmi (vector unsigned __int128, 20163 vector unsigned __int128, 20164 vector unsigned __int128); 20165@exdent vector signed __int128 vec_rlmi (vector signed __int128, 20166 vector signed __int128, 20167 vector unsigned __int128); 20168@end smallexample 20169 20170Returns the result of rotating the first input and inserting it under mask 20171into the second input. The first bit in the mask, the last bit in the mask are 20172obtained from the two 7-bit fields bits [108:115] and bits [117:123] 20173respectively of the second input. The shift is obtained from the third input 20174in the 7-bit field [125:131] where all bits counted from zero at the left. 20175 20176@smallexample 20177@exdent vector unsigned __int128 vec_rlnm (vector unsigned __int128, 20178 vector unsigned __int128, 20179 vector unsigned __int128); 20180@exdent vector signed __int128 vec_rlnm (vector signed __int128, 20181 vector unsigned __int128, 20182 vector unsigned __int128); 20183@end smallexample 20184 20185Returns the result of rotating the first input and ANDing it with a mask. The 20186first bit in the mask and the last bit in the mask are obtained from the two 201877-bit fields bits [117:123] and bits [125:131] respectively of the second 20188input. The shift is obtained from the third input in the 7-bit field bits 20189[125:131] where all bits counted from zero at the left. 20190 20191@smallexample 20192@exdent vector unsigned __int128 vec_sl(vector unsigned __int128 A, vector unsigned __int128 B); 20193@exdent vector signed __int128 vec_sl(vector signed __int128 A, vector unsigned __int128 B); 20194@end smallexample 20195 20196Result value: Each element of R is obtained by shifting the corresponding element of 20197A left by the number of bits specified by the corresponding element of B. 20198 20199@smallexample 20200@exdent vector unsigned __int128 vec_sr(vector unsigned __int128 A, vector unsigned __int128 B); 20201@exdent vector signed __int128 vec_sr(vector signed __int128 A, vector unsigned __int128 B); 20202@end smallexample 20203 20204Result value: Each element of R is obtained by shifting the corresponding element of 20205A right by the number of bits specified by the corresponding element of B. 20206 20207@smallexample 20208@exdent vector unsigned __int128 vec_sra(vector unsigned __int128 A, vector unsigned __int128 B); 20209@exdent vector signed __int128 vec_sra(vector signed __int128 A, vector unsigned __int128 B); 20210@end smallexample 20211 20212Result value: Each element of R is obtained by arithmetic shifting the corresponding 20213element of A right by the number of bits specified by the corresponding element of B. 20214 20215@smallexample 20216@exdent vector unsigned __int128 vec_mule (vector unsigned long long, 20217 vector unsigned long long); 20218@exdent vector signed __int128 vec_mule (vector signed long long, 20219 vector signed long long); 20220@end smallexample 20221 20222Returns a vector containing a 128-bit integer result of multiplying the even 20223doubleword elements of the two inputs. 20224 20225@smallexample 20226@exdent vector unsigned __int128 vec_mulo (vector unsigned long long, 20227 vector unsigned long long); 20228@exdent vector signed __int128 vec_mulo (vector signed long long, 20229 vector signed long long); 20230@end smallexample 20231 20232Returns a vector containing a 128-bit integer result of multiplying the odd 20233doubleword elements of the two inputs. 20234 20235@smallexample 20236@exdent vector unsigned __int128 vec_div (vector unsigned __int128, 20237 vector unsigned __int128); 20238@exdent vector signed __int128 vec_div (vector signed __int128, 20239 vector signed __int128); 20240@end smallexample 20241 20242Returns the result of dividing the first operand by the second operand. An 20243attempt to divide any value by zero or to divide the most negative signed 20244128-bit integer by negative one results in an undefined value. 20245 20246@smallexample 20247@exdent vector unsigned __int128 vec_dive (vector unsigned __int128, 20248 vector unsigned __int128); 20249@exdent vector signed __int128 vec_dive (vector signed __int128, 20250 vector signed __int128); 20251@end smallexample 20252 20253The result is produced by shifting the first input left by 128 bits and 20254dividing by the second. If an attempt is made to divide by zero or the result 20255is larger than 128 bits, the result is undefined. 20256 20257@smallexample 20258@exdent vector unsigned __int128 vec_mod (vector unsigned __int128, 20259 vector unsigned __int128); 20260@exdent vector signed __int128 vec_mod (vector signed __int128, 20261 vector signed __int128); 20262@end smallexample 20263 20264The result is the modulo result of dividing the first input by the second 20265input. 20266 20267The following builtins perform 128-bit vector comparisons. The 20268@code{vec_all_xx}, @code{vec_any_xx}, and @code{vec_cmpxx}, where @code{xx} is 20269one of the operations @code{eq, ne, gt, lt, ge, le} perform pairwise 20270comparisons between the elements at the same positions within their two vector 20271arguments. The @code{vec_all_xx}function returns a non-zero value if and only 20272if all pairwise comparisons are true. The @code{vec_any_xx} function returns 20273a non-zero value if and only if at least one pairwise comparison is true. The 20274@code{vec_cmpxx}function returns a vector of the same type as its two 20275arguments, within which each element consists of all ones to denote that 20276specified logical comparison of the corresponding elements was true. 20277Otherwise, the element of the returned vector contains all zeros. 20278 20279@smallexample 20280vector bool __int128 vec_cmpeq (vector signed __int128, vector signed __int128); 20281vector bool __int128 vec_cmpeq (vector unsigned __int128, vector unsigned __int128); 20282vector bool __int128 vec_cmpne (vector signed __int128, vector signed __int128); 20283vector bool __int128 vec_cmpne (vector unsigned __int128, vector unsigned __int128); 20284vector bool __int128 vec_cmpgt (vector signed __int128, vector signed __int128); 20285vector bool __int128 vec_cmpgt (vector unsigned __int128, vector unsigned __int128); 20286vector bool __int128 vec_cmplt (vector signed __int128, vector signed __int128); 20287vector bool __int128 vec_cmplt (vector unsigned __int128, vector unsigned __int128); 20288vector bool __int128 vec_cmpge (vector signed __int128, vector signed __int128); 20289vector bool __int128 vec_cmpge (vector unsigned __int128, vector unsigned __int128); 20290vector bool __int128 vec_cmple (vector signed __int128, vector signed __int128); 20291vector bool __int128 vec_cmple (vector unsigned __int128, vector unsigned __int128); 20292 20293int vec_all_eq (vector signed __int128, vector signed __int128); 20294int vec_all_eq (vector unsigned __int128, vector unsigned __int128); 20295int vec_all_ne (vector signed __int128, vector signed __int128); 20296int vec_all_ne (vector unsigned __int128, vector unsigned __int128); 20297int vec_all_gt (vector signed __int128, vector signed __int128); 20298int vec_all_gt (vector unsigned __int128, vector unsigned __int128); 20299int vec_all_lt (vector signed __int128, vector signed __int128); 20300int vec_all_lt (vector unsigned __int128, vector unsigned __int128); 20301int vec_all_ge (vector signed __int128, vector signed __int128); 20302int vec_all_ge (vector unsigned __int128, vector unsigned __int128); 20303int vec_all_le (vector signed __int128, vector signed __int128); 20304int vec_all_le (vector unsigned __int128, vector unsigned __int128); 20305 20306int vec_any_eq (vector signed __int128, vector signed __int128); 20307int vec_any_eq (vector unsigned __int128, vector unsigned __int128); 20308int vec_any_ne (vector signed __int128, vector signed __int128); 20309int vec_any_ne (vector unsigned __int128, vector unsigned __int128); 20310int vec_any_gt (vector signed __int128, vector signed __int128); 20311int vec_any_gt (vector unsigned __int128, vector unsigned __int128); 20312int vec_any_lt (vector signed __int128, vector signed __int128); 20313int vec_any_lt (vector unsigned __int128, vector unsigned __int128); 20314int vec_any_ge (vector signed __int128, vector signed __int128); 20315int vec_any_ge (vector unsigned __int128, vector unsigned __int128); 20316int vec_any_le (vector signed __int128, vector signed __int128); 20317int vec_any_le (vector unsigned __int128, vector unsigned __int128); 20318@end smallexample 20319 20320 20321@node PowerPC Hardware Transactional Memory Built-in Functions 20322@subsection PowerPC Hardware Transactional Memory Built-in Functions 20323GCC provides two interfaces for accessing the Hardware Transactional 20324Memory (HTM) instructions available on some of the PowerPC family 20325of processors (eg, POWER8). The two interfaces come in a low level 20326interface, consisting of built-in functions specific to PowerPC and a 20327higher level interface consisting of inline functions that are common 20328between PowerPC and S/390. 20329 20330@subsubsection PowerPC HTM Low Level Built-in Functions 20331 20332The following low level built-in functions are available with 20333@option{-mhtm} or @option{-mcpu=CPU} where CPU is `power8' or later. 20334They all generate the machine instruction that is part of the name. 20335 20336The HTM builtins (with the exception of @code{__builtin_tbegin}) return 20337the full 4-bit condition register value set by their associated hardware 20338instruction. The header file @code{htmintrin.h} defines some macros that can 20339be used to decipher the return value. The @code{__builtin_tbegin} builtin 20340returns a simple @code{true} or @code{false} value depending on whether a transaction was 20341successfully started or not. The arguments of the builtins match exactly the 20342type and order of the associated hardware instruction's operands, except for 20343the @code{__builtin_tcheck} builtin, which does not take any input arguments. 20344Refer to the ISA manual for a description of each instruction's operands. 20345 20346@smallexample 20347unsigned int __builtin_tbegin (unsigned int) 20348unsigned int __builtin_tend (unsigned int) 20349 20350unsigned int __builtin_tabort (unsigned int) 20351unsigned int __builtin_tabortdc (unsigned int, unsigned int, unsigned int) 20352unsigned int __builtin_tabortdci (unsigned int, unsigned int, int) 20353unsigned int __builtin_tabortwc (unsigned int, unsigned int, unsigned int) 20354unsigned int __builtin_tabortwci (unsigned int, unsigned int, int) 20355 20356unsigned int __builtin_tcheck (void) 20357unsigned int __builtin_treclaim (unsigned int) 20358unsigned int __builtin_trechkpt (void) 20359unsigned int __builtin_tsr (unsigned int) 20360@end smallexample 20361 20362In addition to the above HTM built-ins, we have added built-ins for 20363some common extended mnemonics of the HTM instructions: 20364 20365@smallexample 20366unsigned int __builtin_tendall (void) 20367unsigned int __builtin_tresume (void) 20368unsigned int __builtin_tsuspend (void) 20369@end smallexample 20370 20371Note that the semantics of the above HTM builtins are required to mimic 20372the locking semantics used for critical sections. Builtins that are used 20373to create a new transaction or restart a suspended transaction must have 20374lock acquisition like semantics while those builtins that end or suspend a 20375transaction must have lock release like semantics. Specifically, this must 20376mimic lock semantics as specified by C++11, for example: Lock acquisition is 20377as-if an execution of __atomic_exchange_n(&globallock,1,__ATOMIC_ACQUIRE) 20378that returns 0, and lock release is as-if an execution of 20379__atomic_store(&globallock,0,__ATOMIC_RELEASE), with globallock being an 20380implicit implementation-defined lock used for all transactions. The HTM 20381instructions associated with with the builtins inherently provide the 20382correct acquisition and release hardware barriers required. However, 20383the compiler must also be prohibited from moving loads and stores across 20384the builtins in a way that would violate their semantics. This has been 20385accomplished by adding memory barriers to the associated HTM instructions 20386(which is a conservative approach to provide acquire and release semantics). 20387Earlier versions of the compiler did not treat the HTM instructions as 20388memory barriers. A @code{__TM_FENCE__} macro has been added, which can 20389be used to determine whether the current compiler treats HTM instructions 20390as memory barriers or not. This allows the user to explicitly add memory 20391barriers to their code when using an older version of the compiler. 20392 20393The following set of built-in functions are available to gain access 20394to the HTM specific special purpose registers. 20395 20396@smallexample 20397unsigned long __builtin_get_texasr (void) 20398unsigned long __builtin_get_texasru (void) 20399unsigned long __builtin_get_tfhar (void) 20400unsigned long __builtin_get_tfiar (void) 20401 20402void __builtin_set_texasr (unsigned long); 20403void __builtin_set_texasru (unsigned long); 20404void __builtin_set_tfhar (unsigned long); 20405void __builtin_set_tfiar (unsigned long); 20406@end smallexample 20407 20408Example usage of these low level built-in functions may look like: 20409 20410@smallexample 20411#include <htmintrin.h> 20412 20413int num_retries = 10; 20414 20415while (1) 20416 @{ 20417 if (__builtin_tbegin (0)) 20418 @{ 20419 /* Transaction State Initiated. */ 20420 if (is_locked (lock)) 20421 __builtin_tabort (0); 20422 ... transaction code... 20423 __builtin_tend (0); 20424 break; 20425 @} 20426 else 20427 @{ 20428 /* Transaction State Failed. Use locks if the transaction 20429 failure is "persistent" or we've tried too many times. */ 20430 if (num_retries-- <= 0 20431 || _TEXASRU_FAILURE_PERSISTENT (__builtin_get_texasru ())) 20432 @{ 20433 acquire_lock (lock); 20434 ... non transactional fallback path... 20435 release_lock (lock); 20436 break; 20437 @} 20438 @} 20439 @} 20440@end smallexample 20441 20442One final built-in function has been added that returns the value of 20443the 2-bit Transaction State field of the Machine Status Register (MSR) 20444as stored in @code{CR0}. 20445 20446@smallexample 20447unsigned long __builtin_ttest (void) 20448@end smallexample 20449 20450This built-in can be used to determine the current transaction state 20451using the following code example: 20452 20453@smallexample 20454#include <htmintrin.h> 20455 20456unsigned char tx_state = _HTM_STATE (__builtin_ttest ()); 20457 20458if (tx_state == _HTM_TRANSACTIONAL) 20459 @{ 20460 /* Code to use in transactional state. */ 20461 @} 20462else if (tx_state == _HTM_NONTRANSACTIONAL) 20463 @{ 20464 /* Code to use in non-transactional state. */ 20465 @} 20466else if (tx_state == _HTM_SUSPENDED) 20467 @{ 20468 /* Code to use in transaction suspended state. */ 20469 @} 20470@end smallexample 20471 20472@subsubsection PowerPC HTM High Level Inline Functions 20473 20474The following high level HTM interface is made available by including 20475@code{<htmxlintrin.h>} and using @option{-mhtm} or @option{-mcpu=CPU} 20476where CPU is `power8' or later. This interface is common between PowerPC 20477and S/390, allowing users to write one HTM source implementation that 20478can be compiled and executed on either system. 20479 20480@smallexample 20481long __TM_simple_begin (void) 20482long __TM_begin (void* const TM_buff) 20483long __TM_end (void) 20484void __TM_abort (void) 20485void __TM_named_abort (unsigned char const code) 20486void __TM_resume (void) 20487void __TM_suspend (void) 20488 20489long __TM_is_user_abort (void* const TM_buff) 20490long __TM_is_named_user_abort (void* const TM_buff, unsigned char *code) 20491long __TM_is_illegal (void* const TM_buff) 20492long __TM_is_footprint_exceeded (void* const TM_buff) 20493long __TM_nesting_depth (void* const TM_buff) 20494long __TM_is_nested_too_deep(void* const TM_buff) 20495long __TM_is_conflict(void* const TM_buff) 20496long __TM_is_failure_persistent(void* const TM_buff) 20497long __TM_failure_address(void* const TM_buff) 20498long long __TM_failure_code(void* const TM_buff) 20499@end smallexample 20500 20501Using these common set of HTM inline functions, we can create 20502a more portable version of the HTM example in the previous 20503section that will work on either PowerPC or S/390: 20504 20505@smallexample 20506#include <htmxlintrin.h> 20507 20508int num_retries = 10; 20509TM_buff_type TM_buff; 20510 20511while (1) 20512 @{ 20513 if (__TM_begin (TM_buff) == _HTM_TBEGIN_STARTED) 20514 @{ 20515 /* Transaction State Initiated. */ 20516 if (is_locked (lock)) 20517 __TM_abort (); 20518 ... transaction code... 20519 __TM_end (); 20520 break; 20521 @} 20522 else 20523 @{ 20524 /* Transaction State Failed. Use locks if the transaction 20525 failure is "persistent" or we've tried too many times. */ 20526 if (num_retries-- <= 0 20527 || __TM_is_failure_persistent (TM_buff)) 20528 @{ 20529 acquire_lock (lock); 20530 ... non transactional fallback path... 20531 release_lock (lock); 20532 break; 20533 @} 20534 @} 20535 @} 20536@end smallexample 20537 20538@node PowerPC Atomic Memory Operation Functions 20539@subsection PowerPC Atomic Memory Operation Functions 20540ISA 3.0 of the PowerPC added new atomic memory operation (amo) 20541instructions. GCC provides support for these instructions in 64-bit 20542environments. All of the functions are declared in the include file 20543@code{amo.h}. 20544 20545The functions supported are: 20546 20547@smallexample 20548#include <amo.h> 20549 20550uint32_t amo_lwat_add (uint32_t *, uint32_t); 20551uint32_t amo_lwat_xor (uint32_t *, uint32_t); 20552uint32_t amo_lwat_ior (uint32_t *, uint32_t); 20553uint32_t amo_lwat_and (uint32_t *, uint32_t); 20554uint32_t amo_lwat_umax (uint32_t *, uint32_t); 20555uint32_t amo_lwat_umin (uint32_t *, uint32_t); 20556uint32_t amo_lwat_swap (uint32_t *, uint32_t); 20557 20558int32_t amo_lwat_sadd (int32_t *, int32_t); 20559int32_t amo_lwat_smax (int32_t *, int32_t); 20560int32_t amo_lwat_smin (int32_t *, int32_t); 20561int32_t amo_lwat_sswap (int32_t *, int32_t); 20562 20563uint64_t amo_ldat_add (uint64_t *, uint64_t); 20564uint64_t amo_ldat_xor (uint64_t *, uint64_t); 20565uint64_t amo_ldat_ior (uint64_t *, uint64_t); 20566uint64_t amo_ldat_and (uint64_t *, uint64_t); 20567uint64_t amo_ldat_umax (uint64_t *, uint64_t); 20568uint64_t amo_ldat_umin (uint64_t *, uint64_t); 20569uint64_t amo_ldat_swap (uint64_t *, uint64_t); 20570 20571int64_t amo_ldat_sadd (int64_t *, int64_t); 20572int64_t amo_ldat_smax (int64_t *, int64_t); 20573int64_t amo_ldat_smin (int64_t *, int64_t); 20574int64_t amo_ldat_sswap (int64_t *, int64_t); 20575 20576void amo_stwat_add (uint32_t *, uint32_t); 20577void amo_stwat_xor (uint32_t *, uint32_t); 20578void amo_stwat_ior (uint32_t *, uint32_t); 20579void amo_stwat_and (uint32_t *, uint32_t); 20580void amo_stwat_umax (uint32_t *, uint32_t); 20581void amo_stwat_umin (uint32_t *, uint32_t); 20582 20583void amo_stwat_sadd (int32_t *, int32_t); 20584void amo_stwat_smax (int32_t *, int32_t); 20585void amo_stwat_smin (int32_t *, int32_t); 20586 20587void amo_stdat_add (uint64_t *, uint64_t); 20588void amo_stdat_xor (uint64_t *, uint64_t); 20589void amo_stdat_ior (uint64_t *, uint64_t); 20590void amo_stdat_and (uint64_t *, uint64_t); 20591void amo_stdat_umax (uint64_t *, uint64_t); 20592void amo_stdat_umin (uint64_t *, uint64_t); 20593 20594void amo_stdat_sadd (int64_t *, int64_t); 20595void amo_stdat_smax (int64_t *, int64_t); 20596void amo_stdat_smin (int64_t *, int64_t); 20597@end smallexample 20598 20599@node PowerPC Matrix-Multiply Assist Built-in Functions 20600@subsection PowerPC Matrix-Multiply Assist Built-in Functions 20601ISA 3.1 of the PowerPC added new Matrix-Multiply Assist (MMA) instructions. 20602GCC provides support for these instructions through the following built-in 20603functions which are enabled with the @code{-mmma} option. The vec_t type 20604below is defined to be a normal vector unsigned char type. The uint2, uint4 20605and uint8 parameters are 2-bit, 4-bit and 8-bit unsigned integer constants 20606respectively. The compiler will verify that they are constants and that 20607their values are within range. 20608 20609The built-in functions supported are: 20610 20611@smallexample 20612void __builtin_mma_xvi4ger8 (__vector_quad *, vec_t, vec_t); 20613void __builtin_mma_xvi8ger4 (__vector_quad *, vec_t, vec_t); 20614void __builtin_mma_xvi16ger2 (__vector_quad *, vec_t, vec_t); 20615void __builtin_mma_xvi16ger2s (__vector_quad *, vec_t, vec_t); 20616void __builtin_mma_xvf16ger2 (__vector_quad *, vec_t, vec_t); 20617void __builtin_mma_xvbf16ger2 (__vector_quad *, vec_t, vec_t); 20618void __builtin_mma_xvf32ger (__vector_quad *, vec_t, vec_t); 20619 20620void __builtin_mma_xvi4ger8pp (__vector_quad *, vec_t, vec_t); 20621void __builtin_mma_xvi8ger4pp (__vector_quad *, vec_t, vec_t); 20622void __builtin_mma_xvi8ger4spp(__vector_quad *, vec_t, vec_t); 20623void __builtin_mma_xvi16ger2pp (__vector_quad *, vec_t, vec_t); 20624void __builtin_mma_xvi16ger2spp (__vector_quad *, vec_t, vec_t); 20625void __builtin_mma_xvf16ger2pp (__vector_quad *, vec_t, vec_t); 20626void __builtin_mma_xvf16ger2pn (__vector_quad *, vec_t, vec_t); 20627void __builtin_mma_xvf16ger2np (__vector_quad *, vec_t, vec_t); 20628void __builtin_mma_xvf16ger2nn (__vector_quad *, vec_t, vec_t); 20629void __builtin_mma_xvbf16ger2pp (__vector_quad *, vec_t, vec_t); 20630void __builtin_mma_xvbf16ger2pn (__vector_quad *, vec_t, vec_t); 20631void __builtin_mma_xvbf16ger2np (__vector_quad *, vec_t, vec_t); 20632void __builtin_mma_xvbf16ger2nn (__vector_quad *, vec_t, vec_t); 20633void __builtin_mma_xvf32gerpp (__vector_quad *, vec_t, vec_t); 20634void __builtin_mma_xvf32gerpn (__vector_quad *, vec_t, vec_t); 20635void __builtin_mma_xvf32gernp (__vector_quad *, vec_t, vec_t); 20636void __builtin_mma_xvf32gernn (__vector_quad *, vec_t, vec_t); 20637 20638void __builtin_mma_pmxvi4ger8 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint8); 20639void __builtin_mma_pmxvi4ger8pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint8); 20640 20641void __builtin_mma_pmxvi8ger4 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint4); 20642void __builtin_mma_pmxvi8ger4pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint4); 20643void __builtin_mma_pmxvi8ger4spp(__vector_quad *, vec_t, vec_t, uint4, uint4, uint4); 20644 20645void __builtin_mma_pmxvi16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20646void __builtin_mma_pmxvi16ger2s (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20647void __builtin_mma_pmxvf16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20648void __builtin_mma_pmxvbf16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20649 20650void __builtin_mma_pmxvi16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20651void __builtin_mma_pmxvi16ger2spp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20652void __builtin_mma_pmxvf16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20653void __builtin_mma_pmxvf16ger2pn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20654void __builtin_mma_pmxvf16ger2np (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20655void __builtin_mma_pmxvf16ger2nn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20656void __builtin_mma_pmxvbf16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20657void __builtin_mma_pmxvbf16ger2pn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20658void __builtin_mma_pmxvbf16ger2np (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20659void __builtin_mma_pmxvbf16ger2nn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 20660 20661void __builtin_mma_pmxvf32ger (__vector_quad *, vec_t, vec_t, uint4, uint4); 20662void __builtin_mma_pmxvf32gerpp (__vector_quad *, vec_t, vec_t, uint4, uint4); 20663void __builtin_mma_pmxvf32gerpn (__vector_quad *, vec_t, vec_t, uint4, uint4); 20664void __builtin_mma_pmxvf32gernp (__vector_quad *, vec_t, vec_t, uint4, uint4); 20665void __builtin_mma_pmxvf32gernn (__vector_quad *, vec_t, vec_t, uint4, uint4); 20666 20667void __builtin_mma_xvf64ger (__vector_quad *, __vector_pair, vec_t); 20668void __builtin_mma_xvf64gerpp (__vector_quad *, __vector_pair, vec_t); 20669void __builtin_mma_xvf64gerpn (__vector_quad *, __vector_pair, vec_t); 20670void __builtin_mma_xvf64gernp (__vector_quad *, __vector_pair, vec_t); 20671void __builtin_mma_xvf64gernn (__vector_quad *, __vector_pair, vec_t); 20672 20673void __builtin_mma_pmxvf64ger (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 20674void __builtin_mma_pmxvf64gerpp (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 20675void __builtin_mma_pmxvf64gerpn (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 20676void __builtin_mma_pmxvf64gernp (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 20677void __builtin_mma_pmxvf64gernn (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 20678 20679void __builtin_mma_xxmtacc (__vector_quad *); 20680void __builtin_mma_xxmfacc (__vector_quad *); 20681void __builtin_mma_xxsetaccz (__vector_quad *); 20682 20683void __builtin_mma_build_acc (__vector_quad *, vec_t, vec_t, vec_t, vec_t); 20684void __builtin_mma_disassemble_acc (void *, __vector_quad *); 20685 20686void __builtin_vsx_build_pair (__vector_pair *, vec_t, vec_t); 20687void __builtin_vsx_disassemble_pair (void *, __vector_pair *); 20688 20689vec_t __builtin_vsx_xvcvspbf16 (vec_t); 20690vec_t __builtin_vsx_xvcvbf16spn (vec_t); 20691 20692__vector_pair __builtin_vsx_lxvp (size_t, __vector_pair *); 20693void __builtin_vsx_stxvp (__vector_pair, size_t, __vector_pair *); 20694@end smallexample 20695 20696@node PRU Built-in Functions 20697@subsection PRU Built-in Functions 20698 20699GCC provides a couple of special builtin functions to aid in utilizing 20700special PRU instructions. 20701 20702The built-in functions supported are: 20703 20704@table @code 20705@item __delay_cycles (long long @var{cycles}) 20706This inserts an instruction sequence that takes exactly @var{cycles} 20707cycles (between 0 and 0xffffffff) to complete. The inserted sequence 20708may use jumps, loops, or no-ops, and does not interfere with any other 20709instructions. Note that @var{cycles} must be a compile-time constant 20710integer - that is, you must pass a number, not a variable that may be 20711optimized to a constant later. The number of cycles delayed by this 20712builtin is exact. 20713 20714@item __halt (void) 20715This inserts a HALT instruction to stop processor execution. 20716 20717@item unsigned int __lmbd (unsigned int @var{wordval}, unsigned int @var{bitval}) 20718This inserts LMBD instruction to calculate the left-most bit with value 20719@var{bitval} in value @var{wordval}. Only the least significant bit 20720of @var{bitval} is taken into account. 20721@end table 20722 20723@node RISC-V Built-in Functions 20724@subsection RISC-V Built-in Functions 20725 20726These built-in functions are available for the RISC-V family of 20727processors. 20728 20729@deftypefn {Built-in Function} {void *} __builtin_thread_pointer (void) 20730Returns the value that is currently set in the @samp{tp} register. 20731@end deftypefn 20732 20733@node RX Built-in Functions 20734@subsection RX Built-in Functions 20735GCC supports some of the RX instructions which cannot be expressed in 20736the C programming language via the use of built-in functions. The 20737following functions are supported: 20738 20739@deftypefn {Built-in Function} void __builtin_rx_brk (void) 20740Generates the @code{brk} machine instruction. 20741@end deftypefn 20742 20743@deftypefn {Built-in Function} void __builtin_rx_clrpsw (int) 20744Generates the @code{clrpsw} machine instruction to clear the specified 20745bit in the processor status word. 20746@end deftypefn 20747 20748@deftypefn {Built-in Function} void __builtin_rx_int (int) 20749Generates the @code{int} machine instruction to generate an interrupt 20750with the specified value. 20751@end deftypefn 20752 20753@deftypefn {Built-in Function} void __builtin_rx_machi (int, int) 20754Generates the @code{machi} machine instruction to add the result of 20755multiplying the top 16 bits of the two arguments into the 20756accumulator. 20757@end deftypefn 20758 20759@deftypefn {Built-in Function} void __builtin_rx_maclo (int, int) 20760Generates the @code{maclo} machine instruction to add the result of 20761multiplying the bottom 16 bits of the two arguments into the 20762accumulator. 20763@end deftypefn 20764 20765@deftypefn {Built-in Function} void __builtin_rx_mulhi (int, int) 20766Generates the @code{mulhi} machine instruction to place the result of 20767multiplying the top 16 bits of the two arguments into the 20768accumulator. 20769@end deftypefn 20770 20771@deftypefn {Built-in Function} void __builtin_rx_mullo (int, int) 20772Generates the @code{mullo} machine instruction to place the result of 20773multiplying the bottom 16 bits of the two arguments into the 20774accumulator. 20775@end deftypefn 20776 20777@deftypefn {Built-in Function} int __builtin_rx_mvfachi (void) 20778Generates the @code{mvfachi} machine instruction to read the top 2077932 bits of the accumulator. 20780@end deftypefn 20781 20782@deftypefn {Built-in Function} int __builtin_rx_mvfacmi (void) 20783Generates the @code{mvfacmi} machine instruction to read the middle 2078432 bits of the accumulator. 20785@end deftypefn 20786 20787@deftypefn {Built-in Function} int __builtin_rx_mvfc (int) 20788Generates the @code{mvfc} machine instruction which reads the control 20789register specified in its argument and returns its value. 20790@end deftypefn 20791 20792@deftypefn {Built-in Function} void __builtin_rx_mvtachi (int) 20793Generates the @code{mvtachi} machine instruction to set the top 2079432 bits of the accumulator. 20795@end deftypefn 20796 20797@deftypefn {Built-in Function} void __builtin_rx_mvtaclo (int) 20798Generates the @code{mvtaclo} machine instruction to set the bottom 2079932 bits of the accumulator. 20800@end deftypefn 20801 20802@deftypefn {Built-in Function} void __builtin_rx_mvtc (int reg, int val) 20803Generates the @code{mvtc} machine instruction which sets control 20804register number @code{reg} to @code{val}. 20805@end deftypefn 20806 20807@deftypefn {Built-in Function} void __builtin_rx_mvtipl (int) 20808Generates the @code{mvtipl} machine instruction set the interrupt 20809priority level. 20810@end deftypefn 20811 20812@deftypefn {Built-in Function} void __builtin_rx_racw (int) 20813Generates the @code{racw} machine instruction to round the accumulator 20814according to the specified mode. 20815@end deftypefn 20816 20817@deftypefn {Built-in Function} int __builtin_rx_revw (int) 20818Generates the @code{revw} machine instruction which swaps the bytes in 20819the argument so that bits 0--7 now occupy bits 8--15 and vice versa, 20820and also bits 16--23 occupy bits 24--31 and vice versa. 20821@end deftypefn 20822 20823@deftypefn {Built-in Function} void __builtin_rx_rmpa (void) 20824Generates the @code{rmpa} machine instruction which initiates a 20825repeated multiply and accumulate sequence. 20826@end deftypefn 20827 20828@deftypefn {Built-in Function} void __builtin_rx_round (float) 20829Generates the @code{round} machine instruction which returns the 20830floating-point argument rounded according to the current rounding mode 20831set in the floating-point status word register. 20832@end deftypefn 20833 20834@deftypefn {Built-in Function} int __builtin_rx_sat (int) 20835Generates the @code{sat} machine instruction which returns the 20836saturated value of the argument. 20837@end deftypefn 20838 20839@deftypefn {Built-in Function} void __builtin_rx_setpsw (int) 20840Generates the @code{setpsw} machine instruction to set the specified 20841bit in the processor status word. 20842@end deftypefn 20843 20844@deftypefn {Built-in Function} void __builtin_rx_wait (void) 20845Generates the @code{wait} machine instruction. 20846@end deftypefn 20847 20848@node S/390 System z Built-in Functions 20849@subsection S/390 System z Built-in Functions 20850@deftypefn {Built-in Function} int __builtin_tbegin (void*) 20851Generates the @code{tbegin} machine instruction starting a 20852non-constrained hardware transaction. If the parameter is non-NULL the 20853memory area is used to store the transaction diagnostic buffer and 20854will be passed as first operand to @code{tbegin}. This buffer can be 20855defined using the @code{struct __htm_tdb} C struct defined in 20856@code{htmintrin.h} and must reside on a double-word boundary. The 20857second tbegin operand is set to @code{0xff0c}. This enables 20858save/restore of all GPRs and disables aborts for FPR and AR 20859manipulations inside the transaction body. The condition code set by 20860the tbegin instruction is returned as integer value. The tbegin 20861instruction by definition overwrites the content of all FPRs. The 20862compiler will generate code which saves and restores the FPRs. For 20863soft-float code it is recommended to used the @code{*_nofloat} 20864variant. In order to prevent a TDB from being written it is required 20865to pass a constant zero value as parameter. Passing a zero value 20866through a variable is not sufficient. Although modifications of 20867access registers inside the transaction will not trigger an 20868transaction abort it is not supported to actually modify them. Access 20869registers do not get saved when entering a transaction. They will have 20870undefined state when reaching the abort code. 20871@end deftypefn 20872 20873Macros for the possible return codes of tbegin are defined in the 20874@code{htmintrin.h} header file: 20875 20876@table @code 20877@item _HTM_TBEGIN_STARTED 20878@code{tbegin} has been executed as part of normal processing. The 20879transaction body is supposed to be executed. 20880@item _HTM_TBEGIN_INDETERMINATE 20881The transaction was aborted due to an indeterminate condition which 20882might be persistent. 20883@item _HTM_TBEGIN_TRANSIENT 20884The transaction aborted due to a transient failure. The transaction 20885should be re-executed in that case. 20886@item _HTM_TBEGIN_PERSISTENT 20887The transaction aborted due to a persistent failure. Re-execution 20888under same circumstances will not be productive. 20889@end table 20890 20891@defmac _HTM_FIRST_USER_ABORT_CODE 20892The @code{_HTM_FIRST_USER_ABORT_CODE} defined in @code{htmintrin.h} 20893specifies the first abort code which can be used for 20894@code{__builtin_tabort}. Values below this threshold are reserved for 20895machine use. 20896@end defmac 20897 20898@deftp {Data type} {struct __htm_tdb} 20899The @code{struct __htm_tdb} defined in @code{htmintrin.h} describes 20900the structure of the transaction diagnostic block as specified in the 20901Principles of Operation manual chapter 5-91. 20902@end deftp 20903 20904@deftypefn {Built-in Function} int __builtin_tbegin_nofloat (void*) 20905Same as @code{__builtin_tbegin} but without FPR saves and restores. 20906Using this variant in code making use of FPRs will leave the FPRs in 20907undefined state when entering the transaction abort handler code. 20908@end deftypefn 20909 20910@deftypefn {Built-in Function} int __builtin_tbegin_retry (void*, int) 20911In addition to @code{__builtin_tbegin} a loop for transient failures 20912is generated. If tbegin returns a condition code of 2 the transaction 20913will be retried as often as specified in the second argument. The 20914perform processor assist instruction is used to tell the CPU about the 20915number of fails so far. 20916@end deftypefn 20917 20918@deftypefn {Built-in Function} int __builtin_tbegin_retry_nofloat (void*, int) 20919Same as @code{__builtin_tbegin_retry} but without FPR saves and 20920restores. Using this variant in code making use of FPRs will leave 20921the FPRs in undefined state when entering the transaction abort 20922handler code. 20923@end deftypefn 20924 20925@deftypefn {Built-in Function} void __builtin_tbeginc (void) 20926Generates the @code{tbeginc} machine instruction starting a constrained 20927hardware transaction. The second operand is set to @code{0xff08}. 20928@end deftypefn 20929 20930@deftypefn {Built-in Function} int __builtin_tend (void) 20931Generates the @code{tend} machine instruction finishing a transaction 20932and making the changes visible to other threads. The condition code 20933generated by tend is returned as integer value. 20934@end deftypefn 20935 20936@deftypefn {Built-in Function} void __builtin_tabort (int) 20937Generates the @code{tabort} machine instruction with the specified 20938abort code. Abort codes from 0 through 255 are reserved and will 20939result in an error message. 20940@end deftypefn 20941 20942@deftypefn {Built-in Function} void __builtin_tx_assist (int) 20943Generates the @code{ppa rX,rY,1} machine instruction. Where the 20944integer parameter is loaded into rX and a value of zero is loaded into 20945rY. The integer parameter specifies the number of times the 20946transaction repeatedly aborted. 20947@end deftypefn 20948 20949@deftypefn {Built-in Function} int __builtin_tx_nesting_depth (void) 20950Generates the @code{etnd} machine instruction. The current nesting 20951depth is returned as integer value. For a nesting depth of 0 the code 20952is not executed as part of an transaction. 20953@end deftypefn 20954 20955@deftypefn {Built-in Function} void __builtin_non_tx_store (uint64_t *, uint64_t) 20956 20957Generates the @code{ntstg} machine instruction. The second argument 20958is written to the first arguments location. The store operation will 20959not be rolled-back in case of an transaction abort. 20960@end deftypefn 20961 20962@node SH Built-in Functions 20963@subsection SH Built-in Functions 20964The following built-in functions are supported on the SH1, SH2, SH3 and SH4 20965families of processors: 20966 20967@deftypefn {Built-in Function} {void} __builtin_set_thread_pointer (void *@var{ptr}) 20968Sets the @samp{GBR} register to the specified value @var{ptr}. This is usually 20969used by system code that manages threads and execution contexts. The compiler 20970normally does not generate code that modifies the contents of @samp{GBR} and 20971thus the value is preserved across function calls. Changing the @samp{GBR} 20972value in user code must be done with caution, since the compiler might use 20973@samp{GBR} in order to access thread local variables. 20974 20975@end deftypefn 20976 20977@deftypefn {Built-in Function} {void *} __builtin_thread_pointer (void) 20978Returns the value that is currently set in the @samp{GBR} register. 20979Memory loads and stores that use the thread pointer as a base address are 20980turned into @samp{GBR} based displacement loads and stores, if possible. 20981For example: 20982@smallexample 20983struct my_tcb 20984@{ 20985 int a, b, c, d, e; 20986@}; 20987 20988int get_tcb_value (void) 20989@{ 20990 // Generate @samp{mov.l @@(8,gbr),r0} instruction 20991 return ((my_tcb*)__builtin_thread_pointer ())->c; 20992@} 20993 20994@end smallexample 20995@end deftypefn 20996 20997@deftypefn {Built-in Function} {unsigned int} __builtin_sh_get_fpscr (void) 20998Returns the value that is currently set in the @samp{FPSCR} register. 20999@end deftypefn 21000 21001@deftypefn {Built-in Function} {void} __builtin_sh_set_fpscr (unsigned int @var{val}) 21002Sets the @samp{FPSCR} register to the specified value @var{val}, while 21003preserving the current values of the FR, SZ and PR bits. 21004@end deftypefn 21005 21006@node SPARC VIS Built-in Functions 21007@subsection SPARC VIS Built-in Functions 21008 21009GCC supports SIMD operations on the SPARC using both the generic vector 21010extensions (@pxref{Vector Extensions}) as well as built-in functions for 21011the SPARC Visual Instruction Set (VIS). When you use the @option{-mvis} 21012switch, the VIS extension is exposed as the following built-in functions: 21013 21014@smallexample 21015typedef int v1si __attribute__ ((vector_size (4))); 21016typedef int v2si __attribute__ ((vector_size (8))); 21017typedef short v4hi __attribute__ ((vector_size (8))); 21018typedef short v2hi __attribute__ ((vector_size (4))); 21019typedef unsigned char v8qi __attribute__ ((vector_size (8))); 21020typedef unsigned char v4qi __attribute__ ((vector_size (4))); 21021 21022void __builtin_vis_write_gsr (int64_t); 21023int64_t __builtin_vis_read_gsr (void); 21024 21025void * __builtin_vis_alignaddr (void *, long); 21026void * __builtin_vis_alignaddrl (void *, long); 21027int64_t __builtin_vis_faligndatadi (int64_t, int64_t); 21028v2si __builtin_vis_faligndatav2si (v2si, v2si); 21029v4hi __builtin_vis_faligndatav4hi (v4si, v4si); 21030v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi); 21031 21032v4hi __builtin_vis_fexpand (v4qi); 21033 21034v4hi __builtin_vis_fmul8x16 (v4qi, v4hi); 21035v4hi __builtin_vis_fmul8x16au (v4qi, v2hi); 21036v4hi __builtin_vis_fmul8x16al (v4qi, v2hi); 21037v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi); 21038v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi); 21039v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi); 21040v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi); 21041 21042v4qi __builtin_vis_fpack16 (v4hi); 21043v8qi __builtin_vis_fpack32 (v2si, v8qi); 21044v2hi __builtin_vis_fpackfix (v2si); 21045v8qi __builtin_vis_fpmerge (v4qi, v4qi); 21046 21047int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t); 21048 21049long __builtin_vis_edge8 (void *, void *); 21050long __builtin_vis_edge8l (void *, void *); 21051long __builtin_vis_edge16 (void *, void *); 21052long __builtin_vis_edge16l (void *, void *); 21053long __builtin_vis_edge32 (void *, void *); 21054long __builtin_vis_edge32l (void *, void *); 21055 21056long __builtin_vis_fcmple16 (v4hi, v4hi); 21057long __builtin_vis_fcmple32 (v2si, v2si); 21058long __builtin_vis_fcmpne16 (v4hi, v4hi); 21059long __builtin_vis_fcmpne32 (v2si, v2si); 21060long __builtin_vis_fcmpgt16 (v4hi, v4hi); 21061long __builtin_vis_fcmpgt32 (v2si, v2si); 21062long __builtin_vis_fcmpeq16 (v4hi, v4hi); 21063long __builtin_vis_fcmpeq32 (v2si, v2si); 21064 21065v4hi __builtin_vis_fpadd16 (v4hi, v4hi); 21066v2hi __builtin_vis_fpadd16s (v2hi, v2hi); 21067v2si __builtin_vis_fpadd32 (v2si, v2si); 21068v1si __builtin_vis_fpadd32s (v1si, v1si); 21069v4hi __builtin_vis_fpsub16 (v4hi, v4hi); 21070v2hi __builtin_vis_fpsub16s (v2hi, v2hi); 21071v2si __builtin_vis_fpsub32 (v2si, v2si); 21072v1si __builtin_vis_fpsub32s (v1si, v1si); 21073 21074long __builtin_vis_array8 (long, long); 21075long __builtin_vis_array16 (long, long); 21076long __builtin_vis_array32 (long, long); 21077@end smallexample 21078 21079When you use the @option{-mvis2} switch, the VIS version 2.0 built-in 21080functions also become available: 21081 21082@smallexample 21083long __builtin_vis_bmask (long, long); 21084int64_t __builtin_vis_bshuffledi (int64_t, int64_t); 21085v2si __builtin_vis_bshufflev2si (v2si, v2si); 21086v4hi __builtin_vis_bshufflev2si (v4hi, v4hi); 21087v8qi __builtin_vis_bshufflev2si (v8qi, v8qi); 21088 21089long __builtin_vis_edge8n (void *, void *); 21090long __builtin_vis_edge8ln (void *, void *); 21091long __builtin_vis_edge16n (void *, void *); 21092long __builtin_vis_edge16ln (void *, void *); 21093long __builtin_vis_edge32n (void *, void *); 21094long __builtin_vis_edge32ln (void *, void *); 21095@end smallexample 21096 21097When you use the @option{-mvis3} switch, the VIS version 3.0 built-in 21098functions also become available: 21099 21100@smallexample 21101void __builtin_vis_cmask8 (long); 21102void __builtin_vis_cmask16 (long); 21103void __builtin_vis_cmask32 (long); 21104 21105v4hi __builtin_vis_fchksm16 (v4hi, v4hi); 21106 21107v4hi __builtin_vis_fsll16 (v4hi, v4hi); 21108v4hi __builtin_vis_fslas16 (v4hi, v4hi); 21109v4hi __builtin_vis_fsrl16 (v4hi, v4hi); 21110v4hi __builtin_vis_fsra16 (v4hi, v4hi); 21111v2si __builtin_vis_fsll16 (v2si, v2si); 21112v2si __builtin_vis_fslas16 (v2si, v2si); 21113v2si __builtin_vis_fsrl16 (v2si, v2si); 21114v2si __builtin_vis_fsra16 (v2si, v2si); 21115 21116long __builtin_vis_pdistn (v8qi, v8qi); 21117 21118v4hi __builtin_vis_fmean16 (v4hi, v4hi); 21119 21120int64_t __builtin_vis_fpadd64 (int64_t, int64_t); 21121int64_t __builtin_vis_fpsub64 (int64_t, int64_t); 21122 21123v4hi __builtin_vis_fpadds16 (v4hi, v4hi); 21124v2hi __builtin_vis_fpadds16s (v2hi, v2hi); 21125v4hi __builtin_vis_fpsubs16 (v4hi, v4hi); 21126v2hi __builtin_vis_fpsubs16s (v2hi, v2hi); 21127v2si __builtin_vis_fpadds32 (v2si, v2si); 21128v1si __builtin_vis_fpadds32s (v1si, v1si); 21129v2si __builtin_vis_fpsubs32 (v2si, v2si); 21130v1si __builtin_vis_fpsubs32s (v1si, v1si); 21131 21132long __builtin_vis_fucmple8 (v8qi, v8qi); 21133long __builtin_vis_fucmpne8 (v8qi, v8qi); 21134long __builtin_vis_fucmpgt8 (v8qi, v8qi); 21135long __builtin_vis_fucmpeq8 (v8qi, v8qi); 21136 21137float __builtin_vis_fhadds (float, float); 21138double __builtin_vis_fhaddd (double, double); 21139float __builtin_vis_fhsubs (float, float); 21140double __builtin_vis_fhsubd (double, double); 21141float __builtin_vis_fnhadds (float, float); 21142double __builtin_vis_fnhaddd (double, double); 21143 21144int64_t __builtin_vis_umulxhi (int64_t, int64_t); 21145int64_t __builtin_vis_xmulx (int64_t, int64_t); 21146int64_t __builtin_vis_xmulxhi (int64_t, int64_t); 21147@end smallexample 21148 21149When you use the @option{-mvis4} switch, the VIS version 4.0 built-in 21150functions also become available: 21151 21152@smallexample 21153v8qi __builtin_vis_fpadd8 (v8qi, v8qi); 21154v8qi __builtin_vis_fpadds8 (v8qi, v8qi); 21155v8qi __builtin_vis_fpaddus8 (v8qi, v8qi); 21156v4hi __builtin_vis_fpaddus16 (v4hi, v4hi); 21157 21158v8qi __builtin_vis_fpsub8 (v8qi, v8qi); 21159v8qi __builtin_vis_fpsubs8 (v8qi, v8qi); 21160v8qi __builtin_vis_fpsubus8 (v8qi, v8qi); 21161v4hi __builtin_vis_fpsubus16 (v4hi, v4hi); 21162 21163long __builtin_vis_fpcmple8 (v8qi, v8qi); 21164long __builtin_vis_fpcmpgt8 (v8qi, v8qi); 21165long __builtin_vis_fpcmpule16 (v4hi, v4hi); 21166long __builtin_vis_fpcmpugt16 (v4hi, v4hi); 21167long __builtin_vis_fpcmpule32 (v2si, v2si); 21168long __builtin_vis_fpcmpugt32 (v2si, v2si); 21169 21170v8qi __builtin_vis_fpmax8 (v8qi, v8qi); 21171v4hi __builtin_vis_fpmax16 (v4hi, v4hi); 21172v2si __builtin_vis_fpmax32 (v2si, v2si); 21173 21174v8qi __builtin_vis_fpmaxu8 (v8qi, v8qi); 21175v4hi __builtin_vis_fpmaxu16 (v4hi, v4hi); 21176v2si __builtin_vis_fpmaxu32 (v2si, v2si); 21177 21178v8qi __builtin_vis_fpmin8 (v8qi, v8qi); 21179v4hi __builtin_vis_fpmin16 (v4hi, v4hi); 21180v2si __builtin_vis_fpmin32 (v2si, v2si); 21181 21182v8qi __builtin_vis_fpminu8 (v8qi, v8qi); 21183v4hi __builtin_vis_fpminu16 (v4hi, v4hi); 21184v2si __builtin_vis_fpminu32 (v2si, v2si); 21185@end smallexample 21186 21187When you use the @option{-mvis4b} switch, the VIS version 4.0B 21188built-in functions also become available: 21189 21190@smallexample 21191v8qi __builtin_vis_dictunpack8 (double, int); 21192v4hi __builtin_vis_dictunpack16 (double, int); 21193v2si __builtin_vis_dictunpack32 (double, int); 21194 21195long __builtin_vis_fpcmple8shl (v8qi, v8qi, int); 21196long __builtin_vis_fpcmpgt8shl (v8qi, v8qi, int); 21197long __builtin_vis_fpcmpeq8shl (v8qi, v8qi, int); 21198long __builtin_vis_fpcmpne8shl (v8qi, v8qi, int); 21199 21200long __builtin_vis_fpcmple16shl (v4hi, v4hi, int); 21201long __builtin_vis_fpcmpgt16shl (v4hi, v4hi, int); 21202long __builtin_vis_fpcmpeq16shl (v4hi, v4hi, int); 21203long __builtin_vis_fpcmpne16shl (v4hi, v4hi, int); 21204 21205long __builtin_vis_fpcmple32shl (v2si, v2si, int); 21206long __builtin_vis_fpcmpgt32shl (v2si, v2si, int); 21207long __builtin_vis_fpcmpeq32shl (v2si, v2si, int); 21208long __builtin_vis_fpcmpne32shl (v2si, v2si, int); 21209 21210long __builtin_vis_fpcmpule8shl (v8qi, v8qi, int); 21211long __builtin_vis_fpcmpugt8shl (v8qi, v8qi, int); 21212long __builtin_vis_fpcmpule16shl (v4hi, v4hi, int); 21213long __builtin_vis_fpcmpugt16shl (v4hi, v4hi, int); 21214long __builtin_vis_fpcmpule32shl (v2si, v2si, int); 21215long __builtin_vis_fpcmpugt32shl (v2si, v2si, int); 21216 21217long __builtin_vis_fpcmpde8shl (v8qi, v8qi, int); 21218long __builtin_vis_fpcmpde16shl (v4hi, v4hi, int); 21219long __builtin_vis_fpcmpde32shl (v2si, v2si, int); 21220 21221long __builtin_vis_fpcmpur8shl (v8qi, v8qi, int); 21222long __builtin_vis_fpcmpur16shl (v4hi, v4hi, int); 21223long __builtin_vis_fpcmpur32shl (v2si, v2si, int); 21224@end smallexample 21225 21226@node TI C6X Built-in Functions 21227@subsection TI C6X Built-in Functions 21228 21229GCC provides intrinsics to access certain instructions of the TI C6X 21230processors. These intrinsics, listed below, are available after 21231inclusion of the @code{c6x_intrinsics.h} header file. They map directly 21232to C6X instructions. 21233 21234@smallexample 21235 21236int _sadd (int, int) 21237int _ssub (int, int) 21238int _sadd2 (int, int) 21239int _ssub2 (int, int) 21240long long _mpy2 (int, int) 21241long long _smpy2 (int, int) 21242int _add4 (int, int) 21243int _sub4 (int, int) 21244int _saddu4 (int, int) 21245 21246int _smpy (int, int) 21247int _smpyh (int, int) 21248int _smpyhl (int, int) 21249int _smpylh (int, int) 21250 21251int _sshl (int, int) 21252int _subc (int, int) 21253 21254int _avg2 (int, int) 21255int _avgu4 (int, int) 21256 21257int _clrr (int, int) 21258int _extr (int, int) 21259int _extru (int, int) 21260int _abs (int) 21261int _abs2 (int) 21262 21263@end smallexample 21264 21265@node TILE-Gx Built-in Functions 21266@subsection TILE-Gx Built-in Functions 21267 21268GCC provides intrinsics to access every instruction of the TILE-Gx 21269processor. The intrinsics are of the form: 21270 21271@smallexample 21272 21273unsigned long long __insn_@var{op} (...) 21274 21275@end smallexample 21276 21277Where @var{op} is the name of the instruction. Refer to the ISA manual 21278for the complete list of instructions. 21279 21280GCC also provides intrinsics to directly access the network registers. 21281The intrinsics are: 21282 21283@smallexample 21284 21285unsigned long long __tile_idn0_receive (void) 21286unsigned long long __tile_idn1_receive (void) 21287unsigned long long __tile_udn0_receive (void) 21288unsigned long long __tile_udn1_receive (void) 21289unsigned long long __tile_udn2_receive (void) 21290unsigned long long __tile_udn3_receive (void) 21291void __tile_idn_send (unsigned long long) 21292void __tile_udn_send (unsigned long long) 21293 21294@end smallexample 21295 21296The intrinsic @code{void __tile_network_barrier (void)} is used to 21297guarantee that no network operations before it are reordered with 21298those after it. 21299 21300@node TILEPro Built-in Functions 21301@subsection TILEPro Built-in Functions 21302 21303GCC provides intrinsics to access every instruction of the TILEPro 21304processor. The intrinsics are of the form: 21305 21306@smallexample 21307 21308unsigned __insn_@var{op} (...) 21309 21310@end smallexample 21311 21312@noindent 21313where @var{op} is the name of the instruction. Refer to the ISA manual 21314for the complete list of instructions. 21315 21316GCC also provides intrinsics to directly access the network registers. 21317The intrinsics are: 21318 21319@smallexample 21320 21321unsigned __tile_idn0_receive (void) 21322unsigned __tile_idn1_receive (void) 21323unsigned __tile_sn_receive (void) 21324unsigned __tile_udn0_receive (void) 21325unsigned __tile_udn1_receive (void) 21326unsigned __tile_udn2_receive (void) 21327unsigned __tile_udn3_receive (void) 21328void __tile_idn_send (unsigned) 21329void __tile_sn_send (unsigned) 21330void __tile_udn_send (unsigned) 21331 21332@end smallexample 21333 21334The intrinsic @code{void __tile_network_barrier (void)} is used to 21335guarantee that no network operations before it are reordered with 21336those after it. 21337 21338@node x86 Built-in Functions 21339@subsection x86 Built-in Functions 21340 21341These built-in functions are available for the x86-32 and x86-64 family 21342of computers, depending on the command-line switches used. 21343 21344If you specify command-line switches such as @option{-msse}, 21345the compiler could use the extended instruction sets even if the built-ins 21346are not used explicitly in the program. For this reason, applications 21347that perform run-time CPU detection must compile separate files for each 21348supported architecture, using the appropriate flags. In particular, 21349the file containing the CPU detection code should be compiled without 21350these options. 21351 21352The following machine modes are available for use with MMX built-in functions 21353(@pxref{Vector Extensions}): @code{V2SI} for a vector of two 32-bit integers, 21354@code{V4HI} for a vector of four 16-bit integers, and @code{V8QI} for a 21355vector of eight 8-bit integers. Some of the built-in functions operate on 21356MMX registers as a whole 64-bit entity, these use @code{V1DI} as their mode. 21357 21358If 3DNow!@: extensions are enabled, @code{V2SF} is used as a mode for a vector 21359of two 32-bit floating-point values. 21360 21361If SSE extensions are enabled, @code{V4SF} is used for a vector of four 32-bit 21362floating-point values. Some instructions use a vector of four 32-bit 21363integers, these use @code{V4SI}. Finally, some instructions operate on an 21364entire vector register, interpreting it as a 128-bit integer, these use mode 21365@code{TI}. 21366 21367The x86-32 and x86-64 family of processors use additional built-in 21368functions for efficient use of @code{TF} (@code{__float128}) 128-bit 21369floating point and @code{TC} 128-bit complex floating-point values. 21370 21371The following floating-point built-in functions are always available. All 21372of them implement the function that is part of the name. 21373 21374@smallexample 21375__float128 __builtin_fabsq (__float128) 21376__float128 __builtin_copysignq (__float128, __float128) 21377@end smallexample 21378 21379The following built-in functions are always available. 21380 21381@table @code 21382@item __float128 __builtin_infq (void) 21383Similar to @code{__builtin_inf}, except the return type is @code{__float128}. 21384@findex __builtin_infq 21385 21386@item __float128 __builtin_huge_valq (void) 21387Similar to @code{__builtin_huge_val}, except the return type is @code{__float128}. 21388@findex __builtin_huge_valq 21389 21390@item __float128 __builtin_nanq (void) 21391Similar to @code{__builtin_nan}, except the return type is @code{__float128}. 21392@findex __builtin_nanq 21393 21394@item __float128 __builtin_nansq (void) 21395Similar to @code{__builtin_nans}, except the return type is @code{__float128}. 21396@findex __builtin_nansq 21397@end table 21398 21399The following built-in function is always available. 21400 21401@table @code 21402@item void __builtin_ia32_pause (void) 21403Generates the @code{pause} machine instruction with a compiler memory 21404barrier. 21405@end table 21406 21407The following built-in functions are always available and can be used to 21408check the target platform type. 21409 21410@deftypefn {Built-in Function} void __builtin_cpu_init (void) 21411This function runs the CPU detection code to check the type of CPU and the 21412features supported. This built-in function needs to be invoked along with the built-in functions 21413to check CPU type and features, @code{__builtin_cpu_is} and 21414@code{__builtin_cpu_supports}, only when used in a function that is 21415executed before any constructors are called. The CPU detection code is 21416automatically executed in a very high priority constructor. 21417 21418For example, this function has to be used in @code{ifunc} resolvers that 21419check for CPU type using the built-in functions @code{__builtin_cpu_is} 21420and @code{__builtin_cpu_supports}, or in constructors on targets that 21421don't support constructor priority. 21422@smallexample 21423 21424static void (*resolve_memcpy (void)) (void) 21425@{ 21426 // ifunc resolvers fire before constructors, explicitly call the init 21427 // function. 21428 __builtin_cpu_init (); 21429 if (__builtin_cpu_supports ("ssse3")) 21430 return ssse3_memcpy; // super fast memcpy with ssse3 instructions. 21431 else 21432 return default_memcpy; 21433@} 21434 21435void *memcpy (void *, const void *, size_t) 21436 __attribute__ ((ifunc ("resolve_memcpy"))); 21437@end smallexample 21438 21439@end deftypefn 21440 21441@deftypefn {Built-in Function} int __builtin_cpu_is (const char *@var{cpuname}) 21442This function returns a positive integer if the run-time CPU 21443is of type @var{cpuname} 21444and returns @code{0} otherwise. The following CPU names can be detected: 21445 21446@table @samp 21447@item amd 21448AMD CPU. 21449 21450@item intel 21451Intel CPU. 21452 21453@item atom 21454Intel Atom CPU. 21455 21456@item slm 21457Intel Silvermont CPU. 21458 21459@item core2 21460Intel Core 2 CPU. 21461 21462@item corei7 21463Intel Core i7 CPU. 21464 21465@item nehalem 21466Intel Core i7 Nehalem CPU. 21467 21468@item westmere 21469Intel Core i7 Westmere CPU. 21470 21471@item sandybridge 21472Intel Core i7 Sandy Bridge CPU. 21473 21474@item ivybridge 21475Intel Core i7 Ivy Bridge CPU. 21476 21477@item haswell 21478Intel Core i7 Haswell CPU. 21479 21480@item broadwell 21481Intel Core i7 Broadwell CPU. 21482 21483@item skylake 21484Intel Core i7 Skylake CPU. 21485 21486@item skylake-avx512 21487Intel Core i7 Skylake AVX512 CPU. 21488 21489@item cannonlake 21490Intel Core i7 Cannon Lake CPU. 21491 21492@item icelake-client 21493Intel Core i7 Ice Lake Client CPU. 21494 21495@item icelake-server 21496Intel Core i7 Ice Lake Server CPU. 21497 21498@item cascadelake 21499Intel Core i7 Cascadelake CPU. 21500 21501@item tigerlake 21502Intel Core i7 Tigerlake CPU. 21503 21504@item cooperlake 21505Intel Core i7 Cooperlake CPU. 21506 21507@item sapphirerapids 21508Intel Core i7 sapphirerapids CPU. 21509 21510@item alderlake 21511Intel Core i7 Alderlake CPU. 21512 21513@item rocketlake 21514Intel Core i7 Rocketlake CPU. 21515 21516@item bonnell 21517Intel Atom Bonnell CPU. 21518 21519@item silvermont 21520Intel Atom Silvermont CPU. 21521 21522@item goldmont 21523Intel Atom Goldmont CPU. 21524 21525@item goldmont-plus 21526Intel Atom Goldmont Plus CPU. 21527 21528@item tremont 21529Intel Atom Tremont CPU. 21530 21531@item knl 21532Intel Knights Landing CPU. 21533 21534@item knm 21535Intel Knights Mill CPU. 21536 21537@item amdfam10h 21538AMD Family 10h CPU. 21539 21540@item barcelona 21541AMD Family 10h Barcelona CPU. 21542 21543@item shanghai 21544AMD Family 10h Shanghai CPU. 21545 21546@item istanbul 21547AMD Family 10h Istanbul CPU. 21548 21549@item btver1 21550AMD Family 14h CPU. 21551 21552@item amdfam15h 21553AMD Family 15h CPU. 21554 21555@item bdver1 21556AMD Family 15h Bulldozer version 1. 21557 21558@item bdver2 21559AMD Family 15h Bulldozer version 2. 21560 21561@item bdver3 21562AMD Family 15h Bulldozer version 3. 21563 21564@item bdver4 21565AMD Family 15h Bulldozer version 4. 21566 21567@item btver2 21568AMD Family 16h CPU. 21569 21570@item amdfam17h 21571AMD Family 17h CPU. 21572 21573@item znver1 21574AMD Family 17h Zen version 1. 21575 21576@item znver2 21577AMD Family 17h Zen version 2. 21578 21579@item amdfam19h 21580AMD Family 19h CPU. 21581 21582@item znver3 21583AMD Family 19h Zen version 3. 21584@end table 21585 21586Here is an example: 21587@smallexample 21588if (__builtin_cpu_is ("corei7")) 21589 @{ 21590 do_corei7 (); // Core i7 specific implementation. 21591 @} 21592else 21593 @{ 21594 do_generic (); // Generic implementation. 21595 @} 21596@end smallexample 21597@end deftypefn 21598 21599@deftypefn {Built-in Function} int __builtin_cpu_supports (const char *@var{feature}) 21600This function returns a positive integer if the run-time CPU 21601supports @var{feature} 21602and returns @code{0} otherwise. The following features can be detected: 21603 21604@table @samp 21605@item cmov 21606CMOV instruction. 21607@item mmx 21608MMX instructions. 21609@item popcnt 21610POPCNT instruction. 21611@item sse 21612SSE instructions. 21613@item sse2 21614SSE2 instructions. 21615@item sse3 21616SSE3 instructions. 21617@item ssse3 21618SSSE3 instructions. 21619@item sse4.1 21620SSE4.1 instructions. 21621@item sse4.2 21622SSE4.2 instructions. 21623@item avx 21624AVX instructions. 21625@item avx2 21626AVX2 instructions. 21627@item sse4a 21628SSE4A instructions. 21629@item fma4 21630FMA4 instructions. 21631@item xop 21632XOP instructions. 21633@item fma 21634FMA instructions. 21635@item avx512f 21636AVX512F instructions. 21637@item bmi 21638BMI instructions. 21639@item bmi2 21640BMI2 instructions. 21641@item aes 21642AES instructions. 21643@item pclmul 21644PCLMUL instructions. 21645@item avx512vl 21646AVX512VL instructions. 21647@item avx512bw 21648AVX512BW instructions. 21649@item avx512dq 21650AVX512DQ instructions. 21651@item avx512cd 21652AVX512CD instructions. 21653@item avx512er 21654AVX512ER instructions. 21655@item avx512pf 21656AVX512PF instructions. 21657@item avx512vbmi 21658AVX512VBMI instructions. 21659@item avx512ifma 21660AVX512IFMA instructions. 21661@item avx5124vnniw 21662AVX5124VNNIW instructions. 21663@item avx5124fmaps 21664AVX5124FMAPS instructions. 21665@item avx512vpopcntdq 21666AVX512VPOPCNTDQ instructions. 21667@item avx512vbmi2 21668AVX512VBMI2 instructions. 21669@item gfni 21670GFNI instructions. 21671@item vpclmulqdq 21672VPCLMULQDQ instructions. 21673@item avx512vnni 21674AVX512VNNI instructions. 21675@item avx512bitalg 21676AVX512BITALG instructions. 21677@end table 21678 21679Here is an example: 21680@smallexample 21681if (__builtin_cpu_supports ("popcnt")) 21682 @{ 21683 asm("popcnt %1,%0" : "=r"(count) : "rm"(n) : "cc"); 21684 @} 21685else 21686 @{ 21687 count = generic_countbits (n); //generic implementation. 21688 @} 21689@end smallexample 21690@end deftypefn 21691 21692The following built-in functions are made available by @option{-mmmx}. 21693All of them generate the machine instruction that is part of the name. 21694 21695@smallexample 21696v8qi __builtin_ia32_paddb (v8qi, v8qi) 21697v4hi __builtin_ia32_paddw (v4hi, v4hi) 21698v2si __builtin_ia32_paddd (v2si, v2si) 21699v8qi __builtin_ia32_psubb (v8qi, v8qi) 21700v4hi __builtin_ia32_psubw (v4hi, v4hi) 21701v2si __builtin_ia32_psubd (v2si, v2si) 21702v8qi __builtin_ia32_paddsb (v8qi, v8qi) 21703v4hi __builtin_ia32_paddsw (v4hi, v4hi) 21704v8qi __builtin_ia32_psubsb (v8qi, v8qi) 21705v4hi __builtin_ia32_psubsw (v4hi, v4hi) 21706v8qi __builtin_ia32_paddusb (v8qi, v8qi) 21707v4hi __builtin_ia32_paddusw (v4hi, v4hi) 21708v8qi __builtin_ia32_psubusb (v8qi, v8qi) 21709v4hi __builtin_ia32_psubusw (v4hi, v4hi) 21710v4hi __builtin_ia32_pmullw (v4hi, v4hi) 21711v4hi __builtin_ia32_pmulhw (v4hi, v4hi) 21712di __builtin_ia32_pand (di, di) 21713di __builtin_ia32_pandn (di,di) 21714di __builtin_ia32_por (di, di) 21715di __builtin_ia32_pxor (di, di) 21716v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi) 21717v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi) 21718v2si __builtin_ia32_pcmpeqd (v2si, v2si) 21719v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi) 21720v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi) 21721v2si __builtin_ia32_pcmpgtd (v2si, v2si) 21722v8qi __builtin_ia32_punpckhbw (v8qi, v8qi) 21723v4hi __builtin_ia32_punpckhwd (v4hi, v4hi) 21724v2si __builtin_ia32_punpckhdq (v2si, v2si) 21725v8qi __builtin_ia32_punpcklbw (v8qi, v8qi) 21726v4hi __builtin_ia32_punpcklwd (v4hi, v4hi) 21727v2si __builtin_ia32_punpckldq (v2si, v2si) 21728v8qi __builtin_ia32_packsswb (v4hi, v4hi) 21729v4hi __builtin_ia32_packssdw (v2si, v2si) 21730v8qi __builtin_ia32_packuswb (v4hi, v4hi) 21731 21732v4hi __builtin_ia32_psllw (v4hi, v4hi) 21733v2si __builtin_ia32_pslld (v2si, v2si) 21734v1di __builtin_ia32_psllq (v1di, v1di) 21735v4hi __builtin_ia32_psrlw (v4hi, v4hi) 21736v2si __builtin_ia32_psrld (v2si, v2si) 21737v1di __builtin_ia32_psrlq (v1di, v1di) 21738v4hi __builtin_ia32_psraw (v4hi, v4hi) 21739v2si __builtin_ia32_psrad (v2si, v2si) 21740v4hi __builtin_ia32_psllwi (v4hi, int) 21741v2si __builtin_ia32_pslldi (v2si, int) 21742v1di __builtin_ia32_psllqi (v1di, int) 21743v4hi __builtin_ia32_psrlwi (v4hi, int) 21744v2si __builtin_ia32_psrldi (v2si, int) 21745v1di __builtin_ia32_psrlqi (v1di, int) 21746v4hi __builtin_ia32_psrawi (v4hi, int) 21747v2si __builtin_ia32_psradi (v2si, int) 21748 21749@end smallexample 21750 21751The following built-in functions are made available either with 21752@option{-msse}, or with @option{-m3dnowa}. All of them generate 21753the machine instruction that is part of the name. 21754 21755@smallexample 21756v4hi __builtin_ia32_pmulhuw (v4hi, v4hi) 21757v8qi __builtin_ia32_pavgb (v8qi, v8qi) 21758v4hi __builtin_ia32_pavgw (v4hi, v4hi) 21759v1di __builtin_ia32_psadbw (v8qi, v8qi) 21760v8qi __builtin_ia32_pmaxub (v8qi, v8qi) 21761v4hi __builtin_ia32_pmaxsw (v4hi, v4hi) 21762v8qi __builtin_ia32_pminub (v8qi, v8qi) 21763v4hi __builtin_ia32_pminsw (v4hi, v4hi) 21764int __builtin_ia32_pmovmskb (v8qi) 21765void __builtin_ia32_maskmovq (v8qi, v8qi, char *) 21766void __builtin_ia32_movntq (di *, di) 21767void __builtin_ia32_sfence (void) 21768@end smallexample 21769 21770The following built-in functions are available when @option{-msse} is used. 21771All of them generate the machine instruction that is part of the name. 21772 21773@smallexample 21774int __builtin_ia32_comieq (v4sf, v4sf) 21775int __builtin_ia32_comineq (v4sf, v4sf) 21776int __builtin_ia32_comilt (v4sf, v4sf) 21777int __builtin_ia32_comile (v4sf, v4sf) 21778int __builtin_ia32_comigt (v4sf, v4sf) 21779int __builtin_ia32_comige (v4sf, v4sf) 21780int __builtin_ia32_ucomieq (v4sf, v4sf) 21781int __builtin_ia32_ucomineq (v4sf, v4sf) 21782int __builtin_ia32_ucomilt (v4sf, v4sf) 21783int __builtin_ia32_ucomile (v4sf, v4sf) 21784int __builtin_ia32_ucomigt (v4sf, v4sf) 21785int __builtin_ia32_ucomige (v4sf, v4sf) 21786v4sf __builtin_ia32_addps (v4sf, v4sf) 21787v4sf __builtin_ia32_subps (v4sf, v4sf) 21788v4sf __builtin_ia32_mulps (v4sf, v4sf) 21789v4sf __builtin_ia32_divps (v4sf, v4sf) 21790v4sf __builtin_ia32_addss (v4sf, v4sf) 21791v4sf __builtin_ia32_subss (v4sf, v4sf) 21792v4sf __builtin_ia32_mulss (v4sf, v4sf) 21793v4sf __builtin_ia32_divss (v4sf, v4sf) 21794v4sf __builtin_ia32_cmpeqps (v4sf, v4sf) 21795v4sf __builtin_ia32_cmpltps (v4sf, v4sf) 21796v4sf __builtin_ia32_cmpleps (v4sf, v4sf) 21797v4sf __builtin_ia32_cmpgtps (v4sf, v4sf) 21798v4sf __builtin_ia32_cmpgeps (v4sf, v4sf) 21799v4sf __builtin_ia32_cmpunordps (v4sf, v4sf) 21800v4sf __builtin_ia32_cmpneqps (v4sf, v4sf) 21801v4sf __builtin_ia32_cmpnltps (v4sf, v4sf) 21802v4sf __builtin_ia32_cmpnleps (v4sf, v4sf) 21803v4sf __builtin_ia32_cmpngtps (v4sf, v4sf) 21804v4sf __builtin_ia32_cmpngeps (v4sf, v4sf) 21805v4sf __builtin_ia32_cmpordps (v4sf, v4sf) 21806v4sf __builtin_ia32_cmpeqss (v4sf, v4sf) 21807v4sf __builtin_ia32_cmpltss (v4sf, v4sf) 21808v4sf __builtin_ia32_cmpless (v4sf, v4sf) 21809v4sf __builtin_ia32_cmpunordss (v4sf, v4sf) 21810v4sf __builtin_ia32_cmpneqss (v4sf, v4sf) 21811v4sf __builtin_ia32_cmpnltss (v4sf, v4sf) 21812v4sf __builtin_ia32_cmpnless (v4sf, v4sf) 21813v4sf __builtin_ia32_cmpordss (v4sf, v4sf) 21814v4sf __builtin_ia32_maxps (v4sf, v4sf) 21815v4sf __builtin_ia32_maxss (v4sf, v4sf) 21816v4sf __builtin_ia32_minps (v4sf, v4sf) 21817v4sf __builtin_ia32_minss (v4sf, v4sf) 21818v4sf __builtin_ia32_andps (v4sf, v4sf) 21819v4sf __builtin_ia32_andnps (v4sf, v4sf) 21820v4sf __builtin_ia32_orps (v4sf, v4sf) 21821v4sf __builtin_ia32_xorps (v4sf, v4sf) 21822v4sf __builtin_ia32_movss (v4sf, v4sf) 21823v4sf __builtin_ia32_movhlps (v4sf, v4sf) 21824v4sf __builtin_ia32_movlhps (v4sf, v4sf) 21825v4sf __builtin_ia32_unpckhps (v4sf, v4sf) 21826v4sf __builtin_ia32_unpcklps (v4sf, v4sf) 21827v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si) 21828v4sf __builtin_ia32_cvtsi2ss (v4sf, int) 21829v2si __builtin_ia32_cvtps2pi (v4sf) 21830int __builtin_ia32_cvtss2si (v4sf) 21831v2si __builtin_ia32_cvttps2pi (v4sf) 21832int __builtin_ia32_cvttss2si (v4sf) 21833v4sf __builtin_ia32_rcpps (v4sf) 21834v4sf __builtin_ia32_rsqrtps (v4sf) 21835v4sf __builtin_ia32_sqrtps (v4sf) 21836v4sf __builtin_ia32_rcpss (v4sf) 21837v4sf __builtin_ia32_rsqrtss (v4sf) 21838v4sf __builtin_ia32_sqrtss (v4sf) 21839v4sf __builtin_ia32_shufps (v4sf, v4sf, int) 21840void __builtin_ia32_movntps (float *, v4sf) 21841int __builtin_ia32_movmskps (v4sf) 21842@end smallexample 21843 21844The following built-in functions are available when @option{-msse} is used. 21845 21846@table @code 21847@item v4sf __builtin_ia32_loadups (float *) 21848Generates the @code{movups} machine instruction as a load from memory. 21849@item void __builtin_ia32_storeups (float *, v4sf) 21850Generates the @code{movups} machine instruction as a store to memory. 21851@item v4sf __builtin_ia32_loadss (float *) 21852Generates the @code{movss} machine instruction as a load from memory. 21853@item v4sf __builtin_ia32_loadhps (v4sf, const v2sf *) 21854Generates the @code{movhps} machine instruction as a load from memory. 21855@item v4sf __builtin_ia32_loadlps (v4sf, const v2sf *) 21856Generates the @code{movlps} machine instruction as a load from memory 21857@item void __builtin_ia32_storehps (v2sf *, v4sf) 21858Generates the @code{movhps} machine instruction as a store to memory. 21859@item void __builtin_ia32_storelps (v2sf *, v4sf) 21860Generates the @code{movlps} machine instruction as a store to memory. 21861@end table 21862 21863The following built-in functions are available when @option{-msse2} is used. 21864All of them generate the machine instruction that is part of the name. 21865 21866@smallexample 21867int __builtin_ia32_comisdeq (v2df, v2df) 21868int __builtin_ia32_comisdlt (v2df, v2df) 21869int __builtin_ia32_comisdle (v2df, v2df) 21870int __builtin_ia32_comisdgt (v2df, v2df) 21871int __builtin_ia32_comisdge (v2df, v2df) 21872int __builtin_ia32_comisdneq (v2df, v2df) 21873int __builtin_ia32_ucomisdeq (v2df, v2df) 21874int __builtin_ia32_ucomisdlt (v2df, v2df) 21875int __builtin_ia32_ucomisdle (v2df, v2df) 21876int __builtin_ia32_ucomisdgt (v2df, v2df) 21877int __builtin_ia32_ucomisdge (v2df, v2df) 21878int __builtin_ia32_ucomisdneq (v2df, v2df) 21879v2df __builtin_ia32_cmpeqpd (v2df, v2df) 21880v2df __builtin_ia32_cmpltpd (v2df, v2df) 21881v2df __builtin_ia32_cmplepd (v2df, v2df) 21882v2df __builtin_ia32_cmpgtpd (v2df, v2df) 21883v2df __builtin_ia32_cmpgepd (v2df, v2df) 21884v2df __builtin_ia32_cmpunordpd (v2df, v2df) 21885v2df __builtin_ia32_cmpneqpd (v2df, v2df) 21886v2df __builtin_ia32_cmpnltpd (v2df, v2df) 21887v2df __builtin_ia32_cmpnlepd (v2df, v2df) 21888v2df __builtin_ia32_cmpngtpd (v2df, v2df) 21889v2df __builtin_ia32_cmpngepd (v2df, v2df) 21890v2df __builtin_ia32_cmpordpd (v2df, v2df) 21891v2df __builtin_ia32_cmpeqsd (v2df, v2df) 21892v2df __builtin_ia32_cmpltsd (v2df, v2df) 21893v2df __builtin_ia32_cmplesd (v2df, v2df) 21894v2df __builtin_ia32_cmpunordsd (v2df, v2df) 21895v2df __builtin_ia32_cmpneqsd (v2df, v2df) 21896v2df __builtin_ia32_cmpnltsd (v2df, v2df) 21897v2df __builtin_ia32_cmpnlesd (v2df, v2df) 21898v2df __builtin_ia32_cmpordsd (v2df, v2df) 21899v2di __builtin_ia32_paddq (v2di, v2di) 21900v2di __builtin_ia32_psubq (v2di, v2di) 21901v2df __builtin_ia32_addpd (v2df, v2df) 21902v2df __builtin_ia32_subpd (v2df, v2df) 21903v2df __builtin_ia32_mulpd (v2df, v2df) 21904v2df __builtin_ia32_divpd (v2df, v2df) 21905v2df __builtin_ia32_addsd (v2df, v2df) 21906v2df __builtin_ia32_subsd (v2df, v2df) 21907v2df __builtin_ia32_mulsd (v2df, v2df) 21908v2df __builtin_ia32_divsd (v2df, v2df) 21909v2df __builtin_ia32_minpd (v2df, v2df) 21910v2df __builtin_ia32_maxpd (v2df, v2df) 21911v2df __builtin_ia32_minsd (v2df, v2df) 21912v2df __builtin_ia32_maxsd (v2df, v2df) 21913v2df __builtin_ia32_andpd (v2df, v2df) 21914v2df __builtin_ia32_andnpd (v2df, v2df) 21915v2df __builtin_ia32_orpd (v2df, v2df) 21916v2df __builtin_ia32_xorpd (v2df, v2df) 21917v2df __builtin_ia32_movsd (v2df, v2df) 21918v2df __builtin_ia32_unpckhpd (v2df, v2df) 21919v2df __builtin_ia32_unpcklpd (v2df, v2df) 21920v16qi __builtin_ia32_paddb128 (v16qi, v16qi) 21921v8hi __builtin_ia32_paddw128 (v8hi, v8hi) 21922v4si __builtin_ia32_paddd128 (v4si, v4si) 21923v2di __builtin_ia32_paddq128 (v2di, v2di) 21924v16qi __builtin_ia32_psubb128 (v16qi, v16qi) 21925v8hi __builtin_ia32_psubw128 (v8hi, v8hi) 21926v4si __builtin_ia32_psubd128 (v4si, v4si) 21927v2di __builtin_ia32_psubq128 (v2di, v2di) 21928v8hi __builtin_ia32_pmullw128 (v8hi, v8hi) 21929v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi) 21930v2di __builtin_ia32_pand128 (v2di, v2di) 21931v2di __builtin_ia32_pandn128 (v2di, v2di) 21932v2di __builtin_ia32_por128 (v2di, v2di) 21933v2di __builtin_ia32_pxor128 (v2di, v2di) 21934v16qi __builtin_ia32_pavgb128 (v16qi, v16qi) 21935v8hi __builtin_ia32_pavgw128 (v8hi, v8hi) 21936v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi) 21937v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi) 21938v4si __builtin_ia32_pcmpeqd128 (v4si, v4si) 21939v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi) 21940v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi) 21941v4si __builtin_ia32_pcmpgtd128 (v4si, v4si) 21942v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi) 21943v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi) 21944v16qi __builtin_ia32_pminub128 (v16qi, v16qi) 21945v8hi __builtin_ia32_pminsw128 (v8hi, v8hi) 21946v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi) 21947v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi) 21948v4si __builtin_ia32_punpckhdq128 (v4si, v4si) 21949v2di __builtin_ia32_punpckhqdq128 (v2di, v2di) 21950v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi) 21951v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi) 21952v4si __builtin_ia32_punpckldq128 (v4si, v4si) 21953v2di __builtin_ia32_punpcklqdq128 (v2di, v2di) 21954v16qi __builtin_ia32_packsswb128 (v8hi, v8hi) 21955v8hi __builtin_ia32_packssdw128 (v4si, v4si) 21956v16qi __builtin_ia32_packuswb128 (v8hi, v8hi) 21957v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi) 21958void __builtin_ia32_maskmovdqu (v16qi, v16qi) 21959v2df __builtin_ia32_loadupd (double *) 21960void __builtin_ia32_storeupd (double *, v2df) 21961v2df __builtin_ia32_loadhpd (v2df, double const *) 21962v2df __builtin_ia32_loadlpd (v2df, double const *) 21963int __builtin_ia32_movmskpd (v2df) 21964int __builtin_ia32_pmovmskb128 (v16qi) 21965void __builtin_ia32_movnti (int *, int) 21966void __builtin_ia32_movnti64 (long long int *, long long int) 21967void __builtin_ia32_movntpd (double *, v2df) 21968void __builtin_ia32_movntdq (v2df *, v2df) 21969v4si __builtin_ia32_pshufd (v4si, int) 21970v8hi __builtin_ia32_pshuflw (v8hi, int) 21971v8hi __builtin_ia32_pshufhw (v8hi, int) 21972v2di __builtin_ia32_psadbw128 (v16qi, v16qi) 21973v2df __builtin_ia32_sqrtpd (v2df) 21974v2df __builtin_ia32_sqrtsd (v2df) 21975v2df __builtin_ia32_shufpd (v2df, v2df, int) 21976v2df __builtin_ia32_cvtdq2pd (v4si) 21977v4sf __builtin_ia32_cvtdq2ps (v4si) 21978v4si __builtin_ia32_cvtpd2dq (v2df) 21979v2si __builtin_ia32_cvtpd2pi (v2df) 21980v4sf __builtin_ia32_cvtpd2ps (v2df) 21981v4si __builtin_ia32_cvttpd2dq (v2df) 21982v2si __builtin_ia32_cvttpd2pi (v2df) 21983v2df __builtin_ia32_cvtpi2pd (v2si) 21984int __builtin_ia32_cvtsd2si (v2df) 21985int __builtin_ia32_cvttsd2si (v2df) 21986long long __builtin_ia32_cvtsd2si64 (v2df) 21987long long __builtin_ia32_cvttsd2si64 (v2df) 21988v4si __builtin_ia32_cvtps2dq (v4sf) 21989v2df __builtin_ia32_cvtps2pd (v4sf) 21990v4si __builtin_ia32_cvttps2dq (v4sf) 21991v2df __builtin_ia32_cvtsi2sd (v2df, int) 21992v2df __builtin_ia32_cvtsi642sd (v2df, long long) 21993v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df) 21994v2df __builtin_ia32_cvtss2sd (v2df, v4sf) 21995void __builtin_ia32_clflush (const void *) 21996void __builtin_ia32_lfence (void) 21997void __builtin_ia32_mfence (void) 21998v16qi __builtin_ia32_loaddqu (const char *) 21999void __builtin_ia32_storedqu (char *, v16qi) 22000v1di __builtin_ia32_pmuludq (v2si, v2si) 22001v2di __builtin_ia32_pmuludq128 (v4si, v4si) 22002v8hi __builtin_ia32_psllw128 (v8hi, v8hi) 22003v4si __builtin_ia32_pslld128 (v4si, v4si) 22004v2di __builtin_ia32_psllq128 (v2di, v2di) 22005v8hi __builtin_ia32_psrlw128 (v8hi, v8hi) 22006v4si __builtin_ia32_psrld128 (v4si, v4si) 22007v2di __builtin_ia32_psrlq128 (v2di, v2di) 22008v8hi __builtin_ia32_psraw128 (v8hi, v8hi) 22009v4si __builtin_ia32_psrad128 (v4si, v4si) 22010v2di __builtin_ia32_pslldqi128 (v2di, int) 22011v8hi __builtin_ia32_psllwi128 (v8hi, int) 22012v4si __builtin_ia32_pslldi128 (v4si, int) 22013v2di __builtin_ia32_psllqi128 (v2di, int) 22014v2di __builtin_ia32_psrldqi128 (v2di, int) 22015v8hi __builtin_ia32_psrlwi128 (v8hi, int) 22016v4si __builtin_ia32_psrldi128 (v4si, int) 22017v2di __builtin_ia32_psrlqi128 (v2di, int) 22018v8hi __builtin_ia32_psrawi128 (v8hi, int) 22019v4si __builtin_ia32_psradi128 (v4si, int) 22020v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi) 22021v2di __builtin_ia32_movq128 (v2di) 22022@end smallexample 22023 22024The following built-in functions are available when @option{-msse3} is used. 22025All of them generate the machine instruction that is part of the name. 22026 22027@smallexample 22028v2df __builtin_ia32_addsubpd (v2df, v2df) 22029v4sf __builtin_ia32_addsubps (v4sf, v4sf) 22030v2df __builtin_ia32_haddpd (v2df, v2df) 22031v4sf __builtin_ia32_haddps (v4sf, v4sf) 22032v2df __builtin_ia32_hsubpd (v2df, v2df) 22033v4sf __builtin_ia32_hsubps (v4sf, v4sf) 22034v16qi __builtin_ia32_lddqu (char const *) 22035void __builtin_ia32_monitor (void *, unsigned int, unsigned int) 22036v4sf __builtin_ia32_movshdup (v4sf) 22037v4sf __builtin_ia32_movsldup (v4sf) 22038void __builtin_ia32_mwait (unsigned int, unsigned int) 22039@end smallexample 22040 22041The following built-in functions are available when @option{-mssse3} is used. 22042All of them generate the machine instruction that is part of the name. 22043 22044@smallexample 22045v2si __builtin_ia32_phaddd (v2si, v2si) 22046v4hi __builtin_ia32_phaddw (v4hi, v4hi) 22047v4hi __builtin_ia32_phaddsw (v4hi, v4hi) 22048v2si __builtin_ia32_phsubd (v2si, v2si) 22049v4hi __builtin_ia32_phsubw (v4hi, v4hi) 22050v4hi __builtin_ia32_phsubsw (v4hi, v4hi) 22051v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi) 22052v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi) 22053v8qi __builtin_ia32_pshufb (v8qi, v8qi) 22054v8qi __builtin_ia32_psignb (v8qi, v8qi) 22055v2si __builtin_ia32_psignd (v2si, v2si) 22056v4hi __builtin_ia32_psignw (v4hi, v4hi) 22057v1di __builtin_ia32_palignr (v1di, v1di, int) 22058v8qi __builtin_ia32_pabsb (v8qi) 22059v2si __builtin_ia32_pabsd (v2si) 22060v4hi __builtin_ia32_pabsw (v4hi) 22061@end smallexample 22062 22063The following built-in functions are available when @option{-mssse3} is used. 22064All of them generate the machine instruction that is part of the name. 22065 22066@smallexample 22067v4si __builtin_ia32_phaddd128 (v4si, v4si) 22068v8hi __builtin_ia32_phaddw128 (v8hi, v8hi) 22069v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi) 22070v4si __builtin_ia32_phsubd128 (v4si, v4si) 22071v8hi __builtin_ia32_phsubw128 (v8hi, v8hi) 22072v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi) 22073v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi) 22074v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi) 22075v16qi __builtin_ia32_pshufb128 (v16qi, v16qi) 22076v16qi __builtin_ia32_psignb128 (v16qi, v16qi) 22077v4si __builtin_ia32_psignd128 (v4si, v4si) 22078v8hi __builtin_ia32_psignw128 (v8hi, v8hi) 22079v2di __builtin_ia32_palignr128 (v2di, v2di, int) 22080v16qi __builtin_ia32_pabsb128 (v16qi) 22081v4si __builtin_ia32_pabsd128 (v4si) 22082v8hi __builtin_ia32_pabsw128 (v8hi) 22083@end smallexample 22084 22085The following built-in functions are available when @option{-msse4.1} is 22086used. All of them generate the machine instruction that is part of the 22087name. 22088 22089@smallexample 22090v2df __builtin_ia32_blendpd (v2df, v2df, const int) 22091v4sf __builtin_ia32_blendps (v4sf, v4sf, const int) 22092v2df __builtin_ia32_blendvpd (v2df, v2df, v2df) 22093v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf) 22094v2df __builtin_ia32_dppd (v2df, v2df, const int) 22095v4sf __builtin_ia32_dpps (v4sf, v4sf, const int) 22096v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int) 22097v2di __builtin_ia32_movntdqa (v2di *); 22098v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int) 22099v8hi __builtin_ia32_packusdw128 (v4si, v4si) 22100v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi) 22101v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int) 22102v2di __builtin_ia32_pcmpeqq (v2di, v2di) 22103v8hi __builtin_ia32_phminposuw128 (v8hi) 22104v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi) 22105v4si __builtin_ia32_pmaxsd128 (v4si, v4si) 22106v4si __builtin_ia32_pmaxud128 (v4si, v4si) 22107v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi) 22108v16qi __builtin_ia32_pminsb128 (v16qi, v16qi) 22109v4si __builtin_ia32_pminsd128 (v4si, v4si) 22110v4si __builtin_ia32_pminud128 (v4si, v4si) 22111v8hi __builtin_ia32_pminuw128 (v8hi, v8hi) 22112v4si __builtin_ia32_pmovsxbd128 (v16qi) 22113v2di __builtin_ia32_pmovsxbq128 (v16qi) 22114v8hi __builtin_ia32_pmovsxbw128 (v16qi) 22115v2di __builtin_ia32_pmovsxdq128 (v4si) 22116v4si __builtin_ia32_pmovsxwd128 (v8hi) 22117v2di __builtin_ia32_pmovsxwq128 (v8hi) 22118v4si __builtin_ia32_pmovzxbd128 (v16qi) 22119v2di __builtin_ia32_pmovzxbq128 (v16qi) 22120v8hi __builtin_ia32_pmovzxbw128 (v16qi) 22121v2di __builtin_ia32_pmovzxdq128 (v4si) 22122v4si __builtin_ia32_pmovzxwd128 (v8hi) 22123v2di __builtin_ia32_pmovzxwq128 (v8hi) 22124v2di __builtin_ia32_pmuldq128 (v4si, v4si) 22125v4si __builtin_ia32_pmulld128 (v4si, v4si) 22126int __builtin_ia32_ptestc128 (v2di, v2di) 22127int __builtin_ia32_ptestnzc128 (v2di, v2di) 22128int __builtin_ia32_ptestz128 (v2di, v2di) 22129v2df __builtin_ia32_roundpd (v2df, const int) 22130v4sf __builtin_ia32_roundps (v4sf, const int) 22131v2df __builtin_ia32_roundsd (v2df, v2df, const int) 22132v4sf __builtin_ia32_roundss (v4sf, v4sf, const int) 22133@end smallexample 22134 22135The following built-in functions are available when @option{-msse4.1} is 22136used. 22137 22138@table @code 22139@item v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int) 22140Generates the @code{insertps} machine instruction. 22141@item int __builtin_ia32_vec_ext_v16qi (v16qi, const int) 22142Generates the @code{pextrb} machine instruction. 22143@item v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int) 22144Generates the @code{pinsrb} machine instruction. 22145@item v4si __builtin_ia32_vec_set_v4si (v4si, int, const int) 22146Generates the @code{pinsrd} machine instruction. 22147@item v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int) 22148Generates the @code{pinsrq} machine instruction in 64bit mode. 22149@end table 22150 22151The following built-in functions are changed to generate new SSE4.1 22152instructions when @option{-msse4.1} is used. 22153 22154@table @code 22155@item float __builtin_ia32_vec_ext_v4sf (v4sf, const int) 22156Generates the @code{extractps} machine instruction. 22157@item int __builtin_ia32_vec_ext_v4si (v4si, const int) 22158Generates the @code{pextrd} machine instruction. 22159@item long long __builtin_ia32_vec_ext_v2di (v2di, const int) 22160Generates the @code{pextrq} machine instruction in 64bit mode. 22161@end table 22162 22163The following built-in functions are available when @option{-msse4.2} is 22164used. All of them generate the machine instruction that is part of the 22165name. 22166 22167@smallexample 22168v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int) 22169int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int) 22170int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int) 22171int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int) 22172int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int) 22173int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int) 22174int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int) 22175v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int) 22176int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int) 22177int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int) 22178int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int) 22179int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int) 22180int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int) 22181int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int) 22182v2di __builtin_ia32_pcmpgtq (v2di, v2di) 22183@end smallexample 22184 22185The following built-in functions are available when @option{-msse4.2} is 22186used. 22187 22188@table @code 22189@item unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char) 22190Generates the @code{crc32b} machine instruction. 22191@item unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short) 22192Generates the @code{crc32w} machine instruction. 22193@item unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int) 22194Generates the @code{crc32l} machine instruction. 22195@item unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long) 22196Generates the @code{crc32q} machine instruction. 22197@end table 22198 22199The following built-in functions are changed to generate new SSE4.2 22200instructions when @option{-msse4.2} is used. 22201 22202@table @code 22203@item int __builtin_popcount (unsigned int) 22204Generates the @code{popcntl} machine instruction. 22205@item int __builtin_popcountl (unsigned long) 22206Generates the @code{popcntl} or @code{popcntq} machine instruction, 22207depending on the size of @code{unsigned long}. 22208@item int __builtin_popcountll (unsigned long long) 22209Generates the @code{popcntq} machine instruction. 22210@end table 22211 22212The following built-in functions are available when @option{-mavx} is 22213used. All of them generate the machine instruction that is part of the 22214name. 22215 22216@smallexample 22217v4df __builtin_ia32_addpd256 (v4df,v4df) 22218v8sf __builtin_ia32_addps256 (v8sf,v8sf) 22219v4df __builtin_ia32_addsubpd256 (v4df,v4df) 22220v8sf __builtin_ia32_addsubps256 (v8sf,v8sf) 22221v4df __builtin_ia32_andnpd256 (v4df,v4df) 22222v8sf __builtin_ia32_andnps256 (v8sf,v8sf) 22223v4df __builtin_ia32_andpd256 (v4df,v4df) 22224v8sf __builtin_ia32_andps256 (v8sf,v8sf) 22225v4df __builtin_ia32_blendpd256 (v4df,v4df,int) 22226v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int) 22227v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df) 22228v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf) 22229v2df __builtin_ia32_cmppd (v2df,v2df,int) 22230v4df __builtin_ia32_cmppd256 (v4df,v4df,int) 22231v4sf __builtin_ia32_cmpps (v4sf,v4sf,int) 22232v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int) 22233v2df __builtin_ia32_cmpsd (v2df,v2df,int) 22234v4sf __builtin_ia32_cmpss (v4sf,v4sf,int) 22235v4df __builtin_ia32_cvtdq2pd256 (v4si) 22236v8sf __builtin_ia32_cvtdq2ps256 (v8si) 22237v4si __builtin_ia32_cvtpd2dq256 (v4df) 22238v4sf __builtin_ia32_cvtpd2ps256 (v4df) 22239v8si __builtin_ia32_cvtps2dq256 (v8sf) 22240v4df __builtin_ia32_cvtps2pd256 (v4sf) 22241v4si __builtin_ia32_cvttpd2dq256 (v4df) 22242v8si __builtin_ia32_cvttps2dq256 (v8sf) 22243v4df __builtin_ia32_divpd256 (v4df,v4df) 22244v8sf __builtin_ia32_divps256 (v8sf,v8sf) 22245v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int) 22246v4df __builtin_ia32_haddpd256 (v4df,v4df) 22247v8sf __builtin_ia32_haddps256 (v8sf,v8sf) 22248v4df __builtin_ia32_hsubpd256 (v4df,v4df) 22249v8sf __builtin_ia32_hsubps256 (v8sf,v8sf) 22250v32qi __builtin_ia32_lddqu256 (pcchar) 22251v32qi __builtin_ia32_loaddqu256 (pcchar) 22252v4df __builtin_ia32_loadupd256 (pcdouble) 22253v8sf __builtin_ia32_loadups256 (pcfloat) 22254v2df __builtin_ia32_maskloadpd (pcv2df,v2df) 22255v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df) 22256v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf) 22257v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf) 22258void __builtin_ia32_maskstorepd (pv2df,v2df,v2df) 22259void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df) 22260void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf) 22261void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf) 22262v4df __builtin_ia32_maxpd256 (v4df,v4df) 22263v8sf __builtin_ia32_maxps256 (v8sf,v8sf) 22264v4df __builtin_ia32_minpd256 (v4df,v4df) 22265v8sf __builtin_ia32_minps256 (v8sf,v8sf) 22266v4df __builtin_ia32_movddup256 (v4df) 22267int __builtin_ia32_movmskpd256 (v4df) 22268int __builtin_ia32_movmskps256 (v8sf) 22269v8sf __builtin_ia32_movshdup256 (v8sf) 22270v8sf __builtin_ia32_movsldup256 (v8sf) 22271v4df __builtin_ia32_mulpd256 (v4df,v4df) 22272v8sf __builtin_ia32_mulps256 (v8sf,v8sf) 22273v4df __builtin_ia32_orpd256 (v4df,v4df) 22274v8sf __builtin_ia32_orps256 (v8sf,v8sf) 22275v2df __builtin_ia32_pd_pd256 (v4df) 22276v4df __builtin_ia32_pd256_pd (v2df) 22277v4sf __builtin_ia32_ps_ps256 (v8sf) 22278v8sf __builtin_ia32_ps256_ps (v4sf) 22279int __builtin_ia32_ptestc256 (v4di,v4di,ptest) 22280int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest) 22281int __builtin_ia32_ptestz256 (v4di,v4di,ptest) 22282v8sf __builtin_ia32_rcpps256 (v8sf) 22283v4df __builtin_ia32_roundpd256 (v4df,int) 22284v8sf __builtin_ia32_roundps256 (v8sf,int) 22285v8sf __builtin_ia32_rsqrtps_nr256 (v8sf) 22286v8sf __builtin_ia32_rsqrtps256 (v8sf) 22287v4df __builtin_ia32_shufpd256 (v4df,v4df,int) 22288v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int) 22289v4si __builtin_ia32_si_si256 (v8si) 22290v8si __builtin_ia32_si256_si (v4si) 22291v4df __builtin_ia32_sqrtpd256 (v4df) 22292v8sf __builtin_ia32_sqrtps_nr256 (v8sf) 22293v8sf __builtin_ia32_sqrtps256 (v8sf) 22294void __builtin_ia32_storedqu256 (pchar,v32qi) 22295void __builtin_ia32_storeupd256 (pdouble,v4df) 22296void __builtin_ia32_storeups256 (pfloat,v8sf) 22297v4df __builtin_ia32_subpd256 (v4df,v4df) 22298v8sf __builtin_ia32_subps256 (v8sf,v8sf) 22299v4df __builtin_ia32_unpckhpd256 (v4df,v4df) 22300v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf) 22301v4df __builtin_ia32_unpcklpd256 (v4df,v4df) 22302v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf) 22303v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df) 22304v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf) 22305v4df __builtin_ia32_vbroadcastsd256 (pcdouble) 22306v4sf __builtin_ia32_vbroadcastss (pcfloat) 22307v8sf __builtin_ia32_vbroadcastss256 (pcfloat) 22308v2df __builtin_ia32_vextractf128_pd256 (v4df,int) 22309v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int) 22310v4si __builtin_ia32_vextractf128_si256 (v8si,int) 22311v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int) 22312v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int) 22313v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int) 22314v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int) 22315v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int) 22316v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int) 22317v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int) 22318v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int) 22319v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int) 22320v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int) 22321v2df __builtin_ia32_vpermilpd (v2df,int) 22322v4df __builtin_ia32_vpermilpd256 (v4df,int) 22323v4sf __builtin_ia32_vpermilps (v4sf,int) 22324v8sf __builtin_ia32_vpermilps256 (v8sf,int) 22325v2df __builtin_ia32_vpermilvarpd (v2df,v2di) 22326v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di) 22327v4sf __builtin_ia32_vpermilvarps (v4sf,v4si) 22328v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si) 22329int __builtin_ia32_vtestcpd (v2df,v2df,ptest) 22330int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest) 22331int __builtin_ia32_vtestcps (v4sf,v4sf,ptest) 22332int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest) 22333int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest) 22334int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest) 22335int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest) 22336int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest) 22337int __builtin_ia32_vtestzpd (v2df,v2df,ptest) 22338int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest) 22339int __builtin_ia32_vtestzps (v4sf,v4sf,ptest) 22340int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest) 22341void __builtin_ia32_vzeroall (void) 22342void __builtin_ia32_vzeroupper (void) 22343v4df __builtin_ia32_xorpd256 (v4df,v4df) 22344v8sf __builtin_ia32_xorps256 (v8sf,v8sf) 22345@end smallexample 22346 22347The following built-in functions are available when @option{-mavx2} is 22348used. All of them generate the machine instruction that is part of the 22349name. 22350 22351@smallexample 22352v32qi __builtin_ia32_mpsadbw256 (v32qi,v32qi,int) 22353v32qi __builtin_ia32_pabsb256 (v32qi) 22354v16hi __builtin_ia32_pabsw256 (v16hi) 22355v8si __builtin_ia32_pabsd256 (v8si) 22356v16hi __builtin_ia32_packssdw256 (v8si,v8si) 22357v32qi __builtin_ia32_packsswb256 (v16hi,v16hi) 22358v16hi __builtin_ia32_packusdw256 (v8si,v8si) 22359v32qi __builtin_ia32_packuswb256 (v16hi,v16hi) 22360v32qi __builtin_ia32_paddb256 (v32qi,v32qi) 22361v16hi __builtin_ia32_paddw256 (v16hi,v16hi) 22362v8si __builtin_ia32_paddd256 (v8si,v8si) 22363v4di __builtin_ia32_paddq256 (v4di,v4di) 22364v32qi __builtin_ia32_paddsb256 (v32qi,v32qi) 22365v16hi __builtin_ia32_paddsw256 (v16hi,v16hi) 22366v32qi __builtin_ia32_paddusb256 (v32qi,v32qi) 22367v16hi __builtin_ia32_paddusw256 (v16hi,v16hi) 22368v4di __builtin_ia32_palignr256 (v4di,v4di,int) 22369v4di __builtin_ia32_andsi256 (v4di,v4di) 22370v4di __builtin_ia32_andnotsi256 (v4di,v4di) 22371v32qi __builtin_ia32_pavgb256 (v32qi,v32qi) 22372v16hi __builtin_ia32_pavgw256 (v16hi,v16hi) 22373v32qi __builtin_ia32_pblendvb256 (v32qi,v32qi,v32qi) 22374v16hi __builtin_ia32_pblendw256 (v16hi,v16hi,int) 22375v32qi __builtin_ia32_pcmpeqb256 (v32qi,v32qi) 22376v16hi __builtin_ia32_pcmpeqw256 (v16hi,v16hi) 22377v8si __builtin_ia32_pcmpeqd256 (c8si,v8si) 22378v4di __builtin_ia32_pcmpeqq256 (v4di,v4di) 22379v32qi __builtin_ia32_pcmpgtb256 (v32qi,v32qi) 22380v16hi __builtin_ia32_pcmpgtw256 (16hi,v16hi) 22381v8si __builtin_ia32_pcmpgtd256 (v8si,v8si) 22382v4di __builtin_ia32_pcmpgtq256 (v4di,v4di) 22383v16hi __builtin_ia32_phaddw256 (v16hi,v16hi) 22384v8si __builtin_ia32_phaddd256 (v8si,v8si) 22385v16hi __builtin_ia32_phaddsw256 (v16hi,v16hi) 22386v16hi __builtin_ia32_phsubw256 (v16hi,v16hi) 22387v8si __builtin_ia32_phsubd256 (v8si,v8si) 22388v16hi __builtin_ia32_phsubsw256 (v16hi,v16hi) 22389v32qi __builtin_ia32_pmaddubsw256 (v32qi,v32qi) 22390v16hi __builtin_ia32_pmaddwd256 (v16hi,v16hi) 22391v32qi __builtin_ia32_pmaxsb256 (v32qi,v32qi) 22392v16hi __builtin_ia32_pmaxsw256 (v16hi,v16hi) 22393v8si __builtin_ia32_pmaxsd256 (v8si,v8si) 22394v32qi __builtin_ia32_pmaxub256 (v32qi,v32qi) 22395v16hi __builtin_ia32_pmaxuw256 (v16hi,v16hi) 22396v8si __builtin_ia32_pmaxud256 (v8si,v8si) 22397v32qi __builtin_ia32_pminsb256 (v32qi,v32qi) 22398v16hi __builtin_ia32_pminsw256 (v16hi,v16hi) 22399v8si __builtin_ia32_pminsd256 (v8si,v8si) 22400v32qi __builtin_ia32_pminub256 (v32qi,v32qi) 22401v16hi __builtin_ia32_pminuw256 (v16hi,v16hi) 22402v8si __builtin_ia32_pminud256 (v8si,v8si) 22403int __builtin_ia32_pmovmskb256 (v32qi) 22404v16hi __builtin_ia32_pmovsxbw256 (v16qi) 22405v8si __builtin_ia32_pmovsxbd256 (v16qi) 22406v4di __builtin_ia32_pmovsxbq256 (v16qi) 22407v8si __builtin_ia32_pmovsxwd256 (v8hi) 22408v4di __builtin_ia32_pmovsxwq256 (v8hi) 22409v4di __builtin_ia32_pmovsxdq256 (v4si) 22410v16hi __builtin_ia32_pmovzxbw256 (v16qi) 22411v8si __builtin_ia32_pmovzxbd256 (v16qi) 22412v4di __builtin_ia32_pmovzxbq256 (v16qi) 22413v8si __builtin_ia32_pmovzxwd256 (v8hi) 22414v4di __builtin_ia32_pmovzxwq256 (v8hi) 22415v4di __builtin_ia32_pmovzxdq256 (v4si) 22416v4di __builtin_ia32_pmuldq256 (v8si,v8si) 22417v16hi __builtin_ia32_pmulhrsw256 (v16hi, v16hi) 22418v16hi __builtin_ia32_pmulhuw256 (v16hi,v16hi) 22419v16hi __builtin_ia32_pmulhw256 (v16hi,v16hi) 22420v16hi __builtin_ia32_pmullw256 (v16hi,v16hi) 22421v8si __builtin_ia32_pmulld256 (v8si,v8si) 22422v4di __builtin_ia32_pmuludq256 (v8si,v8si) 22423v4di __builtin_ia32_por256 (v4di,v4di) 22424v16hi __builtin_ia32_psadbw256 (v32qi,v32qi) 22425v32qi __builtin_ia32_pshufb256 (v32qi,v32qi) 22426v8si __builtin_ia32_pshufd256 (v8si,int) 22427v16hi __builtin_ia32_pshufhw256 (v16hi,int) 22428v16hi __builtin_ia32_pshuflw256 (v16hi,int) 22429v32qi __builtin_ia32_psignb256 (v32qi,v32qi) 22430v16hi __builtin_ia32_psignw256 (v16hi,v16hi) 22431v8si __builtin_ia32_psignd256 (v8si,v8si) 22432v4di __builtin_ia32_pslldqi256 (v4di,int) 22433v16hi __builtin_ia32_psllwi256 (16hi,int) 22434v16hi __builtin_ia32_psllw256(v16hi,v8hi) 22435v8si __builtin_ia32_pslldi256 (v8si,int) 22436v8si __builtin_ia32_pslld256(v8si,v4si) 22437v4di __builtin_ia32_psllqi256 (v4di,int) 22438v4di __builtin_ia32_psllq256(v4di,v2di) 22439v16hi __builtin_ia32_psrawi256 (v16hi,int) 22440v16hi __builtin_ia32_psraw256 (v16hi,v8hi) 22441v8si __builtin_ia32_psradi256 (v8si,int) 22442v8si __builtin_ia32_psrad256 (v8si,v4si) 22443v4di __builtin_ia32_psrldqi256 (v4di, int) 22444v16hi __builtin_ia32_psrlwi256 (v16hi,int) 22445v16hi __builtin_ia32_psrlw256 (v16hi,v8hi) 22446v8si __builtin_ia32_psrldi256 (v8si,int) 22447v8si __builtin_ia32_psrld256 (v8si,v4si) 22448v4di __builtin_ia32_psrlqi256 (v4di,int) 22449v4di __builtin_ia32_psrlq256(v4di,v2di) 22450v32qi __builtin_ia32_psubb256 (v32qi,v32qi) 22451v32hi __builtin_ia32_psubw256 (v16hi,v16hi) 22452v8si __builtin_ia32_psubd256 (v8si,v8si) 22453v4di __builtin_ia32_psubq256 (v4di,v4di) 22454v32qi __builtin_ia32_psubsb256 (v32qi,v32qi) 22455v16hi __builtin_ia32_psubsw256 (v16hi,v16hi) 22456v32qi __builtin_ia32_psubusb256 (v32qi,v32qi) 22457v16hi __builtin_ia32_psubusw256 (v16hi,v16hi) 22458v32qi __builtin_ia32_punpckhbw256 (v32qi,v32qi) 22459v16hi __builtin_ia32_punpckhwd256 (v16hi,v16hi) 22460v8si __builtin_ia32_punpckhdq256 (v8si,v8si) 22461v4di __builtin_ia32_punpckhqdq256 (v4di,v4di) 22462v32qi __builtin_ia32_punpcklbw256 (v32qi,v32qi) 22463v16hi __builtin_ia32_punpcklwd256 (v16hi,v16hi) 22464v8si __builtin_ia32_punpckldq256 (v8si,v8si) 22465v4di __builtin_ia32_punpcklqdq256 (v4di,v4di) 22466v4di __builtin_ia32_pxor256 (v4di,v4di) 22467v4di __builtin_ia32_movntdqa256 (pv4di) 22468v4sf __builtin_ia32_vbroadcastss_ps (v4sf) 22469v8sf __builtin_ia32_vbroadcastss_ps256 (v4sf) 22470v4df __builtin_ia32_vbroadcastsd_pd256 (v2df) 22471v4di __builtin_ia32_vbroadcastsi256 (v2di) 22472v4si __builtin_ia32_pblendd128 (v4si,v4si) 22473v8si __builtin_ia32_pblendd256 (v8si,v8si) 22474v32qi __builtin_ia32_pbroadcastb256 (v16qi) 22475v16hi __builtin_ia32_pbroadcastw256 (v8hi) 22476v8si __builtin_ia32_pbroadcastd256 (v4si) 22477v4di __builtin_ia32_pbroadcastq256 (v2di) 22478v16qi __builtin_ia32_pbroadcastb128 (v16qi) 22479v8hi __builtin_ia32_pbroadcastw128 (v8hi) 22480v4si __builtin_ia32_pbroadcastd128 (v4si) 22481v2di __builtin_ia32_pbroadcastq128 (v2di) 22482v8si __builtin_ia32_permvarsi256 (v8si,v8si) 22483v4df __builtin_ia32_permdf256 (v4df,int) 22484v8sf __builtin_ia32_permvarsf256 (v8sf,v8sf) 22485v4di __builtin_ia32_permdi256 (v4di,int) 22486v4di __builtin_ia32_permti256 (v4di,v4di,int) 22487v4di __builtin_ia32_extract128i256 (v4di,int) 22488v4di __builtin_ia32_insert128i256 (v4di,v2di,int) 22489v8si __builtin_ia32_maskloadd256 (pcv8si,v8si) 22490v4di __builtin_ia32_maskloadq256 (pcv4di,v4di) 22491v4si __builtin_ia32_maskloadd (pcv4si,v4si) 22492v2di __builtin_ia32_maskloadq (pcv2di,v2di) 22493void __builtin_ia32_maskstored256 (pv8si,v8si,v8si) 22494void __builtin_ia32_maskstoreq256 (pv4di,v4di,v4di) 22495void __builtin_ia32_maskstored (pv4si,v4si,v4si) 22496void __builtin_ia32_maskstoreq (pv2di,v2di,v2di) 22497v8si __builtin_ia32_psllv8si (v8si,v8si) 22498v4si __builtin_ia32_psllv4si (v4si,v4si) 22499v4di __builtin_ia32_psllv4di (v4di,v4di) 22500v2di __builtin_ia32_psllv2di (v2di,v2di) 22501v8si __builtin_ia32_psrav8si (v8si,v8si) 22502v4si __builtin_ia32_psrav4si (v4si,v4si) 22503v8si __builtin_ia32_psrlv8si (v8si,v8si) 22504v4si __builtin_ia32_psrlv4si (v4si,v4si) 22505v4di __builtin_ia32_psrlv4di (v4di,v4di) 22506v2di __builtin_ia32_psrlv2di (v2di,v2di) 22507v2df __builtin_ia32_gathersiv2df (v2df, pcdouble,v4si,v2df,int) 22508v4df __builtin_ia32_gathersiv4df (v4df, pcdouble,v4si,v4df,int) 22509v2df __builtin_ia32_gatherdiv2df (v2df, pcdouble,v2di,v2df,int) 22510v4df __builtin_ia32_gatherdiv4df (v4df, pcdouble,v4di,v4df,int) 22511v4sf __builtin_ia32_gathersiv4sf (v4sf, pcfloat,v4si,v4sf,int) 22512v8sf __builtin_ia32_gathersiv8sf (v8sf, pcfloat,v8si,v8sf,int) 22513v4sf __builtin_ia32_gatherdiv4sf (v4sf, pcfloat,v2di,v4sf,int) 22514v4sf __builtin_ia32_gatherdiv4sf256 (v4sf, pcfloat,v4di,v4sf,int) 22515v2di __builtin_ia32_gathersiv2di (v2di, pcint64,v4si,v2di,int) 22516v4di __builtin_ia32_gathersiv4di (v4di, pcint64,v4si,v4di,int) 22517v2di __builtin_ia32_gatherdiv2di (v2di, pcint64,v2di,v2di,int) 22518v4di __builtin_ia32_gatherdiv4di (v4di, pcint64,v4di,v4di,int) 22519v4si __builtin_ia32_gathersiv4si (v4si, pcint,v4si,v4si,int) 22520v8si __builtin_ia32_gathersiv8si (v8si, pcint,v8si,v8si,int) 22521v4si __builtin_ia32_gatherdiv4si (v4si, pcint,v2di,v4si,int) 22522v4si __builtin_ia32_gatherdiv4si256 (v4si, pcint,v4di,v4si,int) 22523@end smallexample 22524 22525The following built-in functions are available when @option{-maes} is 22526used. All of them generate the machine instruction that is part of the 22527name. 22528 22529@smallexample 22530v2di __builtin_ia32_aesenc128 (v2di, v2di) 22531v2di __builtin_ia32_aesenclast128 (v2di, v2di) 22532v2di __builtin_ia32_aesdec128 (v2di, v2di) 22533v2di __builtin_ia32_aesdeclast128 (v2di, v2di) 22534v2di __builtin_ia32_aeskeygenassist128 (v2di, const int) 22535v2di __builtin_ia32_aesimc128 (v2di) 22536@end smallexample 22537 22538The following built-in function is available when @option{-mpclmul} is 22539used. 22540 22541@table @code 22542@item v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int) 22543Generates the @code{pclmulqdq} machine instruction. 22544@end table 22545 22546The following built-in function is available when @option{-mfsgsbase} is 22547used. All of them generate the machine instruction that is part of the 22548name. 22549 22550@smallexample 22551unsigned int __builtin_ia32_rdfsbase32 (void) 22552unsigned long long __builtin_ia32_rdfsbase64 (void) 22553unsigned int __builtin_ia32_rdgsbase32 (void) 22554unsigned long long __builtin_ia32_rdgsbase64 (void) 22555void _writefsbase_u32 (unsigned int) 22556void _writefsbase_u64 (unsigned long long) 22557void _writegsbase_u32 (unsigned int) 22558void _writegsbase_u64 (unsigned long long) 22559@end smallexample 22560 22561The following built-in function is available when @option{-mrdrnd} is 22562used. All of them generate the machine instruction that is part of the 22563name. 22564 22565@smallexample 22566unsigned int __builtin_ia32_rdrand16_step (unsigned short *) 22567unsigned int __builtin_ia32_rdrand32_step (unsigned int *) 22568unsigned int __builtin_ia32_rdrand64_step (unsigned long long *) 22569@end smallexample 22570 22571The following built-in function is available when @option{-mptwrite} is 22572used. All of them generate the machine instruction that is part of the 22573name. 22574 22575@smallexample 22576void __builtin_ia32_ptwrite32 (unsigned) 22577void __builtin_ia32_ptwrite64 (unsigned long long) 22578@end smallexample 22579 22580The following built-in functions are available when @option{-msse4a} is used. 22581All of them generate the machine instruction that is part of the name. 22582 22583@smallexample 22584void __builtin_ia32_movntsd (double *, v2df) 22585void __builtin_ia32_movntss (float *, v4sf) 22586v2di __builtin_ia32_extrq (v2di, v16qi) 22587v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int) 22588v2di __builtin_ia32_insertq (v2di, v2di) 22589v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int) 22590@end smallexample 22591 22592The following built-in functions are available when @option{-mxop} is used. 22593@smallexample 22594v2df __builtin_ia32_vfrczpd (v2df) 22595v4sf __builtin_ia32_vfrczps (v4sf) 22596v2df __builtin_ia32_vfrczsd (v2df) 22597v4sf __builtin_ia32_vfrczss (v4sf) 22598v4df __builtin_ia32_vfrczpd256 (v4df) 22599v8sf __builtin_ia32_vfrczps256 (v8sf) 22600v2di __builtin_ia32_vpcmov (v2di, v2di, v2di) 22601v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di) 22602v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si) 22603v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi) 22604v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi) 22605v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df) 22606v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf) 22607v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di) 22608v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si) 22609v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi) 22610v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi) 22611v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df) 22612v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf) 22613v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi) 22614v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi) 22615v4si __builtin_ia32_vpcomeqd (v4si, v4si) 22616v2di __builtin_ia32_vpcomeqq (v2di, v2di) 22617v16qi __builtin_ia32_vpcomequb (v16qi, v16qi) 22618v4si __builtin_ia32_vpcomequd (v4si, v4si) 22619v2di __builtin_ia32_vpcomequq (v2di, v2di) 22620v8hi __builtin_ia32_vpcomequw (v8hi, v8hi) 22621v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi) 22622v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi) 22623v4si __builtin_ia32_vpcomfalsed (v4si, v4si) 22624v2di __builtin_ia32_vpcomfalseq (v2di, v2di) 22625v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi) 22626v4si __builtin_ia32_vpcomfalseud (v4si, v4si) 22627v2di __builtin_ia32_vpcomfalseuq (v2di, v2di) 22628v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi) 22629v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi) 22630v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi) 22631v4si __builtin_ia32_vpcomged (v4si, v4si) 22632v2di __builtin_ia32_vpcomgeq (v2di, v2di) 22633v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi) 22634v4si __builtin_ia32_vpcomgeud (v4si, v4si) 22635v2di __builtin_ia32_vpcomgeuq (v2di, v2di) 22636v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi) 22637v8hi __builtin_ia32_vpcomgew (v8hi, v8hi) 22638v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi) 22639v4si __builtin_ia32_vpcomgtd (v4si, v4si) 22640v2di __builtin_ia32_vpcomgtq (v2di, v2di) 22641v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi) 22642v4si __builtin_ia32_vpcomgtud (v4si, v4si) 22643v2di __builtin_ia32_vpcomgtuq (v2di, v2di) 22644v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi) 22645v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi) 22646v16qi __builtin_ia32_vpcomleb (v16qi, v16qi) 22647v4si __builtin_ia32_vpcomled (v4si, v4si) 22648v2di __builtin_ia32_vpcomleq (v2di, v2di) 22649v16qi __builtin_ia32_vpcomleub (v16qi, v16qi) 22650v4si __builtin_ia32_vpcomleud (v4si, v4si) 22651v2di __builtin_ia32_vpcomleuq (v2di, v2di) 22652v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi) 22653v8hi __builtin_ia32_vpcomlew (v8hi, v8hi) 22654v16qi __builtin_ia32_vpcomltb (v16qi, v16qi) 22655v4si __builtin_ia32_vpcomltd (v4si, v4si) 22656v2di __builtin_ia32_vpcomltq (v2di, v2di) 22657v16qi __builtin_ia32_vpcomltub (v16qi, v16qi) 22658v4si __builtin_ia32_vpcomltud (v4si, v4si) 22659v2di __builtin_ia32_vpcomltuq (v2di, v2di) 22660v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi) 22661v8hi __builtin_ia32_vpcomltw (v8hi, v8hi) 22662v16qi __builtin_ia32_vpcomneb (v16qi, v16qi) 22663v4si __builtin_ia32_vpcomned (v4si, v4si) 22664v2di __builtin_ia32_vpcomneq (v2di, v2di) 22665v16qi __builtin_ia32_vpcomneub (v16qi, v16qi) 22666v4si __builtin_ia32_vpcomneud (v4si, v4si) 22667v2di __builtin_ia32_vpcomneuq (v2di, v2di) 22668v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi) 22669v8hi __builtin_ia32_vpcomnew (v8hi, v8hi) 22670v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi) 22671v4si __builtin_ia32_vpcomtrued (v4si, v4si) 22672v2di __builtin_ia32_vpcomtrueq (v2di, v2di) 22673v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi) 22674v4si __builtin_ia32_vpcomtrueud (v4si, v4si) 22675v2di __builtin_ia32_vpcomtrueuq (v2di, v2di) 22676v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi) 22677v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi) 22678v4si __builtin_ia32_vphaddbd (v16qi) 22679v2di __builtin_ia32_vphaddbq (v16qi) 22680v8hi __builtin_ia32_vphaddbw (v16qi) 22681v2di __builtin_ia32_vphadddq (v4si) 22682v4si __builtin_ia32_vphaddubd (v16qi) 22683v2di __builtin_ia32_vphaddubq (v16qi) 22684v8hi __builtin_ia32_vphaddubw (v16qi) 22685v2di __builtin_ia32_vphaddudq (v4si) 22686v4si __builtin_ia32_vphadduwd (v8hi) 22687v2di __builtin_ia32_vphadduwq (v8hi) 22688v4si __builtin_ia32_vphaddwd (v8hi) 22689v2di __builtin_ia32_vphaddwq (v8hi) 22690v8hi __builtin_ia32_vphsubbw (v16qi) 22691v2di __builtin_ia32_vphsubdq (v4si) 22692v4si __builtin_ia32_vphsubwd (v8hi) 22693v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si) 22694v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di) 22695v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di) 22696v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si) 22697v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di) 22698v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di) 22699v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si) 22700v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi) 22701v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si) 22702v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi) 22703v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si) 22704v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si) 22705v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi) 22706v16qi __builtin_ia32_vprotb (v16qi, v16qi) 22707v4si __builtin_ia32_vprotd (v4si, v4si) 22708v2di __builtin_ia32_vprotq (v2di, v2di) 22709v8hi __builtin_ia32_vprotw (v8hi, v8hi) 22710v16qi __builtin_ia32_vpshab (v16qi, v16qi) 22711v4si __builtin_ia32_vpshad (v4si, v4si) 22712v2di __builtin_ia32_vpshaq (v2di, v2di) 22713v8hi __builtin_ia32_vpshaw (v8hi, v8hi) 22714v16qi __builtin_ia32_vpshlb (v16qi, v16qi) 22715v4si __builtin_ia32_vpshld (v4si, v4si) 22716v2di __builtin_ia32_vpshlq (v2di, v2di) 22717v8hi __builtin_ia32_vpshlw (v8hi, v8hi) 22718@end smallexample 22719 22720The following built-in functions are available when @option{-mfma4} is used. 22721All of them generate the machine instruction that is part of the name. 22722 22723@smallexample 22724v2df __builtin_ia32_vfmaddpd (v2df, v2df, v2df) 22725v4sf __builtin_ia32_vfmaddps (v4sf, v4sf, v4sf) 22726v2df __builtin_ia32_vfmaddsd (v2df, v2df, v2df) 22727v4sf __builtin_ia32_vfmaddss (v4sf, v4sf, v4sf) 22728v2df __builtin_ia32_vfmsubpd (v2df, v2df, v2df) 22729v4sf __builtin_ia32_vfmsubps (v4sf, v4sf, v4sf) 22730v2df __builtin_ia32_vfmsubsd (v2df, v2df, v2df) 22731v4sf __builtin_ia32_vfmsubss (v4sf, v4sf, v4sf) 22732v2df __builtin_ia32_vfnmaddpd (v2df, v2df, v2df) 22733v4sf __builtin_ia32_vfnmaddps (v4sf, v4sf, v4sf) 22734v2df __builtin_ia32_vfnmaddsd (v2df, v2df, v2df) 22735v4sf __builtin_ia32_vfnmaddss (v4sf, v4sf, v4sf) 22736v2df __builtin_ia32_vfnmsubpd (v2df, v2df, v2df) 22737v4sf __builtin_ia32_vfnmsubps (v4sf, v4sf, v4sf) 22738v2df __builtin_ia32_vfnmsubsd (v2df, v2df, v2df) 22739v4sf __builtin_ia32_vfnmsubss (v4sf, v4sf, v4sf) 22740v2df __builtin_ia32_vfmaddsubpd (v2df, v2df, v2df) 22741v4sf __builtin_ia32_vfmaddsubps (v4sf, v4sf, v4sf) 22742v2df __builtin_ia32_vfmsubaddpd (v2df, v2df, v2df) 22743v4sf __builtin_ia32_vfmsubaddps (v4sf, v4sf, v4sf) 22744v4df __builtin_ia32_vfmaddpd256 (v4df, v4df, v4df) 22745v8sf __builtin_ia32_vfmaddps256 (v8sf, v8sf, v8sf) 22746v4df __builtin_ia32_vfmsubpd256 (v4df, v4df, v4df) 22747v8sf __builtin_ia32_vfmsubps256 (v8sf, v8sf, v8sf) 22748v4df __builtin_ia32_vfnmaddpd256 (v4df, v4df, v4df) 22749v8sf __builtin_ia32_vfnmaddps256 (v8sf, v8sf, v8sf) 22750v4df __builtin_ia32_vfnmsubpd256 (v4df, v4df, v4df) 22751v8sf __builtin_ia32_vfnmsubps256 (v8sf, v8sf, v8sf) 22752v4df __builtin_ia32_vfmaddsubpd256 (v4df, v4df, v4df) 22753v8sf __builtin_ia32_vfmaddsubps256 (v8sf, v8sf, v8sf) 22754v4df __builtin_ia32_vfmsubaddpd256 (v4df, v4df, v4df) 22755v8sf __builtin_ia32_vfmsubaddps256 (v8sf, v8sf, v8sf) 22756 22757@end smallexample 22758 22759The following built-in functions are available when @option{-mlwp} is used. 22760 22761@smallexample 22762void __builtin_ia32_llwpcb16 (void *); 22763void __builtin_ia32_llwpcb32 (void *); 22764void __builtin_ia32_llwpcb64 (void *); 22765void * __builtin_ia32_llwpcb16 (void); 22766void * __builtin_ia32_llwpcb32 (void); 22767void * __builtin_ia32_llwpcb64 (void); 22768void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short) 22769void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int) 22770void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int) 22771unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short) 22772unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int) 22773unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int) 22774@end smallexample 22775 22776The following built-in functions are available when @option{-mbmi} is used. 22777All of them generate the machine instruction that is part of the name. 22778@smallexample 22779unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int); 22780unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long); 22781@end smallexample 22782 22783The following built-in functions are available when @option{-mbmi2} is used. 22784All of them generate the machine instruction that is part of the name. 22785@smallexample 22786unsigned int _bzhi_u32 (unsigned int, unsigned int) 22787unsigned int _pdep_u32 (unsigned int, unsigned int) 22788unsigned int _pext_u32 (unsigned int, unsigned int) 22789unsigned long long _bzhi_u64 (unsigned long long, unsigned long long) 22790unsigned long long _pdep_u64 (unsigned long long, unsigned long long) 22791unsigned long long _pext_u64 (unsigned long long, unsigned long long) 22792@end smallexample 22793 22794The following built-in functions are available when @option{-mlzcnt} is used. 22795All of them generate the machine instruction that is part of the name. 22796@smallexample 22797unsigned short __builtin_ia32_lzcnt_u16(unsigned short); 22798unsigned int __builtin_ia32_lzcnt_u32(unsigned int); 22799unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long); 22800@end smallexample 22801 22802The following built-in functions are available when @option{-mfxsr} is used. 22803All of them generate the machine instruction that is part of the name. 22804@smallexample 22805void __builtin_ia32_fxsave (void *) 22806void __builtin_ia32_fxrstor (void *) 22807void __builtin_ia32_fxsave64 (void *) 22808void __builtin_ia32_fxrstor64 (void *) 22809@end smallexample 22810 22811The following built-in functions are available when @option{-mxsave} is used. 22812All of them generate the machine instruction that is part of the name. 22813@smallexample 22814void __builtin_ia32_xsave (void *, long long) 22815void __builtin_ia32_xrstor (void *, long long) 22816void __builtin_ia32_xsave64 (void *, long long) 22817void __builtin_ia32_xrstor64 (void *, long long) 22818@end smallexample 22819 22820The following built-in functions are available when @option{-mxsaveopt} is used. 22821All of them generate the machine instruction that is part of the name. 22822@smallexample 22823void __builtin_ia32_xsaveopt (void *, long long) 22824void __builtin_ia32_xsaveopt64 (void *, long long) 22825@end smallexample 22826 22827The following built-in functions are available when @option{-mtbm} is used. 22828Both of them generate the immediate form of the bextr machine instruction. 22829@smallexample 22830unsigned int __builtin_ia32_bextri_u32 (unsigned int, 22831 const unsigned int); 22832unsigned long long __builtin_ia32_bextri_u64 (unsigned long long, 22833 const unsigned long long); 22834@end smallexample 22835 22836 22837The following built-in functions are available when @option{-m3dnow} is used. 22838All of them generate the machine instruction that is part of the name. 22839 22840@smallexample 22841void __builtin_ia32_femms (void) 22842v8qi __builtin_ia32_pavgusb (v8qi, v8qi) 22843v2si __builtin_ia32_pf2id (v2sf) 22844v2sf __builtin_ia32_pfacc (v2sf, v2sf) 22845v2sf __builtin_ia32_pfadd (v2sf, v2sf) 22846v2si __builtin_ia32_pfcmpeq (v2sf, v2sf) 22847v2si __builtin_ia32_pfcmpge (v2sf, v2sf) 22848v2si __builtin_ia32_pfcmpgt (v2sf, v2sf) 22849v2sf __builtin_ia32_pfmax (v2sf, v2sf) 22850v2sf __builtin_ia32_pfmin (v2sf, v2sf) 22851v2sf __builtin_ia32_pfmul (v2sf, v2sf) 22852v2sf __builtin_ia32_pfrcp (v2sf) 22853v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf) 22854v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf) 22855v2sf __builtin_ia32_pfrsqrt (v2sf) 22856v2sf __builtin_ia32_pfsub (v2sf, v2sf) 22857v2sf __builtin_ia32_pfsubr (v2sf, v2sf) 22858v2sf __builtin_ia32_pi2fd (v2si) 22859v4hi __builtin_ia32_pmulhrw (v4hi, v4hi) 22860@end smallexample 22861 22862The following built-in functions are available when @option{-m3dnowa} is used. 22863All of them generate the machine instruction that is part of the name. 22864 22865@smallexample 22866v2si __builtin_ia32_pf2iw (v2sf) 22867v2sf __builtin_ia32_pfnacc (v2sf, v2sf) 22868v2sf __builtin_ia32_pfpnacc (v2sf, v2sf) 22869v2sf __builtin_ia32_pi2fw (v2si) 22870v2sf __builtin_ia32_pswapdsf (v2sf) 22871v2si __builtin_ia32_pswapdsi (v2si) 22872@end smallexample 22873 22874The following built-in functions are available when @option{-mrtm} is used 22875They are used for restricted transactional memory. These are the internal 22876low level functions. Normally the functions in 22877@ref{x86 transactional memory intrinsics} should be used instead. 22878 22879@smallexample 22880int __builtin_ia32_xbegin () 22881void __builtin_ia32_xend () 22882void __builtin_ia32_xabort (status) 22883int __builtin_ia32_xtest () 22884@end smallexample 22885 22886The following built-in functions are available when @option{-mmwaitx} is used. 22887All of them generate the machine instruction that is part of the name. 22888@smallexample 22889void __builtin_ia32_monitorx (void *, unsigned int, unsigned int) 22890void __builtin_ia32_mwaitx (unsigned int, unsigned int, unsigned int) 22891@end smallexample 22892 22893The following built-in functions are available when @option{-mclzero} is used. 22894All of them generate the machine instruction that is part of the name. 22895@smallexample 22896void __builtin_i32_clzero (void *) 22897@end smallexample 22898 22899The following built-in functions are available when @option{-mpku} is used. 22900They generate reads and writes to PKRU. 22901@smallexample 22902void __builtin_ia32_wrpkru (unsigned int) 22903unsigned int __builtin_ia32_rdpkru () 22904@end smallexample 22905 22906The following built-in functions are available when 22907@option{-mshstk} option is used. They support shadow stack 22908machine instructions from Intel Control-flow Enforcement Technology (CET). 22909Each built-in function generates the machine instruction that is part 22910of the function's name. These are the internal low-level functions. 22911Normally the functions in @ref{x86 control-flow protection intrinsics} 22912should be used instead. 22913 22914@smallexample 22915unsigned int __builtin_ia32_rdsspd (void) 22916unsigned long long __builtin_ia32_rdsspq (void) 22917void __builtin_ia32_incsspd (unsigned int) 22918void __builtin_ia32_incsspq (unsigned long long) 22919void __builtin_ia32_saveprevssp(void); 22920void __builtin_ia32_rstorssp(void *); 22921void __builtin_ia32_wrssd(unsigned int, void *); 22922void __builtin_ia32_wrssq(unsigned long long, void *); 22923void __builtin_ia32_wrussd(unsigned int, void *); 22924void __builtin_ia32_wrussq(unsigned long long, void *); 22925void __builtin_ia32_setssbsy(void); 22926void __builtin_ia32_clrssbsy(void *); 22927@end smallexample 22928 22929@node x86 transactional memory intrinsics 22930@subsection x86 Transactional Memory Intrinsics 22931 22932These hardware transactional memory intrinsics for x86 allow you to use 22933memory transactions with RTM (Restricted Transactional Memory). 22934This support is enabled with the @option{-mrtm} option. 22935For using HLE (Hardware Lock Elision) see 22936@ref{x86 specific memory model extensions for transactional memory} instead. 22937 22938A memory transaction commits all changes to memory in an atomic way, 22939as visible to other threads. If the transaction fails it is rolled back 22940and all side effects discarded. 22941 22942Generally there is no guarantee that a memory transaction ever succeeds 22943and suitable fallback code always needs to be supplied. 22944 22945@deftypefn {RTM Function} {unsigned} _xbegin () 22946Start a RTM (Restricted Transactional Memory) transaction. 22947Returns @code{_XBEGIN_STARTED} when the transaction 22948started successfully (note this is not 0, so the constant has to be 22949explicitly tested). 22950 22951If the transaction aborts, all side effects 22952are undone and an abort code encoded as a bit mask is returned. 22953The following macros are defined: 22954 22955@table @code 22956@item _XABORT_EXPLICIT 22957Transaction was explicitly aborted with @code{_xabort}. The parameter passed 22958to @code{_xabort} is available with @code{_XABORT_CODE(status)}. 22959@item _XABORT_RETRY 22960Transaction retry is possible. 22961@item _XABORT_CONFLICT 22962Transaction abort due to a memory conflict with another thread. 22963@item _XABORT_CAPACITY 22964Transaction abort due to the transaction using too much memory. 22965@item _XABORT_DEBUG 22966Transaction abort due to a debug trap. 22967@item _XABORT_NESTED 22968Transaction abort in an inner nested transaction. 22969@end table 22970 22971There is no guarantee 22972any transaction ever succeeds, so there always needs to be a valid 22973fallback path. 22974@end deftypefn 22975 22976@deftypefn {RTM Function} {void} _xend () 22977Commit the current transaction. When no transaction is active this faults. 22978All memory side effects of the transaction become visible 22979to other threads in an atomic manner. 22980@end deftypefn 22981 22982@deftypefn {RTM Function} {int} _xtest () 22983Return a nonzero value if a transaction is currently active, otherwise 0. 22984@end deftypefn 22985 22986@deftypefn {RTM Function} {void} _xabort (status) 22987Abort the current transaction. When no transaction is active this is a no-op. 22988The @var{status} is an 8-bit constant; its value is encoded in the return 22989value from @code{_xbegin}. 22990@end deftypefn 22991 22992Here is an example showing handling for @code{_XABORT_RETRY} 22993and a fallback path for other failures: 22994 22995@smallexample 22996#include <immintrin.h> 22997 22998int n_tries, max_tries; 22999unsigned status = _XABORT_EXPLICIT; 23000... 23001 23002for (n_tries = 0; n_tries < max_tries; n_tries++) 23003 @{ 23004 status = _xbegin (); 23005 if (status == _XBEGIN_STARTED || !(status & _XABORT_RETRY)) 23006 break; 23007 @} 23008if (status == _XBEGIN_STARTED) 23009 @{ 23010 ... transaction code... 23011 _xend (); 23012 @} 23013else 23014 @{ 23015 ... non-transactional fallback path... 23016 @} 23017@end smallexample 23018 23019@noindent 23020Note that, in most cases, the transactional and non-transactional code 23021must synchronize together to ensure consistency. 23022 23023@node x86 control-flow protection intrinsics 23024@subsection x86 Control-Flow Protection Intrinsics 23025 23026@deftypefn {CET Function} {ret_type} _get_ssp (void) 23027Get the current value of shadow stack pointer if shadow stack support 23028from Intel CET is enabled in the hardware or @code{0} otherwise. 23029The @code{ret_type} is @code{unsigned long long} for 64-bit targets 23030and @code{unsigned int} for 32-bit targets. 23031@end deftypefn 23032 23033@deftypefn {CET Function} void _inc_ssp (unsigned int) 23034Increment the current shadow stack pointer by the size specified by the 23035function argument. The argument is masked to a byte value for security 23036reasons, so to increment by more than 255 bytes you must call the function 23037multiple times. 23038@end deftypefn 23039 23040The shadow stack unwind code looks like: 23041 23042@smallexample 23043#include <immintrin.h> 23044 23045/* Unwind the shadow stack for EH. */ 23046#define _Unwind_Frames_Extra(x) \ 23047 do \ 23048 @{ \ 23049 _Unwind_Word ssp = _get_ssp (); \ 23050 if (ssp != 0) \ 23051 @{ \ 23052 _Unwind_Word tmp = (x); \ 23053 while (tmp > 255) \ 23054 @{ \ 23055 _inc_ssp (tmp); \ 23056 tmp -= 255; \ 23057 @} \ 23058 _inc_ssp (tmp); \ 23059 @} \ 23060 @} \ 23061 while (0) 23062@end smallexample 23063 23064@noindent 23065This code runs unconditionally on all 64-bit processors. For 32-bit 23066processors the code runs on those that support multi-byte NOP instructions. 23067 23068@node Target Format Checks 23069@section Format Checks Specific to Particular Target Machines 23070 23071For some target machines, GCC supports additional options to the 23072format attribute 23073(@pxref{Function Attributes,,Declaring Attributes of Functions}). 23074 23075@menu 23076* Solaris Format Checks:: 23077* Darwin Format Checks:: 23078@end menu 23079 23080@node Solaris Format Checks 23081@subsection Solaris Format Checks 23082 23083Solaris targets support the @code{cmn_err} (or @code{__cmn_err__}) format 23084check. @code{cmn_err} accepts a subset of the standard @code{printf} 23085conversions, and the two-argument @code{%b} conversion for displaying 23086bit-fields. See the Solaris man page for @code{cmn_err} for more information. 23087 23088@node Darwin Format Checks 23089@subsection Darwin Format Checks 23090 23091In addition to the full set of format archetypes (attribute format style 23092arguments such as @code{printf}, @code{scanf}, @code{strftime}, and 23093@code{strfmon}), Darwin targets also support the @code{CFString} (or 23094@code{__CFString__}) archetype in the @code{format} attribute. 23095Declarations with this archetype are parsed for correct syntax 23096and argument types. However, parsing of the format string itself and 23097validating arguments against it in calls to such functions is currently 23098not performed. 23099 23100Additionally, @code{CFStringRefs} (defined by the @code{CoreFoundation} headers) may 23101also be used as format arguments. Note that the relevant headers are only likely to be 23102available on Darwin (OSX) installations. On such installations, the XCode and system 23103documentation provide descriptions of @code{CFString}, @code{CFStringRefs} and 23104associated functions. 23105 23106@node Pragmas 23107@section Pragmas Accepted by GCC 23108@cindex pragmas 23109@cindex @code{#pragma} 23110 23111GCC supports several types of pragmas, primarily in order to compile 23112code originally written for other compilers. Note that in general 23113we do not recommend the use of pragmas; @xref{Function Attributes}, 23114for further explanation. 23115 23116The GNU C preprocessor recognizes several pragmas in addition to the 23117compiler pragmas documented here. Refer to the CPP manual for more 23118information. 23119 23120@menu 23121* AArch64 Pragmas:: 23122* ARM Pragmas:: 23123* M32C Pragmas:: 23124* MeP Pragmas:: 23125* PRU Pragmas:: 23126* RS/6000 and PowerPC Pragmas:: 23127* S/390 Pragmas:: 23128* Darwin Pragmas:: 23129* Solaris Pragmas:: 23130* Symbol-Renaming Pragmas:: 23131* Structure-Layout Pragmas:: 23132* Weak Pragmas:: 23133* Diagnostic Pragmas:: 23134* Visibility Pragmas:: 23135* Push/Pop Macro Pragmas:: 23136* Function Specific Option Pragmas:: 23137* Loop-Specific Pragmas:: 23138@end menu 23139 23140@node AArch64 Pragmas 23141@subsection AArch64 Pragmas 23142 23143The pragmas defined by the AArch64 target correspond to the AArch64 23144target function attributes. They can be specified as below: 23145@smallexample 23146#pragma GCC target("string") 23147@end smallexample 23148 23149where @code{@var{string}} can be any string accepted as an AArch64 target 23150attribute. @xref{AArch64 Function Attributes}, for more details 23151on the permissible values of @code{string}. 23152 23153@node ARM Pragmas 23154@subsection ARM Pragmas 23155 23156The ARM target defines pragmas for controlling the default addition of 23157@code{long_call} and @code{short_call} attributes to functions. 23158@xref{Function Attributes}, for information about the effects of these 23159attributes. 23160 23161@table @code 23162@item long_calls 23163@cindex pragma, long_calls 23164Set all subsequent functions to have the @code{long_call} attribute. 23165 23166@item no_long_calls 23167@cindex pragma, no_long_calls 23168Set all subsequent functions to have the @code{short_call} attribute. 23169 23170@item long_calls_off 23171@cindex pragma, long_calls_off 23172Do not affect the @code{long_call} or @code{short_call} attributes of 23173subsequent functions. 23174@end table 23175 23176@node M32C Pragmas 23177@subsection M32C Pragmas 23178 23179@table @code 23180@item GCC memregs @var{number} 23181@cindex pragma, memregs 23182Overrides the command-line option @code{-memregs=} for the current 23183file. Use with care! This pragma must be before any function in the 23184file, and mixing different memregs values in different objects may 23185make them incompatible. This pragma is useful when a 23186performance-critical function uses a memreg for temporary values, 23187as it may allow you to reduce the number of memregs used. 23188 23189@item ADDRESS @var{name} @var{address} 23190@cindex pragma, address 23191For any declared symbols matching @var{name}, this does three things 23192to that symbol: it forces the symbol to be located at the given 23193address (a number), it forces the symbol to be volatile, and it 23194changes the symbol's scope to be static. This pragma exists for 23195compatibility with other compilers, but note that the common 23196@code{1234H} numeric syntax is not supported (use @code{0x1234} 23197instead). Example: 23198 23199@smallexample 23200#pragma ADDRESS port3 0x103 23201char port3; 23202@end smallexample 23203 23204@end table 23205 23206@node MeP Pragmas 23207@subsection MeP Pragmas 23208 23209@table @code 23210 23211@item custom io_volatile (on|off) 23212@cindex pragma, custom io_volatile 23213Overrides the command-line option @code{-mio-volatile} for the current 23214file. Note that for compatibility with future GCC releases, this 23215option should only be used once before any @code{io} variables in each 23216file. 23217 23218@item GCC coprocessor available @var{registers} 23219@cindex pragma, coprocessor available 23220Specifies which coprocessor registers are available to the register 23221allocator. @var{registers} may be a single register, register range 23222separated by ellipses, or comma-separated list of those. Example: 23223 23224@smallexample 23225#pragma GCC coprocessor available $c0...$c10, $c28 23226@end smallexample 23227 23228@item GCC coprocessor call_saved @var{registers} 23229@cindex pragma, coprocessor call_saved 23230Specifies which coprocessor registers are to be saved and restored by 23231any function using them. @var{registers} may be a single register, 23232register range separated by ellipses, or comma-separated list of 23233those. Example: 23234 23235@smallexample 23236#pragma GCC coprocessor call_saved $c4...$c6, $c31 23237@end smallexample 23238 23239@item GCC coprocessor subclass '(A|B|C|D)' = @var{registers} 23240@cindex pragma, coprocessor subclass 23241Creates and defines a register class. These register classes can be 23242used by inline @code{asm} constructs. @var{registers} may be a single 23243register, register range separated by ellipses, or comma-separated 23244list of those. Example: 23245 23246@smallexample 23247#pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6 23248 23249asm ("cpfoo %0" : "=B" (x)); 23250@end smallexample 23251 23252@item GCC disinterrupt @var{name} , @var{name} @dots{} 23253@cindex pragma, disinterrupt 23254For the named functions, the compiler adds code to disable interrupts 23255for the duration of those functions. If any functions so named 23256are not encountered in the source, a warning is emitted that the pragma is 23257not used. Examples: 23258 23259@smallexample 23260#pragma disinterrupt foo 23261#pragma disinterrupt bar, grill 23262int foo () @{ @dots{} @} 23263@end smallexample 23264 23265@item GCC call @var{name} , @var{name} @dots{} 23266@cindex pragma, call 23267For the named functions, the compiler always uses a register-indirect 23268call model when calling the named functions. Examples: 23269 23270@smallexample 23271extern int foo (); 23272#pragma call foo 23273@end smallexample 23274 23275@end table 23276 23277@node PRU Pragmas 23278@subsection PRU Pragmas 23279 23280@table @code 23281 23282@item ctable_entry @var{index} @var{constant_address} 23283@cindex pragma, ctable_entry 23284Specifies that the PRU CTABLE entry given by @var{index} has the value 23285@var{constant_address}. This enables GCC to emit LBCO/SBCO instructions 23286when the load/store address is known and can be addressed with some CTABLE 23287entry. For example: 23288 23289@smallexample 23290/* will compile to "sbco Rx, 2, 0x10, 4" */ 23291#pragma ctable_entry 2 0x4802a000 23292*(unsigned int *)0x4802a010 = val; 23293@end smallexample 23294 23295@end table 23296 23297@node RS/6000 and PowerPC Pragmas 23298@subsection RS/6000 and PowerPC Pragmas 23299 23300The RS/6000 and PowerPC targets define one pragma for controlling 23301whether or not the @code{longcall} attribute is added to function 23302declarations by default. This pragma overrides the @option{-mlongcall} 23303option, but not the @code{longcall} and @code{shortcall} attributes. 23304@xref{RS/6000 and PowerPC Options}, for more information about when long 23305calls are and are not necessary. 23306 23307@table @code 23308@item longcall (1) 23309@cindex pragma, longcall 23310Apply the @code{longcall} attribute to all subsequent function 23311declarations. 23312 23313@item longcall (0) 23314Do not apply the @code{longcall} attribute to subsequent function 23315declarations. 23316@end table 23317 23318@c Describe h8300 pragmas here. 23319@c Describe sh pragmas here. 23320@c Describe v850 pragmas here. 23321 23322@node S/390 Pragmas 23323@subsection S/390 Pragmas 23324 23325The pragmas defined by the S/390 target correspond to the S/390 23326target function attributes and some the additional options: 23327 23328@table @samp 23329@item zvector 23330@itemx no-zvector 23331@end table 23332 23333Note that options of the pragma, unlike options of the target 23334attribute, do change the value of preprocessor macros like 23335@code{__VEC__}. They can be specified as below: 23336 23337@smallexample 23338#pragma GCC target("string[,string]...") 23339#pragma GCC target("string"[,"string"]...) 23340@end smallexample 23341 23342@node Darwin Pragmas 23343@subsection Darwin Pragmas 23344 23345The following pragmas are available for all architectures running the 23346Darwin operating system. These are useful for compatibility with other 23347Mac OS compilers. 23348 23349@table @code 23350@item mark @var{tokens}@dots{} 23351@cindex pragma, mark 23352This pragma is accepted, but has no effect. 23353 23354@item options align=@var{alignment} 23355@cindex pragma, options align 23356This pragma sets the alignment of fields in structures. The values of 23357@var{alignment} may be @code{mac68k}, to emulate m68k alignment, or 23358@code{power}, to emulate PowerPC alignment. Uses of this pragma nest 23359properly; to restore the previous setting, use @code{reset} for the 23360@var{alignment}. 23361 23362@item segment @var{tokens}@dots{} 23363@cindex pragma, segment 23364This pragma is accepted, but has no effect. 23365 23366@item unused (@var{var} [, @var{var}]@dots{}) 23367@cindex pragma, unused 23368This pragma declares variables to be possibly unused. GCC does not 23369produce warnings for the listed variables. The effect is similar to 23370that of the @code{unused} attribute, except that this pragma may appear 23371anywhere within the variables' scopes. 23372@end table 23373 23374@node Solaris Pragmas 23375@subsection Solaris Pragmas 23376 23377The Solaris target supports @code{#pragma redefine_extname} 23378(@pxref{Symbol-Renaming Pragmas}). It also supports additional 23379@code{#pragma} directives for compatibility with the system compiler. 23380 23381@table @code 23382@item align @var{alignment} (@var{variable} [, @var{variable}]...) 23383@cindex pragma, align 23384 23385Increase the minimum alignment of each @var{variable} to @var{alignment}. 23386This is the same as GCC's @code{aligned} attribute @pxref{Variable 23387Attributes}). Macro expansion occurs on the arguments to this pragma 23388when compiling C and Objective-C@. It does not currently occur when 23389compiling C++, but this is a bug which may be fixed in a future 23390release. 23391 23392@item fini (@var{function} [, @var{function}]...) 23393@cindex pragma, fini 23394 23395This pragma causes each listed @var{function} to be called after 23396main, or during shared module unloading, by adding a call to the 23397@code{.fini} section. 23398 23399@item init (@var{function} [, @var{function}]...) 23400@cindex pragma, init 23401 23402This pragma causes each listed @var{function} to be called during 23403initialization (before @code{main}) or during shared module loading, by 23404adding a call to the @code{.init} section. 23405 23406@end table 23407 23408@node Symbol-Renaming Pragmas 23409@subsection Symbol-Renaming Pragmas 23410 23411GCC supports a @code{#pragma} directive that changes the name used in 23412assembly for a given declaration. While this pragma is supported on all 23413platforms, it is intended primarily to provide compatibility with the 23414Solaris system headers. This effect can also be achieved using the asm 23415labels extension (@pxref{Asm Labels}). 23416 23417@table @code 23418@item redefine_extname @var{oldname} @var{newname} 23419@cindex pragma, redefine_extname 23420 23421This pragma gives the C function @var{oldname} the assembly symbol 23422@var{newname}. The preprocessor macro @code{__PRAGMA_REDEFINE_EXTNAME} 23423is defined if this pragma is available (currently on all platforms). 23424@end table 23425 23426This pragma and the @code{asm} labels extension interact in a complicated 23427manner. Here are some corner cases you may want to be aware of: 23428 23429@enumerate 23430@item This pragma silently applies only to declarations with external 23431linkage. The @code{asm} label feature does not have this restriction. 23432 23433@item In C++, this pragma silently applies only to declarations with 23434``C'' linkage. Again, @code{asm} labels do not have this restriction. 23435 23436@item If either of the ways of changing the assembly name of a 23437declaration are applied to a declaration whose assembly name has 23438already been determined (either by a previous use of one of these 23439features, or because the compiler needed the assembly name in order to 23440generate code), and the new name is different, a warning issues and 23441the name does not change. 23442 23443@item The @var{oldname} used by @code{#pragma redefine_extname} is 23444always the C-language name. 23445@end enumerate 23446 23447@node Structure-Layout Pragmas 23448@subsection Structure-Layout Pragmas 23449 23450For compatibility with Microsoft Windows compilers, GCC supports a 23451set of @code{#pragma} directives that change the maximum alignment of 23452members of structures (other than zero-width bit-fields), unions, and 23453classes subsequently defined. The @var{n} value below always is required 23454to be a small power of two and specifies the new alignment in bytes. 23455 23456@enumerate 23457@item @code{#pragma pack(@var{n})} simply sets the new alignment. 23458@item @code{#pragma pack()} sets the alignment to the one that was in 23459effect when compilation started (see also command-line option 23460@option{-fpack-struct[=@var{n}]} @pxref{Code Gen Options}). 23461@item @code{#pragma pack(push[,@var{n}])} pushes the current alignment 23462setting on an internal stack and then optionally sets the new alignment. 23463@item @code{#pragma pack(pop)} restores the alignment setting to the one 23464saved at the top of the internal stack (and removes that stack entry). 23465Note that @code{#pragma pack([@var{n}])} does not influence this internal 23466stack; thus it is possible to have @code{#pragma pack(push)} followed by 23467multiple @code{#pragma pack(@var{n})} instances and finalized by a single 23468@code{#pragma pack(pop)}. 23469@end enumerate 23470 23471Some targets, e.g.@: x86 and PowerPC, support the @code{#pragma ms_struct} 23472directive which lays out structures and unions subsequently defined as the 23473documented @code{__attribute__ ((ms_struct))}. 23474 23475@enumerate 23476@item @code{#pragma ms_struct on} turns on the Microsoft layout. 23477@item @code{#pragma ms_struct off} turns off the Microsoft layout. 23478@item @code{#pragma ms_struct reset} goes back to the default layout. 23479@end enumerate 23480 23481Most targets also support the @code{#pragma scalar_storage_order} directive 23482which lays out structures and unions subsequently defined as the documented 23483@code{__attribute__ ((scalar_storage_order))}. 23484 23485@enumerate 23486@item @code{#pragma scalar_storage_order big-endian} sets the storage order 23487of the scalar fields to big-endian. 23488@item @code{#pragma scalar_storage_order little-endian} sets the storage order 23489of the scalar fields to little-endian. 23490@item @code{#pragma scalar_storage_order default} goes back to the endianness 23491that was in effect when compilation started (see also command-line option 23492@option{-fsso-struct=@var{endianness}} @pxref{C Dialect Options}). 23493@end enumerate 23494 23495@node Weak Pragmas 23496@subsection Weak Pragmas 23497 23498For compatibility with SVR4, GCC supports a set of @code{#pragma} 23499directives for declaring symbols to be weak, and defining weak 23500aliases. 23501 23502@table @code 23503@item #pragma weak @var{symbol} 23504@cindex pragma, weak 23505This pragma declares @var{symbol} to be weak, as if the declaration 23506had the attribute of the same name. The pragma may appear before 23507or after the declaration of @var{symbol}. It is not an error for 23508@var{symbol} to never be defined at all. 23509 23510@item #pragma weak @var{symbol1} = @var{symbol2} 23511This pragma declares @var{symbol1} to be a weak alias of @var{symbol2}. 23512It is an error if @var{symbol2} is not defined in the current 23513translation unit. 23514@end table 23515 23516@node Diagnostic Pragmas 23517@subsection Diagnostic Pragmas 23518 23519GCC allows the user to selectively enable or disable certain types of 23520diagnostics, and change the kind of the diagnostic. For example, a 23521project's policy might require that all sources compile with 23522@option{-Werror} but certain files might have exceptions allowing 23523specific types of warnings. Or, a project might selectively enable 23524diagnostics and treat them as errors depending on which preprocessor 23525macros are defined. 23526 23527@table @code 23528@item #pragma GCC diagnostic @var{kind} @var{option} 23529@cindex pragma, diagnostic 23530 23531Modifies the disposition of a diagnostic. Note that not all 23532diagnostics are modifiable; at the moment only warnings (normally 23533controlled by @samp{-W@dots{}}) can be controlled, and not all of them. 23534Use @option{-fdiagnostics-show-option} to determine which diagnostics 23535are controllable and which option controls them. 23536 23537@var{kind} is @samp{error} to treat this diagnostic as an error, 23538@samp{warning} to treat it like a warning (even if @option{-Werror} is 23539in effect), or @samp{ignored} if the diagnostic is to be ignored. 23540@var{option} is a double quoted string that matches the command-line 23541option. 23542 23543@smallexample 23544#pragma GCC diagnostic warning "-Wformat" 23545#pragma GCC diagnostic error "-Wformat" 23546#pragma GCC diagnostic ignored "-Wformat" 23547@end smallexample 23548 23549Note that these pragmas override any command-line options. GCC keeps 23550track of the location of each pragma, and issues diagnostics according 23551to the state as of that point in the source file. Thus, pragmas occurring 23552after a line do not affect diagnostics caused by that line. 23553 23554@item #pragma GCC diagnostic push 23555@itemx #pragma GCC diagnostic pop 23556 23557Causes GCC to remember the state of the diagnostics as of each 23558@code{push}, and restore to that point at each @code{pop}. If a 23559@code{pop} has no matching @code{push}, the command-line options are 23560restored. 23561 23562@smallexample 23563#pragma GCC diagnostic error "-Wuninitialized" 23564 foo(a); /* error is given for this one */ 23565#pragma GCC diagnostic push 23566#pragma GCC diagnostic ignored "-Wuninitialized" 23567 foo(b); /* no diagnostic for this one */ 23568#pragma GCC diagnostic pop 23569 foo(c); /* error is given for this one */ 23570#pragma GCC diagnostic pop 23571 foo(d); /* depends on command-line options */ 23572@end smallexample 23573 23574@end table 23575 23576GCC also offers a simple mechanism for printing messages during 23577compilation. 23578 23579@table @code 23580@item #pragma message @var{string} 23581@cindex pragma, diagnostic 23582 23583Prints @var{string} as a compiler message on compilation. The message 23584is informational only, and is neither a compilation warning nor an 23585error. Newlines can be included in the string by using the @samp{\n} 23586escape sequence. 23587 23588@smallexample 23589#pragma message "Compiling " __FILE__ "..." 23590@end smallexample 23591 23592@var{string} may be parenthesized, and is printed with location 23593information. For example, 23594 23595@smallexample 23596#define DO_PRAGMA(x) _Pragma (#x) 23597#define TODO(x) DO_PRAGMA(message ("TODO - " #x)) 23598 23599TODO(Remember to fix this) 23600@end smallexample 23601 23602@noindent 23603prints @samp{/tmp/file.c:4: note: #pragma message: 23604TODO - Remember to fix this}. 23605 23606@item #pragma GCC error @var{message} 23607@cindex pragma, diagnostic 23608Generates an error message. This pragma @emph{is} considered to 23609indicate an error in the compilation, and it will be treated as such. 23610 23611Newlines can be included in the string by using the @samp{\n} 23612escape sequence. They will be displayed as newlines even if the 23613@option{-fmessage-length} option is set to zero. 23614 23615The error is only generated if the pragma is present in the code after 23616pre-processing has been completed. It does not matter however if the 23617code containing the pragma is unreachable: 23618 23619@smallexample 23620#if 0 23621#pragma GCC error "this error is not seen" 23622#endif 23623void foo (void) 23624@{ 23625 return; 23626#pragma GCC error "this error is seen" 23627@} 23628@end smallexample 23629 23630@item #pragma GCC warning @var{message} 23631@cindex pragma, diagnostic 23632This is just like @samp{pragma GCC error} except that a warning 23633message is issued instead of an error message. Unless 23634@option{-Werror} is in effect, in which case this pragma will generate 23635an error as well. 23636 23637@end table 23638 23639@node Visibility Pragmas 23640@subsection Visibility Pragmas 23641 23642@table @code 23643@item #pragma GCC visibility push(@var{visibility}) 23644@itemx #pragma GCC visibility pop 23645@cindex pragma, visibility 23646 23647This pragma allows the user to set the visibility for multiple 23648declarations without having to give each a visibility attribute 23649(@pxref{Function Attributes}). 23650 23651In C++, @samp{#pragma GCC visibility} affects only namespace-scope 23652declarations. Class members and template specializations are not 23653affected; if you want to override the visibility for a particular 23654member or instantiation, you must use an attribute. 23655 23656@end table 23657 23658 23659@node Push/Pop Macro Pragmas 23660@subsection Push/Pop Macro Pragmas 23661 23662For compatibility with Microsoft Windows compilers, GCC supports 23663@samp{#pragma push_macro(@var{"macro_name"})} 23664and @samp{#pragma pop_macro(@var{"macro_name"})}. 23665 23666@table @code 23667@item #pragma push_macro(@var{"macro_name"}) 23668@cindex pragma, push_macro 23669This pragma saves the value of the macro named as @var{macro_name} to 23670the top of the stack for this macro. 23671 23672@item #pragma pop_macro(@var{"macro_name"}) 23673@cindex pragma, pop_macro 23674This pragma sets the value of the macro named as @var{macro_name} to 23675the value on top of the stack for this macro. If the stack for 23676@var{macro_name} is empty, the value of the macro remains unchanged. 23677@end table 23678 23679For example: 23680 23681@smallexample 23682#define X 1 23683#pragma push_macro("X") 23684#undef X 23685#define X -1 23686#pragma pop_macro("X") 23687int x [X]; 23688@end smallexample 23689 23690@noindent 23691In this example, the definition of X as 1 is saved by @code{#pragma 23692push_macro} and restored by @code{#pragma pop_macro}. 23693 23694@node Function Specific Option Pragmas 23695@subsection Function Specific Option Pragmas 23696 23697@table @code 23698@item #pragma GCC target (@var{string}, @dots{}) 23699@cindex pragma GCC target 23700 23701This pragma allows you to set target-specific options for functions 23702defined later in the source file. One or more strings can be 23703specified. Each function that is defined after this point is treated 23704as if it had been declared with one @code{target(}@var{string}@code{)} 23705attribute for each @var{string} argument. The parentheses around 23706the strings in the pragma are optional. @xref{Function Attributes}, 23707for more information about the @code{target} attribute and the attribute 23708syntax. 23709 23710The @code{#pragma GCC target} pragma is presently implemented for 23711x86, ARM, AArch64, PowerPC, S/390, and Nios II targets only. 23712 23713@item #pragma GCC optimize (@var{string}, @dots{}) 23714@cindex pragma GCC optimize 23715 23716This pragma allows you to set global optimization options for functions 23717defined later in the source file. One or more strings can be 23718specified. Each function that is defined after this point is treated 23719as if it had been declared with one @code{optimize(}@var{string}@code{)} 23720attribute for each @var{string} argument. The parentheses around 23721the strings in the pragma are optional. @xref{Function Attributes}, 23722for more information about the @code{optimize} attribute and the attribute 23723syntax. 23724 23725@item #pragma GCC push_options 23726@itemx #pragma GCC pop_options 23727@cindex pragma GCC push_options 23728@cindex pragma GCC pop_options 23729 23730These pragmas maintain a stack of the current target and optimization 23731options. It is intended for include files where you temporarily want 23732to switch to using a different @samp{#pragma GCC target} or 23733@samp{#pragma GCC optimize} and then to pop back to the previous 23734options. 23735 23736@item #pragma GCC reset_options 23737@cindex pragma GCC reset_options 23738 23739This pragma clears the current @code{#pragma GCC target} and 23740@code{#pragma GCC optimize} to use the default switches as specified 23741on the command line. 23742 23743@end table 23744 23745@node Loop-Specific Pragmas 23746@subsection Loop-Specific Pragmas 23747 23748@table @code 23749@item #pragma GCC ivdep 23750@cindex pragma GCC ivdep 23751 23752With this pragma, the programmer asserts that there are no loop-carried 23753dependencies which would prevent consecutive iterations of 23754the following loop from executing concurrently with SIMD 23755(single instruction multiple data) instructions. 23756 23757For example, the compiler can only unconditionally vectorize the following 23758loop with the pragma: 23759 23760@smallexample 23761void foo (int n, int *a, int *b, int *c) 23762@{ 23763 int i, j; 23764#pragma GCC ivdep 23765 for (i = 0; i < n; ++i) 23766 a[i] = b[i] + c[i]; 23767@} 23768@end smallexample 23769 23770@noindent 23771In this example, using the @code{restrict} qualifier had the same 23772effect. In the following example, that would not be possible. Assume 23773@math{k < -m} or @math{k >= m}. Only with the pragma, the compiler knows 23774that it can unconditionally vectorize the following loop: 23775 23776@smallexample 23777void ignore_vec_dep (int *a, int k, int c, int m) 23778@{ 23779#pragma GCC ivdep 23780 for (int i = 0; i < m; i++) 23781 a[i] = a[i + k] * c; 23782@} 23783@end smallexample 23784 23785@item #pragma GCC unroll @var{n} 23786@cindex pragma GCC unroll @var{n} 23787 23788You can use this pragma to control how many times a loop should be unrolled. 23789It must be placed immediately before a @code{for}, @code{while} or @code{do} 23790loop or a @code{#pragma GCC ivdep}, and applies only to the loop that follows. 23791@var{n} is an integer constant expression specifying the unrolling factor. 23792The values of @math{0} and @math{1} block any unrolling of the loop. 23793 23794@end table 23795 23796@node Unnamed Fields 23797@section Unnamed Structure and Union Fields 23798@cindex @code{struct} 23799@cindex @code{union} 23800 23801As permitted by ISO C11 and for compatibility with other compilers, 23802GCC allows you to define 23803a structure or union that contains, as fields, structures and unions 23804without names. For example: 23805 23806@smallexample 23807struct @{ 23808 int a; 23809 union @{ 23810 int b; 23811 float c; 23812 @}; 23813 int d; 23814@} foo; 23815@end smallexample 23816 23817@noindent 23818In this example, you are able to access members of the unnamed 23819union with code like @samp{foo.b}. Note that only unnamed structs and 23820unions are allowed, you may not have, for example, an unnamed 23821@code{int}. 23822 23823You must never create such structures that cause ambiguous field definitions. 23824For example, in this structure: 23825 23826@smallexample 23827struct @{ 23828 int a; 23829 struct @{ 23830 int a; 23831 @}; 23832@} foo; 23833@end smallexample 23834 23835@noindent 23836it is ambiguous which @code{a} is being referred to with @samp{foo.a}. 23837The compiler gives errors for such constructs. 23838 23839@opindex fms-extensions 23840Unless @option{-fms-extensions} is used, the unnamed field must be a 23841structure or union definition without a tag (for example, @samp{struct 23842@{ int a; @};}). If @option{-fms-extensions} is used, the field may 23843also be a definition with a tag such as @samp{struct foo @{ int a; 23844@};}, a reference to a previously defined structure or union such as 23845@samp{struct foo;}, or a reference to a @code{typedef} name for a 23846previously defined structure or union type. 23847 23848@opindex fplan9-extensions 23849The option @option{-fplan9-extensions} enables 23850@option{-fms-extensions} as well as two other extensions. First, a 23851pointer to a structure is automatically converted to a pointer to an 23852anonymous field for assignments and function calls. For example: 23853 23854@smallexample 23855struct s1 @{ int a; @}; 23856struct s2 @{ struct s1; @}; 23857extern void f1 (struct s1 *); 23858void f2 (struct s2 *p) @{ f1 (p); @} 23859@end smallexample 23860 23861@noindent 23862In the call to @code{f1} inside @code{f2}, the pointer @code{p} is 23863converted into a pointer to the anonymous field. 23864 23865Second, when the type of an anonymous field is a @code{typedef} for a 23866@code{struct} or @code{union}, code may refer to the field using the 23867name of the @code{typedef}. 23868 23869@smallexample 23870typedef struct @{ int a; @} s1; 23871struct s2 @{ s1; @}; 23872s1 f1 (struct s2 *p) @{ return p->s1; @} 23873@end smallexample 23874 23875These usages are only permitted when they are not ambiguous. 23876 23877@node Thread-Local 23878@section Thread-Local Storage 23879@cindex Thread-Local Storage 23880@cindex @acronym{TLS} 23881@cindex @code{__thread} 23882 23883Thread-local storage (@acronym{TLS}) is a mechanism by which variables 23884are allocated such that there is one instance of the variable per extant 23885thread. The runtime model GCC uses to implement this originates 23886in the IA-64 processor-specific ABI, but has since been migrated 23887to other processors as well. It requires significant support from 23888the linker (@command{ld}), dynamic linker (@command{ld.so}), and 23889system libraries (@file{libc.so} and @file{libpthread.so}), so it 23890is not available everywhere. 23891 23892At the user level, the extension is visible with a new storage 23893class keyword: @code{__thread}. For example: 23894 23895@smallexample 23896__thread int i; 23897extern __thread struct state s; 23898static __thread char *p; 23899@end smallexample 23900 23901The @code{__thread} specifier may be used alone, with the @code{extern} 23902or @code{static} specifiers, but with no other storage class specifier. 23903When used with @code{extern} or @code{static}, @code{__thread} must appear 23904immediately after the other storage class specifier. 23905 23906The @code{__thread} specifier may be applied to any global, file-scoped 23907static, function-scoped static, or static data member of a class. It may 23908not be applied to block-scoped automatic or non-static data member. 23909 23910When the address-of operator is applied to a thread-local variable, it is 23911evaluated at run time and returns the address of the current thread's 23912instance of that variable. An address so obtained may be used by any 23913thread. When a thread terminates, any pointers to thread-local variables 23914in that thread become invalid. 23915 23916No static initialization may refer to the address of a thread-local variable. 23917 23918In C++, if an initializer is present for a thread-local variable, it must 23919be a @var{constant-expression}, as defined in 5.19.2 of the ANSI/ISO C++ 23920standard. 23921 23922See @uref{https://www.akkadia.org/drepper/tls.pdf, 23923ELF Handling For Thread-Local Storage} for a detailed explanation of 23924the four thread-local storage addressing models, and how the runtime 23925is expected to function. 23926 23927@menu 23928* C99 Thread-Local Edits:: 23929* C++98 Thread-Local Edits:: 23930@end menu 23931 23932@node C99 Thread-Local Edits 23933@subsection ISO/IEC 9899:1999 Edits for Thread-Local Storage 23934 23935The following are a set of changes to ISO/IEC 9899:1999 (aka C99) 23936that document the exact semantics of the language extension. 23937 23938@itemize @bullet 23939@item 23940@cite{5.1.2 Execution environments} 23941 23942Add new text after paragraph 1 23943 23944@quotation 23945Within either execution environment, a @dfn{thread} is a flow of 23946control within a program. It is implementation defined whether 23947or not there may be more than one thread associated with a program. 23948It is implementation defined how threads beyond the first are 23949created, the name and type of the function called at thread 23950startup, and how threads may be terminated. However, objects 23951with thread storage duration shall be initialized before thread 23952startup. 23953@end quotation 23954 23955@item 23956@cite{6.2.4 Storage durations of objects} 23957 23958Add new text before paragraph 3 23959 23960@quotation 23961An object whose identifier is declared with the storage-class 23962specifier @w{@code{__thread}} has @dfn{thread storage duration}. 23963Its lifetime is the entire execution of the thread, and its 23964stored value is initialized only once, prior to thread startup. 23965@end quotation 23966 23967@item 23968@cite{6.4.1 Keywords} 23969 23970Add @code{__thread}. 23971 23972@item 23973@cite{6.7.1 Storage-class specifiers} 23974 23975Add @code{__thread} to the list of storage class specifiers in 23976paragraph 1. 23977 23978Change paragraph 2 to 23979 23980@quotation 23981With the exception of @code{__thread}, at most one storage-class 23982specifier may be given [@dots{}]. The @code{__thread} specifier may 23983be used alone, or immediately following @code{extern} or 23984@code{static}. 23985@end quotation 23986 23987Add new text after paragraph 6 23988 23989@quotation 23990The declaration of an identifier for a variable that has 23991block scope that specifies @code{__thread} shall also 23992specify either @code{extern} or @code{static}. 23993 23994The @code{__thread} specifier shall be used only with 23995variables. 23996@end quotation 23997@end itemize 23998 23999@node C++98 Thread-Local Edits 24000@subsection ISO/IEC 14882:1998 Edits for Thread-Local Storage 24001 24002The following are a set of changes to ISO/IEC 14882:1998 (aka C++98) 24003that document the exact semantics of the language extension. 24004 24005@itemize @bullet 24006@item 24007@b{[intro.execution]} 24008 24009New text after paragraph 4 24010 24011@quotation 24012A @dfn{thread} is a flow of control within the abstract machine. 24013It is implementation defined whether or not there may be more than 24014one thread. 24015@end quotation 24016 24017New text after paragraph 7 24018 24019@quotation 24020It is unspecified whether additional action must be taken to 24021ensure when and whether side effects are visible to other threads. 24022@end quotation 24023 24024@item 24025@b{[lex.key]} 24026 24027Add @code{__thread}. 24028 24029@item 24030@b{[basic.start.main]} 24031 24032Add after paragraph 5 24033 24034@quotation 24035The thread that begins execution at the @code{main} function is called 24036the @dfn{main thread}. It is implementation defined how functions 24037beginning threads other than the main thread are designated or typed. 24038A function so designated, as well as the @code{main} function, is called 24039a @dfn{thread startup function}. It is implementation defined what 24040happens if a thread startup function returns. It is implementation 24041defined what happens to other threads when any thread calls @code{exit}. 24042@end quotation 24043 24044@item 24045@b{[basic.start.init]} 24046 24047Add after paragraph 4 24048 24049@quotation 24050The storage for an object of thread storage duration shall be 24051statically initialized before the first statement of the thread startup 24052function. An object of thread storage duration shall not require 24053dynamic initialization. 24054@end quotation 24055 24056@item 24057@b{[basic.start.term]} 24058 24059Add after paragraph 3 24060 24061@quotation 24062The type of an object with thread storage duration shall not have a 24063non-trivial destructor, nor shall it be an array type whose elements 24064(directly or indirectly) have non-trivial destructors. 24065@end quotation 24066 24067@item 24068@b{[basic.stc]} 24069 24070Add ``thread storage duration'' to the list in paragraph 1. 24071 24072Change paragraph 2 24073 24074@quotation 24075Thread, static, and automatic storage durations are associated with 24076objects introduced by declarations [@dots{}]. 24077@end quotation 24078 24079Add @code{__thread} to the list of specifiers in paragraph 3. 24080 24081@item 24082@b{[basic.stc.thread]} 24083 24084New section before @b{[basic.stc.static]} 24085 24086@quotation 24087The keyword @code{__thread} applied to a non-local object gives the 24088object thread storage duration. 24089 24090A local variable or class data member declared both @code{static} 24091and @code{__thread} gives the variable or member thread storage 24092duration. 24093@end quotation 24094 24095@item 24096@b{[basic.stc.static]} 24097 24098Change paragraph 1 24099 24100@quotation 24101All objects that have neither thread storage duration, dynamic 24102storage duration nor are local [@dots{}]. 24103@end quotation 24104 24105@item 24106@b{[dcl.stc]} 24107 24108Add @code{__thread} to the list in paragraph 1. 24109 24110Change paragraph 1 24111 24112@quotation 24113With the exception of @code{__thread}, at most one 24114@var{storage-class-specifier} shall appear in a given 24115@var{decl-specifier-seq}. The @code{__thread} specifier may 24116be used alone, or immediately following the @code{extern} or 24117@code{static} specifiers. [@dots{}] 24118@end quotation 24119 24120Add after paragraph 5 24121 24122@quotation 24123The @code{__thread} specifier can be applied only to the names of objects 24124and to anonymous unions. 24125@end quotation 24126 24127@item 24128@b{[class.mem]} 24129 24130Add after paragraph 6 24131 24132@quotation 24133Non-@code{static} members shall not be @code{__thread}. 24134@end quotation 24135@end itemize 24136 24137@node Binary constants 24138@section Binary Constants using the @samp{0b} Prefix 24139@cindex Binary constants using the @samp{0b} prefix 24140 24141Integer constants can be written as binary constants, consisting of a 24142sequence of @samp{0} and @samp{1} digits, prefixed by @samp{0b} or 24143@samp{0B}. This is particularly useful in environments that operate a 24144lot on the bit level (like microcontrollers). 24145 24146The following statements are identical: 24147 24148@smallexample 24149i = 42; 24150i = 0x2a; 24151i = 052; 24152i = 0b101010; 24153@end smallexample 24154 24155The type of these constants follows the same rules as for octal or 24156hexadecimal integer constants, so suffixes like @samp{L} or @samp{UL} 24157can be applied. 24158 24159@node C++ Extensions 24160@chapter Extensions to the C++ Language 24161@cindex extensions, C++ language 24162@cindex C++ language extensions 24163 24164The GNU compiler provides these extensions to the C++ language (and you 24165can also use most of the C language extensions in your C++ programs). If you 24166want to write code that checks whether these features are available, you can 24167test for the GNU compiler the same way as for C programs: check for a 24168predefined macro @code{__GNUC__}. You can also use @code{__GNUG__} to 24169test specifically for GNU C++ (@pxref{Common Predefined Macros,, 24170Predefined Macros,cpp,The GNU C Preprocessor}). 24171 24172@menu 24173* C++ Volatiles:: What constitutes an access to a volatile object. 24174* Restricted Pointers:: C99 restricted pointers and references. 24175* Vague Linkage:: Where G++ puts inlines, vtables and such. 24176* C++ Interface:: You can use a single C++ header file for both 24177 declarations and definitions. 24178* Template Instantiation:: Methods for ensuring that exactly one copy of 24179 each needed template instantiation is emitted. 24180* Bound member functions:: You can extract a function pointer to the 24181 method denoted by a @samp{->*} or @samp{.*} expression. 24182* C++ Attributes:: Variable, function, and type attributes for C++ only. 24183* Function Multiversioning:: Declaring multiple function versions. 24184* Type Traits:: Compiler support for type traits. 24185* C++ Concepts:: Improved support for generic programming. 24186* Deprecated Features:: Things will disappear from G++. 24187* Backwards Compatibility:: Compatibilities with earlier definitions of C++. 24188@end menu 24189 24190@node C++ Volatiles 24191@section When is a Volatile C++ Object Accessed? 24192@cindex accessing volatiles 24193@cindex volatile read 24194@cindex volatile write 24195@cindex volatile access 24196 24197The C++ standard differs from the C standard in its treatment of 24198volatile objects. It fails to specify what constitutes a volatile 24199access, except to say that C++ should behave in a similar manner to C 24200with respect to volatiles, where possible. However, the different 24201lvalueness of expressions between C and C++ complicate the behavior. 24202G++ behaves the same as GCC for volatile access, @xref{C 24203Extensions,,Volatiles}, for a description of GCC's behavior. 24204 24205The C and C++ language specifications differ when an object is 24206accessed in a void context: 24207 24208@smallexample 24209volatile int *src = @var{somevalue}; 24210*src; 24211@end smallexample 24212 24213The C++ standard specifies that such expressions do not undergo lvalue 24214to rvalue conversion, and that the type of the dereferenced object may 24215be incomplete. The C++ standard does not specify explicitly that it 24216is lvalue to rvalue conversion that is responsible for causing an 24217access. There is reason to believe that it is, because otherwise 24218certain simple expressions become undefined. However, because it 24219would surprise most programmers, G++ treats dereferencing a pointer to 24220volatile object of complete type as GCC would do for an equivalent 24221type in C@. When the object has incomplete type, G++ issues a 24222warning; if you wish to force an error, you must force a conversion to 24223rvalue with, for instance, a static cast. 24224 24225When using a reference to volatile, G++ does not treat equivalent 24226expressions as accesses to volatiles, but instead issues a warning that 24227no volatile is accessed. The rationale for this is that otherwise it 24228becomes difficult to determine where volatile access occur, and not 24229possible to ignore the return value from functions returning volatile 24230references. Again, if you wish to force a read, cast the reference to 24231an rvalue. 24232 24233G++ implements the same behavior as GCC does when assigning to a 24234volatile object---there is no reread of the assigned-to object, the 24235assigned rvalue is reused. Note that in C++ assignment expressions 24236are lvalues, and if used as an lvalue, the volatile object is 24237referred to. For instance, @var{vref} refers to @var{vobj}, as 24238expected, in the following example: 24239 24240@smallexample 24241volatile int vobj; 24242volatile int &vref = vobj = @var{something}; 24243@end smallexample 24244 24245@node Restricted Pointers 24246@section Restricting Pointer Aliasing 24247@cindex restricted pointers 24248@cindex restricted references 24249@cindex restricted this pointer 24250 24251As with the C front end, G++ understands the C99 feature of restricted pointers, 24252specified with the @code{__restrict__}, or @code{__restrict} type 24253qualifier. Because you cannot compile C++ by specifying the @option{-std=c99} 24254language flag, @code{restrict} is not a keyword in C++. 24255 24256In addition to allowing restricted pointers, you can specify restricted 24257references, which indicate that the reference is not aliased in the local 24258context. 24259 24260@smallexample 24261void fn (int *__restrict__ rptr, int &__restrict__ rref) 24262@{ 24263 /* @r{@dots{}} */ 24264@} 24265@end smallexample 24266 24267@noindent 24268In the body of @code{fn}, @var{rptr} points to an unaliased integer and 24269@var{rref} refers to a (different) unaliased integer. 24270 24271You may also specify whether a member function's @var{this} pointer is 24272unaliased by using @code{__restrict__} as a member function qualifier. 24273 24274@smallexample 24275void T::fn () __restrict__ 24276@{ 24277 /* @r{@dots{}} */ 24278@} 24279@end smallexample 24280 24281@noindent 24282Within the body of @code{T::fn}, @var{this} has the effective 24283definition @code{T *__restrict__ const this}. Notice that the 24284interpretation of a @code{__restrict__} member function qualifier is 24285different to that of @code{const} or @code{volatile} qualifier, in that it 24286is applied to the pointer rather than the object. This is consistent with 24287other compilers that implement restricted pointers. 24288 24289As with all outermost parameter qualifiers, @code{__restrict__} is 24290ignored in function definition matching. This means you only need to 24291specify @code{__restrict__} in a function definition, rather than 24292in a function prototype as well. 24293 24294@node Vague Linkage 24295@section Vague Linkage 24296@cindex vague linkage 24297 24298There are several constructs in C++ that require space in the object 24299file but are not clearly tied to a single translation unit. We say that 24300these constructs have ``vague linkage''. Typically such constructs are 24301emitted wherever they are needed, though sometimes we can be more 24302clever. 24303 24304@table @asis 24305@item Inline Functions 24306Inline functions are typically defined in a header file which can be 24307included in many different compilations. Hopefully they can usually be 24308inlined, but sometimes an out-of-line copy is necessary, if the address 24309of the function is taken or if inlining fails. In general, we emit an 24310out-of-line copy in all translation units where one is needed. As an 24311exception, we only emit inline virtual functions with the vtable, since 24312it always requires a copy. 24313 24314Local static variables and string constants used in an inline function 24315are also considered to have vague linkage, since they must be shared 24316between all inlined and out-of-line instances of the function. 24317 24318@item VTables 24319@cindex vtable 24320C++ virtual functions are implemented in most compilers using a lookup 24321table, known as a vtable. The vtable contains pointers to the virtual 24322functions provided by a class, and each object of the class contains a 24323pointer to its vtable (or vtables, in some multiple-inheritance 24324situations). If the class declares any non-inline, non-pure virtual 24325functions, the first one is chosen as the ``key method'' for the class, 24326and the vtable is only emitted in the translation unit where the key 24327method is defined. 24328 24329@emph{Note:} If the chosen key method is later defined as inline, the 24330vtable is still emitted in every translation unit that defines it. 24331Make sure that any inline virtuals are declared inline in the class 24332body, even if they are not defined there. 24333 24334@item @code{type_info} objects 24335@cindex @code{type_info} 24336@cindex RTTI 24337C++ requires information about types to be written out in order to 24338implement @samp{dynamic_cast}, @samp{typeid} and exception handling. 24339For polymorphic classes (classes with virtual functions), the @samp{type_info} 24340object is written out along with the vtable so that @samp{dynamic_cast} 24341can determine the dynamic type of a class object at run time. For all 24342other types, we write out the @samp{type_info} object when it is used: when 24343applying @samp{typeid} to an expression, throwing an object, or 24344referring to a type in a catch clause or exception specification. 24345 24346@item Template Instantiations 24347Most everything in this section also applies to template instantiations, 24348but there are other options as well. 24349@xref{Template Instantiation,,Where's the Template?}. 24350 24351@end table 24352 24353When used with GNU ld version 2.8 or later on an ELF system such as 24354GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of 24355these constructs will be discarded at link time. This is known as 24356COMDAT support. 24357 24358On targets that don't support COMDAT, but do support weak symbols, GCC 24359uses them. This way one copy overrides all the others, but 24360the unused copies still take up space in the executable. 24361 24362For targets that do not support either COMDAT or weak symbols, 24363most entities with vague linkage are emitted as local symbols to 24364avoid duplicate definition errors from the linker. This does not happen 24365for local statics in inlines, however, as having multiple copies 24366almost certainly breaks things. 24367 24368@xref{C++ Interface,,Declarations and Definitions in One Header}, for 24369another way to control placement of these constructs. 24370 24371@node C++ Interface 24372@section C++ Interface and Implementation Pragmas 24373 24374@cindex interface and implementation headers, C++ 24375@cindex C++ interface and implementation headers 24376@cindex pragmas, interface and implementation 24377 24378@code{#pragma interface} and @code{#pragma implementation} provide the 24379user with a way of explicitly directing the compiler to emit entities 24380with vague linkage (and debugging information) in a particular 24381translation unit. 24382 24383@emph{Note:} These @code{#pragma}s have been superceded as of GCC 2.7.2 24384by COMDAT support and the ``key method'' heuristic 24385mentioned in @ref{Vague Linkage}. Using them can actually cause your 24386program to grow due to unnecessary out-of-line copies of inline 24387functions. 24388 24389@table @code 24390@item #pragma interface 24391@itemx #pragma interface "@var{subdir}/@var{objects}.h" 24392@kindex #pragma interface 24393Use this directive in @emph{header files} that define object classes, to save 24394space in most of the object files that use those classes. Normally, 24395local copies of certain information (backup copies of inline member 24396functions, debugging information, and the internal tables that implement 24397virtual functions) must be kept in each object file that includes class 24398definitions. You can use this pragma to avoid such duplication. When a 24399header file containing @samp{#pragma interface} is included in a 24400compilation, this auxiliary information is not generated (unless 24401the main input source file itself uses @samp{#pragma implementation}). 24402Instead, the object files contain references to be resolved at link 24403time. 24404 24405The second form of this directive is useful for the case where you have 24406multiple headers with the same name in different directories. If you 24407use this form, you must specify the same string to @samp{#pragma 24408implementation}. 24409 24410@item #pragma implementation 24411@itemx #pragma implementation "@var{objects}.h" 24412@kindex #pragma implementation 24413Use this pragma in a @emph{main input file}, when you want full output from 24414included header files to be generated (and made globally visible). The 24415included header file, in turn, should use @samp{#pragma interface}. 24416Backup copies of inline member functions, debugging information, and the 24417internal tables used to implement virtual functions are all generated in 24418implementation files. 24419 24420@cindex implied @code{#pragma implementation} 24421@cindex @code{#pragma implementation}, implied 24422@cindex naming convention, implementation headers 24423If you use @samp{#pragma implementation} with no argument, it applies to 24424an include file with the same basename@footnote{A file's @dfn{basename} 24425is the name stripped of all leading path information and of trailing 24426suffixes, such as @samp{.h} or @samp{.C} or @samp{.cc}.} as your source 24427file. For example, in @file{allclass.cc}, giving just 24428@samp{#pragma implementation} 24429by itself is equivalent to @samp{#pragma implementation "allclass.h"}. 24430 24431Use the string argument if you want a single implementation file to 24432include code from multiple header files. (You must also use 24433@samp{#include} to include the header file; @samp{#pragma 24434implementation} only specifies how to use the file---it doesn't actually 24435include it.) 24436 24437There is no way to split up the contents of a single header file into 24438multiple implementation files. 24439@end table 24440 24441@cindex inlining and C++ pragmas 24442@cindex C++ pragmas, effect on inlining 24443@cindex pragmas in C++, effect on inlining 24444@samp{#pragma implementation} and @samp{#pragma interface} also have an 24445effect on function inlining. 24446 24447If you define a class in a header file marked with @samp{#pragma 24448interface}, the effect on an inline function defined in that class is 24449similar to an explicit @code{extern} declaration---the compiler emits 24450no code at all to define an independent version of the function. Its 24451definition is used only for inlining with its callers. 24452 24453@opindex fno-implement-inlines 24454Conversely, when you include the same header file in a main source file 24455that declares it as @samp{#pragma implementation}, the compiler emits 24456code for the function itself; this defines a version of the function 24457that can be found via pointers (or by callers compiled without 24458inlining). If all calls to the function can be inlined, you can avoid 24459emitting the function by compiling with @option{-fno-implement-inlines}. 24460If any calls are not inlined, you will get linker errors. 24461 24462@node Template Instantiation 24463@section Where's the Template? 24464@cindex template instantiation 24465 24466C++ templates were the first language feature to require more 24467intelligence from the environment than was traditionally found on a UNIX 24468system. Somehow the compiler and linker have to make sure that each 24469template instance occurs exactly once in the executable if it is needed, 24470and not at all otherwise. There are two basic approaches to this 24471problem, which are referred to as the Borland model and the Cfront model. 24472 24473@table @asis 24474@item Borland model 24475Borland C++ solved the template instantiation problem by adding the code 24476equivalent of common blocks to their linker; the compiler emits template 24477instances in each translation unit that uses them, and the linker 24478collapses them together. The advantage of this model is that the linker 24479only has to consider the object files themselves; there is no external 24480complexity to worry about. The disadvantage is that compilation time 24481is increased because the template code is being compiled repeatedly. 24482Code written for this model tends to include definitions of all 24483templates in the header file, since they must be seen to be 24484instantiated. 24485 24486@item Cfront model 24487The AT&T C++ translator, Cfront, solved the template instantiation 24488problem by creating the notion of a template repository, an 24489automatically maintained place where template instances are stored. A 24490more modern version of the repository works as follows: As individual 24491object files are built, the compiler places any template definitions and 24492instantiations encountered in the repository. At link time, the link 24493wrapper adds in the objects in the repository and compiles any needed 24494instances that were not previously emitted. The advantages of this 24495model are more optimal compilation speed and the ability to use the 24496system linker; to implement the Borland model a compiler vendor also 24497needs to replace the linker. The disadvantages are vastly increased 24498complexity, and thus potential for error; for some code this can be 24499just as transparent, but in practice it can been very difficult to build 24500multiple programs in one directory and one program in multiple 24501directories. Code written for this model tends to separate definitions 24502of non-inline member templates into a separate file, which should be 24503compiled separately. 24504@end table 24505 24506G++ implements the Borland model on targets where the linker supports it, 24507including ELF targets (such as GNU/Linux), Mac OS X and Microsoft Windows. 24508Otherwise G++ implements neither automatic model. 24509 24510You have the following options for dealing with template instantiations: 24511 24512@enumerate 24513@item 24514Do nothing. Code written for the Borland model works fine, but 24515each translation unit contains instances of each of the templates it 24516uses. The duplicate instances will be discarded by the linker, but in 24517a large program, this can lead to an unacceptable amount of code 24518duplication in object files or shared libraries. 24519 24520Duplicate instances of a template can be avoided by defining an explicit 24521instantiation in one object file, and preventing the compiler from doing 24522implicit instantiations in any other object files by using an explicit 24523instantiation declaration, using the @code{extern template} syntax: 24524 24525@smallexample 24526extern template int max (int, int); 24527@end smallexample 24528 24529This syntax is defined in the C++ 2011 standard, but has been supported by 24530G++ and other compilers since well before 2011. 24531 24532Explicit instantiations can be used for the largest or most frequently 24533duplicated instances, without having to know exactly which other instances 24534are used in the rest of the program. You can scatter the explicit 24535instantiations throughout your program, perhaps putting them in the 24536translation units where the instances are used or the translation units 24537that define the templates themselves; you can put all of the explicit 24538instantiations you need into one big file; or you can create small files 24539like 24540 24541@smallexample 24542#include "Foo.h" 24543#include "Foo.cc" 24544 24545template class Foo<int>; 24546template ostream& operator << 24547 (ostream&, const Foo<int>&); 24548@end smallexample 24549 24550@noindent 24551for each of the instances you need, and create a template instantiation 24552library from those. 24553 24554This is the simplest option, but also offers flexibility and 24555fine-grained control when necessary. It is also the most portable 24556alternative and programs using this approach will work with most modern 24557compilers. 24558 24559@item 24560@opindex fno-implicit-templates 24561Compile your code with @option{-fno-implicit-templates} to disable the 24562implicit generation of template instances, and explicitly instantiate 24563all the ones you use. This approach requires more knowledge of exactly 24564which instances you need than do the others, but it's less 24565mysterious and allows greater control if you want to ensure that only 24566the intended instances are used. 24567 24568If you are using Cfront-model code, you can probably get away with not 24569using @option{-fno-implicit-templates} when compiling files that don't 24570@samp{#include} the member template definitions. 24571 24572If you use one big file to do the instantiations, you may want to 24573compile it without @option{-fno-implicit-templates} so you get all of the 24574instances required by your explicit instantiations (but not by any 24575other files) without having to specify them as well. 24576 24577In addition to forward declaration of explicit instantiations 24578(with @code{extern}), G++ has extended the template instantiation 24579syntax to support instantiation of the compiler support data for a 24580template class (i.e.@: the vtable) without instantiating any of its 24581members (with @code{inline}), and instantiation of only the static data 24582members of a template class, without the support data or member 24583functions (with @code{static}): 24584 24585@smallexample 24586inline template class Foo<int>; 24587static template class Foo<int>; 24588@end smallexample 24589@end enumerate 24590 24591@node Bound member functions 24592@section Extracting the Function Pointer from a Bound Pointer to Member Function 24593@cindex pmf 24594@cindex pointer to member function 24595@cindex bound pointer to member function 24596 24597In C++, pointer to member functions (PMFs) are implemented using a wide 24598pointer of sorts to handle all the possible call mechanisms; the PMF 24599needs to store information about how to adjust the @samp{this} pointer, 24600and if the function pointed to is virtual, where to find the vtable, and 24601where in the vtable to look for the member function. If you are using 24602PMFs in an inner loop, you should really reconsider that decision. If 24603that is not an option, you can extract the pointer to the function that 24604would be called for a given object/PMF pair and call it directly inside 24605the inner loop, to save a bit of time. 24606 24607Note that you still pay the penalty for the call through a 24608function pointer; on most modern architectures, such a call defeats the 24609branch prediction features of the CPU@. This is also true of normal 24610virtual function calls. 24611 24612The syntax for this extension is 24613 24614@smallexample 24615extern A a; 24616extern int (A::*fp)(); 24617typedef int (*fptr)(A *); 24618 24619fptr p = (fptr)(a.*fp); 24620@end smallexample 24621 24622For PMF constants (i.e.@: expressions of the form @samp{&Klasse::Member}), 24623no object is needed to obtain the address of the function. They can be 24624converted to function pointers directly: 24625 24626@smallexample 24627fptr p1 = (fptr)(&A::foo); 24628@end smallexample 24629 24630@opindex Wno-pmf-conversions 24631You must specify @option{-Wno-pmf-conversions} to use this extension. 24632 24633@node C++ Attributes 24634@section C++-Specific Variable, Function, and Type Attributes 24635 24636Some attributes only make sense for C++ programs. 24637 24638@table @code 24639@item abi_tag ("@var{tag}", ...) 24640@cindex @code{abi_tag} function attribute 24641@cindex @code{abi_tag} variable attribute 24642@cindex @code{abi_tag} type attribute 24643The @code{abi_tag} attribute can be applied to a function, variable, or class 24644declaration. It modifies the mangled name of the entity to 24645incorporate the tag name, in order to distinguish the function or 24646class from an earlier version with a different ABI; perhaps the class 24647has changed size, or the function has a different return type that is 24648not encoded in the mangled name. 24649 24650The attribute can also be applied to an inline namespace, but does not 24651affect the mangled name of the namespace; in this case it is only used 24652for @option{-Wabi-tag} warnings and automatic tagging of functions and 24653variables. Tagging inline namespaces is generally preferable to 24654tagging individual declarations, but the latter is sometimes 24655necessary, such as when only certain members of a class need to be 24656tagged. 24657 24658The argument can be a list of strings of arbitrary length. The 24659strings are sorted on output, so the order of the list is 24660unimportant. 24661 24662A redeclaration of an entity must not add new ABI tags, 24663since doing so would change the mangled name. 24664 24665The ABI tags apply to a name, so all instantiations and 24666specializations of a template have the same tags. The attribute will 24667be ignored if applied to an explicit specialization or instantiation. 24668 24669The @option{-Wabi-tag} flag enables a warning about a class which does 24670not have all the ABI tags used by its subobjects and virtual functions; for users with code 24671that needs to coexist with an earlier ABI, using this option can help 24672to find all affected types that need to be tagged. 24673 24674When a type involving an ABI tag is used as the type of a variable or 24675return type of a function where that tag is not already present in the 24676signature of the function, the tag is automatically applied to the 24677variable or function. @option{-Wabi-tag} also warns about this 24678situation; this warning can be avoided by explicitly tagging the 24679variable or function or moving it into a tagged inline namespace. 24680 24681@item init_priority (@var{priority}) 24682@cindex @code{init_priority} variable attribute 24683 24684In Standard C++, objects defined at namespace scope are guaranteed to be 24685initialized in an order in strict accordance with that of their definitions 24686@emph{in a given translation unit}. No guarantee is made for initializations 24687across translation units. However, GNU C++ allows users to control the 24688order of initialization of objects defined at namespace scope with the 24689@code{init_priority} attribute by specifying a relative @var{priority}, 24690a constant integral expression currently bounded between 101 and 65535 24691inclusive. Lower numbers indicate a higher priority. 24692 24693In the following example, @code{A} would normally be created before 24694@code{B}, but the @code{init_priority} attribute reverses that order: 24695 24696@smallexample 24697Some_Class A __attribute__ ((init_priority (2000))); 24698Some_Class B __attribute__ ((init_priority (543))); 24699@end smallexample 24700 24701@noindent 24702Note that the particular values of @var{priority} do not matter; only their 24703relative ordering. 24704 24705@item warn_unused 24706@cindex @code{warn_unused} type attribute 24707 24708For C++ types with non-trivial constructors and/or destructors it is 24709impossible for the compiler to determine whether a variable of this 24710type is truly unused if it is not referenced. This type attribute 24711informs the compiler that variables of this type should be warned 24712about if they appear to be unused, just like variables of fundamental 24713types. 24714 24715This attribute is appropriate for types which just represent a value, 24716such as @code{std::string}; it is not appropriate for types which 24717control a resource, such as @code{std::lock_guard}. 24718 24719This attribute is also accepted in C, but it is unnecessary because C 24720does not have constructors or destructors. 24721 24722@end table 24723 24724@node Function Multiversioning 24725@section Function Multiversioning 24726@cindex function versions 24727 24728With the GNU C++ front end, for x86 targets, you may specify multiple 24729versions of a function, where each function is specialized for a 24730specific target feature. At runtime, the appropriate version of the 24731function is automatically executed depending on the characteristics of 24732the execution platform. Here is an example. 24733 24734@smallexample 24735__attribute__ ((target ("default"))) 24736int foo () 24737@{ 24738 // The default version of foo. 24739 return 0; 24740@} 24741 24742__attribute__ ((target ("sse4.2"))) 24743int foo () 24744@{ 24745 // foo version for SSE4.2 24746 return 1; 24747@} 24748 24749__attribute__ ((target ("arch=atom"))) 24750int foo () 24751@{ 24752 // foo version for the Intel ATOM processor 24753 return 2; 24754@} 24755 24756__attribute__ ((target ("arch=amdfam10"))) 24757int foo () 24758@{ 24759 // foo version for the AMD Family 0x10 processors. 24760 return 3; 24761@} 24762 24763int main () 24764@{ 24765 int (*p)() = &foo; 24766 assert ((*p) () == foo ()); 24767 return 0; 24768@} 24769@end smallexample 24770 24771In the above example, four versions of function foo are created. The 24772first version of foo with the target attribute "default" is the default 24773version. This version gets executed when no other target specific 24774version qualifies for execution on a particular platform. A new version 24775of foo is created by using the same function signature but with a 24776different target string. Function foo is called or a pointer to it is 24777taken just like a regular function. GCC takes care of doing the 24778dispatching to call the right version at runtime. Refer to the 24779@uref{http://gcc.gnu.org/wiki/FunctionMultiVersioning, GCC wiki on 24780Function Multiversioning} for more details. 24781 24782@node Type Traits 24783@section Type Traits 24784 24785The C++ front end implements syntactic extensions that allow 24786compile-time determination of 24787various characteristics of a type (or of a 24788pair of types). 24789 24790@table @code 24791@item __has_nothrow_assign (type) 24792If @code{type} is @code{const}-qualified or is a reference type then 24793the trait is @code{false}. Otherwise if @code{__has_trivial_assign (type)} 24794is @code{true} then the trait is @code{true}, else if @code{type} is 24795a cv-qualified class or union type with copy assignment operators that are 24796known not to throw an exception then the trait is @code{true}, else it is 24797@code{false}. 24798Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24799@code{void}, or an array of unknown bound. 24800 24801@item __has_nothrow_copy (type) 24802If @code{__has_trivial_copy (type)} is @code{true} then the trait is 24803@code{true}, else if @code{type} is a cv-qualified class or union type 24804with copy constructors that are known not to throw an exception then 24805the trait is @code{true}, else it is @code{false}. 24806Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24807@code{void}, or an array of unknown bound. 24808 24809@item __has_nothrow_constructor (type) 24810If @code{__has_trivial_constructor (type)} is @code{true} then the trait 24811is @code{true}, else if @code{type} is a cv class or union type (or array 24812thereof) with a default constructor that is known not to throw an 24813exception then the trait is @code{true}, else it is @code{false}. 24814Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24815@code{void}, or an array of unknown bound. 24816 24817@item __has_trivial_assign (type) 24818If @code{type} is @code{const}- qualified or is a reference type then 24819the trait is @code{false}. Otherwise if @code{__is_pod (type)} is 24820@code{true} then the trait is @code{true}, else if @code{type} is 24821a cv-qualified class or union type with a trivial copy assignment 24822([class.copy]) then the trait is @code{true}, else it is @code{false}. 24823Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24824@code{void}, or an array of unknown bound. 24825 24826@item __has_trivial_copy (type) 24827If @code{__is_pod (type)} is @code{true} or @code{type} is a reference 24828type then the trait is @code{true}, else if @code{type} is a cv class 24829or union type with a trivial copy constructor ([class.copy]) then the trait 24830is @code{true}, else it is @code{false}. Requires: @code{type} shall be 24831a complete type, (possibly cv-qualified) @code{void}, or an array of unknown 24832bound. 24833 24834@item __has_trivial_constructor (type) 24835If @code{__is_pod (type)} is @code{true} then the trait is @code{true}, 24836else if @code{type} is a cv-qualified class or union type (or array thereof) 24837with a trivial default constructor ([class.ctor]) then the trait is @code{true}, 24838else it is @code{false}. 24839Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24840@code{void}, or an array of unknown bound. 24841 24842@item __has_trivial_destructor (type) 24843If @code{__is_pod (type)} is @code{true} or @code{type} is a reference type 24844then the trait is @code{true}, else if @code{type} is a cv class or union 24845type (or array thereof) with a trivial destructor ([class.dtor]) then 24846the trait is @code{true}, else it is @code{false}. 24847Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24848@code{void}, or an array of unknown bound. 24849 24850@item __has_virtual_destructor (type) 24851If @code{type} is a class type with a virtual destructor 24852([class.dtor]) then the trait is @code{true}, else it is @code{false}. 24853Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24854@code{void}, or an array of unknown bound. 24855 24856@item __is_abstract (type) 24857If @code{type} is an abstract class ([class.abstract]) then the trait 24858is @code{true}, else it is @code{false}. 24859Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24860@code{void}, or an array of unknown bound. 24861 24862@item __is_base_of (base_type, derived_type) 24863If @code{base_type} is a base class of @code{derived_type} 24864([class.derived]) then the trait is @code{true}, otherwise it is @code{false}. 24865Top-level cv-qualifications of @code{base_type} and 24866@code{derived_type} are ignored. For the purposes of this trait, a 24867class type is considered is own base. 24868Requires: if @code{__is_class (base_type)} and @code{__is_class (derived_type)} 24869are @code{true} and @code{base_type} and @code{derived_type} are not the same 24870type (disregarding cv-qualifiers), @code{derived_type} shall be a complete 24871type. A diagnostic is produced if this requirement is not met. 24872 24873@item __is_class (type) 24874If @code{type} is a cv-qualified class type, and not a union type 24875([basic.compound]) the trait is @code{true}, else it is @code{false}. 24876 24877@item __is_empty (type) 24878If @code{__is_class (type)} is @code{false} then the trait is @code{false}. 24879Otherwise @code{type} is considered empty if and only if: @code{type} 24880has no non-static data members, or all non-static data members, if 24881any, are bit-fields of length 0, and @code{type} has no virtual 24882members, and @code{type} has no virtual base classes, and @code{type} 24883has no base classes @code{base_type} for which 24884@code{__is_empty (base_type)} is @code{false}. 24885Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24886@code{void}, or an array of unknown bound. 24887 24888@item __is_enum (type) 24889If @code{type} is a cv enumeration type ([basic.compound]) the trait is 24890@code{true}, else it is @code{false}. 24891 24892@item __is_literal_type (type) 24893If @code{type} is a literal type ([basic.types]) the trait is 24894@code{true}, else it is @code{false}. 24895Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24896@code{void}, or an array of unknown bound. 24897 24898@item __is_pod (type) 24899If @code{type} is a cv POD type ([basic.types]) then the trait is @code{true}, 24900else it is @code{false}. 24901Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24902@code{void}, or an array of unknown bound. 24903 24904@item __is_polymorphic (type) 24905If @code{type} is a polymorphic class ([class.virtual]) then the trait 24906is @code{true}, else it is @code{false}. 24907Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24908@code{void}, or an array of unknown bound. 24909 24910@item __is_standard_layout (type) 24911If @code{type} is a standard-layout type ([basic.types]) the trait is 24912@code{true}, else it is @code{false}. 24913Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24914@code{void}, or an array of unknown bound. 24915 24916@item __is_trivial (type) 24917If @code{type} is a trivial type ([basic.types]) the trait is 24918@code{true}, else it is @code{false}. 24919Requires: @code{type} shall be a complete type, (possibly cv-qualified) 24920@code{void}, or an array of unknown bound. 24921 24922@item __is_union (type) 24923If @code{type} is a cv union type ([basic.compound]) the trait is 24924@code{true}, else it is @code{false}. 24925 24926@item __underlying_type (type) 24927The underlying type of @code{type}. 24928Requires: @code{type} shall be an enumeration type ([dcl.enum]). 24929 24930@item __integer_pack (length) 24931When used as the pattern of a pack expansion within a template 24932definition, expands to a template argument pack containing integers 24933from @code{0} to @code{length-1}. This is provided for efficient 24934implementation of @code{std::make_integer_sequence}. 24935 24936@end table 24937 24938 24939@node C++ Concepts 24940@section C++ Concepts 24941 24942C++ concepts provide much-improved support for generic programming. In 24943particular, they allow the specification of constraints on template arguments. 24944The constraints are used to extend the usual overloading and partial 24945specialization capabilities of the language, allowing generic data structures 24946and algorithms to be ``refined'' based on their properties rather than their 24947type names. 24948 24949The following keywords are reserved for concepts. 24950 24951@table @code 24952@item assumes 24953States an expression as an assumption, and if possible, verifies that the 24954assumption is valid. For example, @code{assume(n > 0)}. 24955 24956@item axiom 24957Introduces an axiom definition. Axioms introduce requirements on values. 24958 24959@item forall 24960Introduces a universally quantified object in an axiom. For example, 24961@code{forall (int n) n + 0 == n}). 24962 24963@item concept 24964Introduces a concept definition. Concepts are sets of syntactic and semantic 24965requirements on types and their values. 24966 24967@item requires 24968Introduces constraints on template arguments or requirements for a member 24969function of a class template. 24970 24971@end table 24972 24973The front end also exposes a number of internal mechanism that can be used 24974to simplify the writing of type traits. Note that some of these traits are 24975likely to be removed in the future. 24976 24977@table @code 24978@item __is_same (type1, type2) 24979A binary type trait: @code{true} whenever the type arguments are the same. 24980 24981@end table 24982 24983 24984@node Deprecated Features 24985@section Deprecated Features 24986 24987In the past, the GNU C++ compiler was extended to experiment with new 24988features, at a time when the C++ language was still evolving. Now that 24989the C++ standard is complete, some of those features are superseded by 24990superior alternatives. Using the old features might cause a warning in 24991some cases that the feature will be dropped in the future. In other 24992cases, the feature might be gone already. 24993 24994G++ allows a virtual function returning @samp{void *} to be overridden 24995by one returning a different pointer type. This extension to the 24996covariant return type rules is now deprecated and will be removed from a 24997future version. 24998 24999The use of default arguments in function pointers, function typedefs 25000and other places where they are not permitted by the standard is 25001deprecated and will be removed from a future version of G++. 25002 25003G++ allows floating-point literals to appear in integral constant expressions, 25004e.g.@: @samp{ enum E @{ e = int(2.2 * 3.7) @} } 25005This extension is deprecated and will be removed from a future version. 25006 25007G++ allows static data members of const floating-point type to be declared 25008with an initializer in a class definition. The standard only allows 25009initializers for static members of const integral types and const 25010enumeration types so this extension has been deprecated and will be removed 25011from a future version. 25012 25013G++ allows attributes to follow a parenthesized direct initializer, 25014e.g.@: @samp{ int f (0) __attribute__ ((something)); } This extension 25015has been ignored since G++ 3.3 and is deprecated. 25016 25017G++ allows anonymous structs and unions to have members that are not 25018public non-static data members (i.e.@: fields). These extensions are 25019deprecated. 25020 25021@node Backwards Compatibility 25022@section Backwards Compatibility 25023@cindex Backwards Compatibility 25024@cindex ARM [Annotated C++ Reference Manual] 25025 25026Now that there is a definitive ISO standard C++, G++ has a specification 25027to adhere to. The C++ language evolved over time, and features that 25028used to be acceptable in previous drafts of the standard, such as the ARM 25029[Annotated C++ Reference Manual], are no longer accepted. In order to allow 25030compilation of C++ written to such drafts, G++ contains some backwards 25031compatibilities. @emph{All such backwards compatibility features are 25032liable to disappear in future versions of G++.} They should be considered 25033deprecated. @xref{Deprecated Features}. 25034 25035@table @code 25036 25037@item Implicit C language 25038Old C system header files did not contain an @code{extern "C" @{@dots{}@}} 25039scope to set the language. On such systems, all system header files are 25040implicitly scoped inside a C language scope. Such headers must 25041correctly prototype function argument types, there is no leeway for 25042@code{()} to indicate an unspecified set of arguments. 25043 25044@end table 25045 25046@c LocalWords: emph deftypefn builtin ARCv2EM SIMD builtins msimd 25047@c LocalWords: typedef v4si v8hi DMA dma vdiwr vdowr 25048