1 //==--- AttrDocs.td - Attribute documentation ----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===---------------------------------------------------------------------===//
10 def GlobalDocumentation {
12 -------------------------------------------------------------------
13 NOTE: This file is automatically generated by running clang-tblgen
14 -gen-attr-docs. Do not edit this file by hand!!
15 -------------------------------------------------------------------
26 This page lists the attributes currently supported by Clang.
30 def SectionDocs : Documentation {
31 let Category = DocCatVariable;
33 The ``section`` attribute allows you to specify a specific section a
34 global variable or function should be in after translation.
36 let Heading = "section (gnu::section, __declspec(allocate))";
39 def InitSegDocs : Documentation {
40 let Category = DocCatVariable;
42 The attribute applied by ``pragma init_seg()`` controls the section into
43 which global initialization function pointers are emitted. It is only
44 available with ``-fms-extensions``. Typically, this function pointer is
45 emitted into ``.CRT$XCU`` on Windows. The user can change the order of
46 initialization by using a different section name with the same
47 ``.CRT$XC`` prefix and a suffix that sorts lexicographically before or
48 after the standard ``.CRT$XCU`` sections. See the init_seg_
49 documentation on MSDN for more information.
51 .. _init_seg: http://msdn.microsoft.com/en-us/library/7977wcck(v=vs.110).aspx
55 def TLSModelDocs : Documentation {
56 let Category = DocCatVariable;
58 The ``tls_model`` attribute allows you to specify which thread-local storage
59 model to use. It accepts the following strings:
66 TLS models are mutually exclusive.
70 def DLLExportDocs : Documentation {
71 let Category = DocCatVariable;
73 The ``__declspec(dllexport)`` attribute declares a variable, function, or
74 Objective-C interface to be exported from the module. It is available under the
75 ``-fdeclspec`` flag for compatibility with various compilers. The primary use
76 is for COFF object files which explicitly specify what interfaces are available
77 for external use. See the dllexport_ documentation on MSDN for more
80 .. _dllexport: https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx
84 def DLLImportDocs : Documentation {
85 let Category = DocCatVariable;
87 The ``__declspec(dllimport)`` attribute declares a variable, function, or
88 Objective-C interface to be imported from an external module. It is available
89 under the ``-fdeclspec`` flag for compatibility with various compilers. The
90 primary use is for COFF object files which explicitly specify what interfaces
91 are imported from external modules. See the dllimport_ documentation on MSDN
94 .. _dllimport: https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx
98 def ThreadDocs : Documentation {
99 let Category = DocCatVariable;
101 The ``__declspec(thread)`` attribute declares a variable with thread local
102 storage. It is available under the ``-fms-extensions`` flag for MSVC
103 compatibility. See the documentation for `__declspec(thread)`_ on MSDN.
105 .. _`__declspec(thread)`: http://msdn.microsoft.com/en-us/library/9w1sdazb.aspx
107 In Clang, ``__declspec(thread)`` is generally equivalent in functionality to the
108 GNU ``__thread`` keyword. The variable must not have a destructor and must have
109 a constant initializer, if any. The attribute only applies to variables
110 declared with static storage duration, such as globals, class static data
111 members, and static locals.
115 def CarriesDependencyDocs : Documentation {
116 let Category = DocCatFunction;
118 The ``carries_dependency`` attribute specifies dependency propagation into and
121 When specified on a function or Objective-C method, the ``carries_dependency``
122 attribute means that the return value carries a dependency out of the function,
123 so that the implementation need not constrain ordering upon return from that
124 function. Implementations of the function and its caller may choose to preserve
125 dependencies instead of emitting memory ordering instructions such as fences.
127 Note, this attribute does not change the meaning of the program, but may result
128 in generation of more efficient code.
132 def C11NoReturnDocs : Documentation {
133 let Category = DocCatFunction;
135 A function declared as ``_Noreturn`` shall not return to its caller. The
136 compiler will generate a diagnostic for a function declared as ``_Noreturn``
137 that appears to be capable of returning to its caller.
141 def CXX11NoReturnDocs : Documentation {
142 let Category = DocCatFunction;
144 A function declared as ``[[noreturn]]`` shall not return to its caller. The
145 compiler will generate a diagnostic for a function declared as ``[[noreturn]]``
146 that appears to be capable of returning to its caller.
150 def AssertCapabilityDocs : Documentation {
151 let Category = DocCatFunction;
152 let Heading = "assert_capability (assert_shared_capability, clang::assert_capability, clang::assert_shared_capability)";
154 Marks a function that dynamically tests whether a capability is held, and halts
155 the program if it is not held.
159 def AcquireCapabilityDocs : Documentation {
160 let Category = DocCatFunction;
161 let Heading = "acquire_capability (acquire_shared_capability, clang::acquire_capability, clang::acquire_shared_capability)";
163 Marks a function as acquiring a capability.
167 def TryAcquireCapabilityDocs : Documentation {
168 let Category = DocCatFunction;
169 let Heading = "try_acquire_capability (try_acquire_shared_capability, clang::try_acquire_capability, clang::try_acquire_shared_capability)";
171 Marks a function that attempts to acquire a capability. This function may fail to
172 actually acquire the capability; they accept a Boolean value determining
173 whether acquiring the capability means success (true), or failing to acquire
174 the capability means success (false).
178 def ReleaseCapabilityDocs : Documentation {
179 let Category = DocCatFunction;
180 let Heading = "release_capability (release_shared_capability, clang::release_capability, clang::release_shared_capability)";
182 Marks a function as releasing a capability.
186 def AssumeAlignedDocs : Documentation {
187 let Category = DocCatFunction;
189 Use ``__attribute__((assume_aligned(<alignment>[,<offset>]))`` on a function
190 declaration to specify that the return value of the function (which must be a
191 pointer type) has the specified offset, in bytes, from an address with the
192 specified alignment. The offset is taken to be zero if omitted.
196 // The returned pointer value has 32-byte alignment.
197 void *a() __attribute__((assume_aligned (32)));
199 // The returned pointer value is 4 bytes greater than an address having
200 // 32-byte alignment.
201 void *b() __attribute__((assume_aligned (32, 4)));
203 Note that this attribute provides information to the compiler regarding a
204 condition that the code already ensures is true. It does not cause the compiler
205 to enforce the provided alignment assumption.
209 def AllocSizeDocs : Documentation {
210 let Category = DocCatFunction;
212 The ``alloc_size`` attribute can be placed on functions that return pointers in
213 order to hint to the compiler how many bytes of memory will be available at the
214 returned poiner. ``alloc_size`` takes one or two arguments.
216 - ``alloc_size(N)`` implies that argument number N equals the number of
217 available bytes at the returned pointer.
218 - ``alloc_size(N, M)`` implies that the product of argument number N and
219 argument number M equals the number of available bytes at the returned
222 Argument numbers are 1-based.
224 An example of how to use ``alloc_size``
228 void *my_malloc(int a) __attribute__((alloc_size(1)));
229 void *my_calloc(int a, int b) __attribute__((alloc_size(1, 2)));
232 void *const p = my_malloc(100);
233 assert(__builtin_object_size(p, 0) == 100);
234 void *const a = my_calloc(20, 5);
235 assert(__builtin_object_size(a, 0) == 100);
238 .. Note:: This attribute works differently in clang than it does in GCC.
239 Specifically, clang will only trace ``const`` pointers (as above); we give up
240 on pointers that are not marked as ``const``. In the vast majority of cases,
241 this is unimportant, because LLVM has support for the ``alloc_size``
242 attribute. However, this may cause mildly unintuitive behavior when used with
243 other attributes, such as ``enable_if``.
247 def AllocAlignDocs : Documentation {
248 let Category = DocCatFunction;
250 Use ``__attribute__((alloc_align(<alignment>))`` on a function
251 declaration to specify that the return value of the function (which must be a
252 pointer type) is at least as aligned as the value of the indicated parameter. The
253 parameter is given by its index in the list of formal parameters; the first
254 parameter has index 1 unless the function is a C++ non-static member function,
255 in which case the first parameter has index 2 to account for the implicit ``this``
260 // The returned pointer has the alignment specified by the first parameter.
261 void *a(size_t align) __attribute__((alloc_align(1)));
263 // The returned pointer has the alignment specified by the second parameter.
264 void *b(void *v, size_t align) __attribute__((alloc_align(2)));
266 // The returned pointer has the alignment specified by the second visible
267 // parameter, however it must be adjusted for the implicit 'this' parameter.
268 void *Foo::b(void *v, size_t align) __attribute__((alloc_align(3)));
270 Note that this attribute merely informs the compiler that a function always
271 returns a sufficiently aligned pointer. It does not cause the compiler to
272 emit code to enforce that alignment. The behavior is undefined if the returned
273 poitner is not sufficiently aligned.
277 def EnableIfDocs : Documentation {
278 let Category = DocCatFunction;
280 .. Note:: Some features of this attribute are experimental. The meaning of
281 multiple enable_if attributes on a single declaration is subject to change in
282 a future version of clang. Also, the ABI is not standardized and the name
283 mangling may change in future versions. To avoid that, use asm labels.
285 The ``enable_if`` attribute can be placed on function declarations to control
286 which overload is selected based on the values of the function's arguments.
287 When combined with the ``overloadable`` attribute, this feature is also
293 int isdigit(int c) __attribute__((enable_if(c <= -1 || c > 255, "chosen when 'c' is out of range"))) __attribute__((unavailable("'c' must have the value of an unsigned char or EOF")));
298 isdigit(-10); // results in a compile-time error.
301 The enable_if attribute takes two arguments, the first is an expression written
302 in terms of the function parameters, the second is a string explaining why this
303 overload candidate could not be selected to be displayed in diagnostics. The
304 expression is part of the function signature for the purposes of determining
305 whether it is a redeclaration (following the rules used when determining
306 whether a C++ template specialization is ODR-equivalent), but is not part of
309 The enable_if expression is evaluated as if it were the body of a
310 bool-returning constexpr function declared with the arguments of the function
311 it is being applied to, then called with the parameters at the call site. If the
312 result is false or could not be determined through constant expression
313 evaluation, then this overload will not be chosen and the provided string may
314 be used in a diagnostic if the compile fails as a result.
316 Because the enable_if expression is an unevaluated context, there are no global
317 state changes, nor the ability to pass information from the enable_if
318 expression to the function body. For example, suppose we want calls to
319 strnlen(strbuf, maxlen) to resolve to strnlen_chk(strbuf, maxlen, size of
320 strbuf) only if the size of strbuf can be determined:
324 __attribute__((always_inline))
325 static inline size_t strnlen(const char *s, size_t maxlen)
326 __attribute__((overloadable))
327 __attribute__((enable_if(__builtin_object_size(s, 0) != -1))),
328 "chosen when the buffer size is known but 'maxlen' is not")))
330 return strnlen_chk(s, maxlen, __builtin_object_size(s, 0));
333 Multiple enable_if attributes may be applied to a single declaration. In this
334 case, the enable_if expressions are evaluated from left to right in the
335 following manner. First, the candidates whose enable_if expressions evaluate to
336 false or cannot be evaluated are discarded. If the remaining candidates do not
337 share ODR-equivalent enable_if expressions, the overload resolution is
338 ambiguous. Otherwise, enable_if overload resolution continues with the next
339 enable_if attribute on the candidates that have not been discarded and have
340 remaining enable_if attributes. In this way, we pick the most specific
341 overload out of a number of viable overloads using enable_if.
345 void f() __attribute__((enable_if(true, ""))); // #1
346 void f() __attribute__((enable_if(true, ""))) __attribute__((enable_if(true, ""))); // #2
348 void g(int i, int j) __attribute__((enable_if(i, ""))); // #1
349 void g(int i, int j) __attribute__((enable_if(j, ""))) __attribute__((enable_if(true))); // #2
351 In this example, a call to f() is always resolved to #2, as the first enable_if
352 expression is ODR-equivalent for both declarations, but #1 does not have another
353 enable_if expression to continue evaluating, so the next round of evaluation has
354 only a single candidate. In a call to g(1, 1), the call is ambiguous even though
355 #2 has more enable_if attributes, because the first enable_if expressions are
358 Query for this feature with ``__has_attribute(enable_if)``.
360 Note that functions with one or more ``enable_if`` attributes may not have
361 their address taken, unless all of the conditions specified by said
362 ``enable_if`` are constants that evaluate to ``true``. For example:
366 const int TrueConstant = 1;
367 const int FalseConstant = 0;
368 int f(int a) __attribute__((enable_if(a > 0, "")));
369 int g(int a) __attribute__((enable_if(a == 0 || a != 0, "")));
370 int h(int a) __attribute__((enable_if(1, "")));
371 int i(int a) __attribute__((enable_if(TrueConstant, "")));
372 int j(int a) __attribute__((enable_if(FalseConstant, "")));
376 ptr = &f; // error: 'a > 0' is not always true
377 ptr = &g; // error: 'a == 0 || a != 0' is not a truthy constant
378 ptr = &h; // OK: 1 is a truthy constant
379 ptr = &i; // OK: 'TrueConstant' is a truthy constant
380 ptr = &j; // error: 'FalseConstant' is a constant, but not truthy
383 Because ``enable_if`` evaluation happens during overload resolution,
384 ``enable_if`` may give unintuitive results when used with templates, depending
385 on when overloads are resolved. In the example below, clang will emit a
386 diagnostic about no viable overloads for ``foo`` in ``bar``, but not in ``baz``:
390 double foo(int i) __attribute__((enable_if(i > 0, "")));
391 void *foo(int i) __attribute__((enable_if(i <= 0, "")));
393 auto bar() { return foo(I); }
395 template <typename T>
396 auto baz() { return foo(T::number); }
398 struct WithNumber { constexpr static int number = 1; };
400 bar<sizeof(WithNumber)>();
404 This is because, in ``bar``, ``foo`` is resolved prior to template
405 instantiation, so the value for ``I`` isn't known (thus, both ``enable_if``
406 conditions for ``foo`` fail). However, in ``baz``, ``foo`` is resolved during
407 template instantiation, so the value for ``T::number`` is known.
411 def DiagnoseIfDocs : Documentation {
412 let Category = DocCatFunction;
414 The ``diagnose_if`` attribute can be placed on function declarations to emit
415 warnings or errors at compile-time if calls to the attributed function meet
416 certain user-defined criteria. For example:
421 __attribute__((diagnose_if(a >= 0, "Redundant abs call", "warning")));
423 __attribute__((diagnose_if(a >= 0, "Redundant abs call", "error")));
425 int val = abs(1); // warning: Redundant abs call
426 int val2 = must_abs(1); // error: Redundant abs call
428 int val4 = must_abs(val); // Because run-time checks are not emitted for
429 // diagnose_if attributes, this executes without
433 ``diagnose_if`` is closely related to ``enable_if``, with a few key differences:
435 * Overload resolution is not aware of ``diagnose_if`` attributes: they're
436 considered only after we select the best candidate from a given candidate set.
437 * Function declarations that differ only in their ``diagnose_if`` attributes are
438 considered to be redeclarations of the same function (not overloads).
439 * If the condition provided to ``diagnose_if`` cannot be evaluated, no
440 diagnostic will be emitted.
442 Otherwise, ``diagnose_if`` is essentially the logical negation of ``enable_if``.
444 As a result of bullet number two, ``diagnose_if`` attributes will stack on the
445 same function. For example:
449 int foo() __attribute__((diagnose_if(1, "diag1", "warning")));
450 int foo() __attribute__((diagnose_if(1, "diag2", "warning")));
452 int bar = foo(); // warning: diag1
454 int (*fooptr)(void) = foo; // warning: diag1
457 constexpr int supportsAPILevel(int N) { return N < 5; }
459 __attribute__((diagnose_if(!supportsAPILevel(10),
460 "Upgrade to API level 10 to use baz", "error")));
462 __attribute__((diagnose_if(!a, "0 is not recommended.", "warning")));
464 int (*bazptr)(int) = baz; // error: Upgrade to API level 10 to use baz
465 int v = baz(0); // error: Upgrade to API level 10 to use baz
467 Query for this feature with ``__has_attribute(diagnose_if)``.
471 def PassObjectSizeDocs : Documentation {
472 let Category = DocCatVariable; // Technically it's a parameter doc, but eh.
474 .. Note:: The mangling of functions with parameters that are annotated with
475 ``pass_object_size`` is subject to change. You can get around this by
476 using ``__asm__("foo")`` to explicitly name your functions, thus preserving
477 your ABI; also, non-overloadable C functions with ``pass_object_size`` are
480 The ``pass_object_size(Type)`` attribute can be placed on function parameters to
481 instruct clang to call ``__builtin_object_size(param, Type)`` at each callsite
482 of said function, and implicitly pass the result of this call in as an invisible
483 argument of type ``size_t`` directly after the parameter annotated with
484 ``pass_object_size``. Clang will also replace any calls to
485 ``__builtin_object_size(param, Type)`` in the function by said implicit
492 int bzero1(char *const p __attribute__((pass_object_size(0))))
493 __attribute__((noinline)) {
495 for (/**/; i < (int)__builtin_object_size(p, 0); ++i) {
503 int n = bzero1(&chars[0]);
504 assert(n == sizeof(chars));
508 If successfully evaluating ``__builtin_object_size(param, Type)`` at the
509 callsite is not possible, then the "failed" value is passed in. So, using the
510 definition of ``bzero1`` from above, the following code would exit cleanly:
514 int main2(int argc, char *argv[]) {
515 int n = bzero1(argv);
520 ``pass_object_size`` plays a part in overload resolution. If two overload
521 candidates are otherwise equally good, then the overload with one or more
522 parameters with ``pass_object_size`` is preferred. This implies that the choice
523 between two identical overloads both with ``pass_object_size`` on one or more
524 parameters will always be ambiguous; for this reason, having two such overloads
525 is illegal. For example:
529 #define PS(N) __attribute__((pass_object_size(N)))
531 void Foo(char *a, char *b); // Overload A
532 // OK -- overload A has no parameters with pass_object_size.
533 void Foo(char *a PS(0), char *b PS(0)); // Overload B
534 // Error -- Same signature (sans pass_object_size) as overload B, and both
535 // overloads have one or more parameters with the pass_object_size attribute.
536 void Foo(void *a PS(0), void *b);
539 void Bar(void *a PS(0)); // Overload C
541 void Bar(char *c PS(1)); // Overload D
544 char known[10], *unknown;
545 Foo(unknown, unknown); // Calls overload B
546 Foo(known, unknown); // Calls overload B
547 Foo(unknown, known); // Calls overload B
548 Foo(known, known); // Calls overload B
550 Bar(known); // Calls overload D
551 Bar(unknown); // Calls overload D
554 Currently, ``pass_object_size`` is a bit restricted in terms of its usage:
556 * Only one use of ``pass_object_size`` is allowed per parameter.
558 * It is an error to take the address of a function with ``pass_object_size`` on
559 any of its parameters. If you wish to do this, you can create an overload
560 without ``pass_object_size`` on any parameters.
562 * It is an error to apply the ``pass_object_size`` attribute to parameters that
563 are not pointers. Additionally, any parameter that ``pass_object_size`` is
564 applied to must be marked ``const`` at its function's definition.
568 def OverloadableDocs : Documentation {
569 let Category = DocCatFunction;
571 Clang provides support for C++ function overloading in C. Function overloading
572 in C is introduced using the ``overloadable`` attribute. For example, one
573 might provide several overloaded versions of a ``tgsin`` function that invokes
574 the appropriate standard function computing the sine of a value with ``float``,
575 ``double``, or ``long double`` precision:
580 float __attribute__((overloadable)) tgsin(float x) { return sinf(x); }
581 double __attribute__((overloadable)) tgsin(double x) { return sin(x); }
582 long double __attribute__((overloadable)) tgsin(long double x) { return sinl(x); }
584 Given these declarations, one can call ``tgsin`` with a ``float`` value to
585 receive a ``float`` result, with a ``double`` to receive a ``double`` result,
586 etc. Function overloading in C follows the rules of C++ function overloading
587 to pick the best overload given the call arguments, with a few C-specific
590 * Conversion from ``float`` or ``double`` to ``long double`` is ranked as a
591 floating-point promotion (per C99) rather than as a floating-point conversion
594 * A conversion from a pointer of type ``T*`` to a pointer of type ``U*`` is
595 considered a pointer conversion (with conversion rank) if ``T`` and ``U`` are
598 * A conversion from type ``T`` to a value of type ``U`` is permitted if ``T``
599 and ``U`` are compatible types. This conversion is given "conversion" rank.
601 * If no viable candidates are otherwise available, we allow a conversion from a
602 pointer of type ``T*`` to a pointer of type ``U*``, where ``T`` and ``U`` are
603 incompatible. This conversion is ranked below all other types of conversions.
604 Please note: ``U`` lacking qualifiers that are present on ``T`` is sufficient
605 for ``T`` and ``U`` to be incompatible.
607 The declaration of ``overloadable`` functions is restricted to function
608 declarations and definitions. Most importantly, if any function with a given
609 name is given the ``overloadable`` attribute, then all function declarations
610 and definitions with that name (and in that scope) must have the
611 ``overloadable`` attribute. This rule even applies to redeclarations of
612 functions whose original declaration had the ``overloadable`` attribute, e.g.,
616 int f(int) __attribute__((overloadable));
617 float f(float); // error: declaration of "f" must have the "overloadable" attribute
619 int g(int) __attribute__((overloadable));
620 int g(int) { } // error: redeclaration of "g" must also have the "overloadable" attribute
622 Functions marked ``overloadable`` must have prototypes. Therefore, the
623 following code is ill-formed:
627 int h() __attribute__((overloadable)); // error: h does not have a prototype
629 However, ``overloadable`` functions are allowed to use a ellipsis even if there
630 are no named parameters (as is permitted in C++). This feature is particularly
631 useful when combined with the ``unavailable`` attribute:
635 void honeypot(...) __attribute__((overloadable, unavailable)); // calling me is an error
637 Functions declared with the ``overloadable`` attribute have their names mangled
638 according to the same rules as C++ function names. For example, the three
639 ``tgsin`` functions in our motivating example get the mangled names
640 ``_Z5tgsinf``, ``_Z5tgsind``, and ``_Z5tgsine``, respectively. There are two
641 caveats to this use of name mangling:
643 * Future versions of Clang may change the name mangling of functions overloaded
644 in C, so you should not depend on an specific mangling. To be completely
645 safe, we strongly urge the use of ``static inline`` with ``overloadable``
648 * The ``overloadable`` attribute has almost no meaning when used in C++,
649 because names will already be mangled and functions are already overloadable.
650 However, when an ``overloadable`` function occurs within an ``extern "C"``
651 linkage specification, it's name *will* be mangled in the same way as it
654 Query for this feature with ``__has_extension(attribute_overloadable)``.
658 def ObjCMethodFamilyDocs : Documentation {
659 let Category = DocCatFunction;
661 Many methods in Objective-C have conventional meanings determined by their
662 selectors. It is sometimes useful to be able to mark a method as having a
663 particular conventional meaning despite not having the right selector, or as
664 not having the conventional meaning that its selector would suggest. For these
665 use cases, we provide an attribute to specifically describe the "method family"
666 that a method belongs to.
668 **Usage**: ``__attribute__((objc_method_family(X)))``, where ``X`` is one of
669 ``none``, ``alloc``, ``copy``, ``init``, ``mutableCopy``, or ``new``. This
670 attribute can only be placed at the end of a method declaration:
674 - (NSString *)initMyStringValue __attribute__((objc_method_family(none)));
676 Users who do not wish to change the conventional meaning of a method, and who
677 merely want to document its non-standard retain and release semantics, should
678 use the retaining behavior attributes (``ns_returns_retained``,
679 ``ns_returns_not_retained``, etc).
681 Query for this feature with ``__has_attribute(objc_method_family)``.
685 def NoDebugDocs : Documentation {
686 let Category = DocCatVariable;
688 The ``nodebug`` attribute allows you to suppress debugging information for a
689 function or method, or for a variable that is not a parameter or a non-static
694 def NoDuplicateDocs : Documentation {
695 let Category = DocCatFunction;
697 The ``noduplicate`` attribute can be placed on function declarations to control
698 whether function calls to this function can be duplicated or not as a result of
699 optimizations. This is required for the implementation of functions with
700 certain special requirements, like the OpenCL "barrier" function, that might
701 need to be run concurrently by all the threads that are executing in lockstep
702 on the hardware. For example this attribute applied on the function
703 "nodupfunc" in the code below avoids that:
707 void nodupfunc() __attribute__((noduplicate));
708 // Setting it as a C++11 attribute is also valid
709 // void nodupfunc() [[clang::noduplicate]];
720 gets possibly modified by some optimizations into code similar to this:
732 where the call to "nodupfunc" is duplicated and sunk into the two branches
737 def ConvergentDocs : Documentation {
738 let Category = DocCatFunction;
740 The ``convergent`` attribute can be placed on a function declaration. It is
741 translated into the LLVM ``convergent`` attribute, which indicates that the call
742 instructions of a function with this attribute cannot be made control-dependent
743 on any additional values.
745 In languages designed for SPMD/SIMT programming model, e.g. OpenCL or CUDA,
746 the call instructions of a function with this attribute must be executed by
747 all work items or threads in a work group or sub group.
749 This attribute is different from ``noduplicate`` because it allows duplicating
750 function calls if it can be proved that the duplicated function calls are
751 not made control-dependent on any additional values, e.g., unrolling a loop
752 executed by all work items.
757 void convfunc(void) __attribute__((convergent));
758 // Setting it as a C++11 attribute is also valid in a C++ program.
759 // void convfunc(void) [[clang::convergent]];
764 def NoSplitStackDocs : Documentation {
765 let Category = DocCatFunction;
767 The ``no_split_stack`` attribute disables the emission of the split stack
768 preamble for a particular function. It has no effect if ``-fsplit-stack``
773 def ObjCRequiresSuperDocs : Documentation {
774 let Category = DocCatFunction;
776 Some Objective-C classes allow a subclass to override a particular method in a
777 parent class but expect that the overriding method also calls the overridden
778 method in the parent class. For these cases, we provide an attribute to
779 designate that a method requires a "call to ``super``" in the overriding
780 method in the subclass.
782 **Usage**: ``__attribute__((objc_requires_super))``. This attribute can only
783 be placed at the end of a method declaration:
787 - (void)foo __attribute__((objc_requires_super));
789 This attribute can only be applied the method declarations within a class, and
790 not a protocol. Currently this attribute does not enforce any placement of
791 where the call occurs in the overriding method (such as in the case of
792 ``-dealloc`` where the call must appear at the end). It checks only that it
795 Note that on both OS X and iOS that the Foundation framework provides a
796 convenience macro ``NS_REQUIRES_SUPER`` that provides syntactic sugar for this
801 - (void)foo NS_REQUIRES_SUPER;
803 This macro is conditionally defined depending on the compiler's support for
804 this attribute. If the compiler does not support the attribute the macro
807 Operationally, when a method has this annotation the compiler will warn if the
808 implementation of an override in a subclass does not call super. For example:
812 warning: method possibly missing a [super AnnotMeth] call
813 - (void) AnnotMeth{};
818 def ObjCRuntimeNameDocs : Documentation {
819 let Category = DocCatFunction;
821 By default, the Objective-C interface or protocol identifier is used
822 in the metadata name for that object. The `objc_runtime_name`
823 attribute allows annotated interfaces or protocols to use the
824 specified string argument in the object's metadata name instead of the
827 **Usage**: ``__attribute__((objc_runtime_name("MyLocalName")))``. This attribute
828 can only be placed before an @protocol or @interface declaration:
832 __attribute__((objc_runtime_name("MyLocalName")))
839 def ObjCRuntimeVisibleDocs : Documentation {
840 let Category = DocCatFunction;
842 This attribute specifies that the Objective-C class to which it applies is visible to the Objective-C runtime but not to the linker. Classes annotated with this attribute cannot be subclassed and cannot have categories defined for them.
846 def ObjCBoxableDocs : Documentation {
847 let Category = DocCatFunction;
849 Structs and unions marked with the ``objc_boxable`` attribute can be used
850 with the Objective-C boxed expression syntax, ``@(...)``.
852 **Usage**: ``__attribute__((objc_boxable))``. This attribute
853 can only be placed on a declaration of a trivially-copyable struct or union:
857 struct __attribute__((objc_boxable)) some_struct {
860 union __attribute__((objc_boxable)) some_union {
864 typedef struct __attribute__((objc_boxable)) _some_struct some_struct;
869 NSValue *boxed = @(ss);
874 def AvailabilityDocs : Documentation {
875 let Category = DocCatFunction;
877 The ``availability`` attribute can be placed on declarations to describe the
878 lifecycle of that declaration relative to operating system versions. Consider
879 the function declaration for a hypothetical function ``f``:
883 void f(void) __attribute__((availability(macos,introduced=10.4,deprecated=10.6,obsoleted=10.7)));
885 The availability attribute states that ``f`` was introduced in Mac OS X 10.4,
886 deprecated in Mac OS X 10.6, and obsoleted in Mac OS X 10.7. This information
887 is used by Clang to determine when it is safe to use ``f``: for example, if
888 Clang is instructed to compile code for Mac OS X 10.5, a call to ``f()``
889 succeeds. If Clang is instructed to compile code for Mac OS X 10.6, the call
890 succeeds but Clang emits a warning specifying that the function is deprecated.
891 Finally, if Clang is instructed to compile code for Mac OS X 10.7, the call
892 fails because ``f()`` is no longer available.
894 The availability attribute is a comma-separated list starting with the
895 platform name and then including clauses specifying important milestones in the
896 declaration's lifetime (in any order) along with additional information. Those
899 introduced=\ *version*
900 The first version in which this declaration was introduced.
902 deprecated=\ *version*
903 The first version in which this declaration was deprecated, meaning that
904 users should migrate away from this API.
906 obsoleted=\ *version*
907 The first version in which this declaration was obsoleted, meaning that it
908 was removed completely and can no longer be used.
911 This declaration is never available on this platform.
913 message=\ *string-literal*
914 Additional message text that Clang will provide when emitting a warning or
915 error about use of a deprecated or obsoleted declaration. Useful to direct
916 users to replacement APIs.
918 replacement=\ *string-literal*
919 Additional message text that Clang will use to provide Fix-It when emitting
920 a warning about use of a deprecated declaration. The Fix-It will replace
921 the deprecated declaration with the new declaration specified.
923 Multiple availability attributes can be placed on a declaration, which may
924 correspond to different platforms. Only the availability attribute with the
925 platform corresponding to the target platform will be used; any others will be
926 ignored. If no availability attribute specifies availability for the current
927 target platform, the availability attributes are ignored. Supported platforms
931 Apple's iOS operating system. The minimum deployment target is specified by
932 the ``-mios-version-min=*version*`` or ``-miphoneos-version-min=*version*``
933 command-line arguments.
936 Apple's Mac OS X operating system. The minimum deployment target is
937 specified by the ``-mmacosx-version-min=*version*`` command-line argument.
938 ``macosx`` is supported for backward-compatibility reasons, but it is
942 Apple's tvOS operating system. The minimum deployment target is specified by
943 the ``-mtvos-version-min=*version*`` command-line argument.
946 Apple's watchOS operating system. The minimum deployment target is specified by
947 the ``-mwatchos-version-min=*version*`` command-line argument.
949 A declaration can typically be used even when deploying back to a platform
950 version prior to when the declaration was introduced. When this happens, the
951 declaration is `weakly linked
952 <https://developer.apple.com/library/mac/#documentation/MacOSX/Conceptual/BPFrameworks/Concepts/WeakLinking.html>`_,
953 as if the ``weak_import`` attribute were added to the declaration. A
954 weakly-linked declaration may or may not be present a run-time, and a program
955 can determine whether the declaration is present by checking whether the
956 address of that declaration is non-NULL.
958 The flag ``strict`` disallows using API when deploying back to a
959 platform version prior to when the declaration was introduced. An
960 attempt to use such API before its introduction causes a hard error.
961 Weakly-linking is almost always a better API choice, since it allows
962 users to query availability at runtime.
964 If there are multiple declarations of the same entity, the availability
965 attributes must either match on a per-platform basis or later
966 declarations must not have availability attributes for that
967 platform. For example:
971 void g(void) __attribute__((availability(macos,introduced=10.4)));
972 void g(void) __attribute__((availability(macos,introduced=10.4))); // okay, matches
973 void g(void) __attribute__((availability(ios,introduced=4.0))); // okay, adds a new platform
974 void g(void); // okay, inherits both macos and ios availability from above.
975 void g(void) __attribute__((availability(macos,introduced=10.5))); // error: mismatch
977 When one method overrides another, the overriding method can be more widely available than the overridden method, e.g.,:
982 - (id)method __attribute__((availability(macos,introduced=10.4)));
983 - (id)method2 __attribute__((availability(macos,introduced=10.4)));
987 - (id)method __attribute__((availability(macos,introduced=10.3))); // okay: method moved into base class later
988 - (id)method __attribute__((availability(macos,introduced=10.5))); // error: this method was available via the base class in 10.4
993 def ExternalSourceSymbolDocs : Documentation {
994 let Category = DocCatFunction;
996 The ``external_source_symbol`` attribute specifies that a declaration originates
997 from an external source and describes the nature of that source.
999 The fact that Clang is capable of recognizing declarations that were defined
1000 externally can be used to provide better tooling support for mixed-language
1001 projects or projects that rely on auto-generated code. For instance, an IDE that
1002 uses Clang and that supports mixed-language projects can use this attribute to
1003 provide a correct 'jump-to-definition' feature. For a concrete example,
1004 consider a protocol that's defined in a Swift file:
1006 .. code-block:: swift
1008 @objc public protocol SwiftProtocol {
1012 This protocol can be used from Objective-C code by including a header file that
1013 was generated by the Swift compiler. The declarations in that header can use
1014 the ``external_source_symbol`` attribute to make Clang aware of the fact
1015 that ``SwiftProtocol`` actually originates from a Swift module:
1017 .. code-block:: objc
1019 __attribute__((external_source_symbol(language="Swift",defined_in="module")))
1020 @protocol SwiftProtocol
1025 Consequently, when 'jump-to-definition' is performed at a location that
1026 references ``SwiftProtocol``, the IDE can jump to the original definition in
1027 the Swift source file rather than jumping to the Objective-C declaration in the
1028 auto-generated header file.
1030 The ``external_source_symbol`` attribute is a comma-separated list that includes
1031 clauses that describe the origin and the nature of the particular declaration.
1032 Those clauses can be:
1034 language=\ *string-literal*
1035 The name of the source language in which this declaration was defined.
1037 defined_in=\ *string-literal*
1038 The name of the source container in which the declaration was defined. The
1039 exact definition of source container is language-specific, e.g. Swift's
1040 source containers are modules, so ``defined_in`` should specify the Swift
1043 generated_declaration
1044 This declaration was automatically generated by some tool.
1046 The clauses can be specified in any order. The clauses that are listed above are
1047 all optional, but the attribute has to have at least one clause.
1051 def RequireConstantInitDocs : Documentation {
1052 let Category = DocCatVariable;
1054 This attribute specifies that the variable to which it is attached is intended
1055 to have a `constant initializer <http://en.cppreference.com/w/cpp/language/constant_initialization>`_
1056 according to the rules of [basic.start.static]. The variable is required to
1057 have static or thread storage duration. If the initialization of the variable
1058 is not a constant initializer an error will be produced. This attribute may
1059 only be used in C++.
1061 Note that in C++03 strict constant expression checking is not done. Instead
1062 the attribute reports if Clang can emit the variable as a constant, even if it's
1063 not technically a 'constant initializer'. This behavior is non-portable.
1065 Static storage duration variables with constant initializers avoid hard-to-find
1066 bugs caused by the indeterminate order of dynamic initialization. They can also
1067 be safely used during dynamic initialization across translation units.
1069 This attribute acts as a compile time assertion that the requirements
1070 for constant initialization have been met. Since these requirements change
1071 between dialects and have subtle pitfalls it's important to fail fast instead
1072 of silently falling back on dynamic initialization.
1077 #define SAFE_STATIC [[clang::require_constant_initialization]]
1080 ~T(); // non-trivial
1082 SAFE_STATIC T x = {42}; // Initialization OK. Doesn't check destructor.
1083 SAFE_STATIC T y = 42; // error: variable does not have a constant initializer
1084 // copy initialization is not a constant expression on a non-literal type.
1088 def WarnMaybeUnusedDocs : Documentation {
1089 let Category = DocCatVariable;
1090 let Heading = "maybe_unused, unused, gnu::unused";
1092 When passing the ``-Wunused`` flag to Clang, entities that are unused by the
1093 program may be diagnosed. The ``[[maybe_unused]]`` (or
1094 ``__attribute__((unused))``) attribute can be used to silence such diagnostics
1095 when the entity cannot be removed. For instance, a local variable may exist
1096 solely for use in an ``assert()`` statement, which makes the local variable
1097 unused when ``NDEBUG`` is defined.
1099 The attribute may be applied to the declaration of a class, a typedef, a
1100 variable, a function or method, a function parameter, an enumeration, an
1101 enumerator, a non-static data member, or a label.
1106 [[maybe_unused]] void f([[maybe_unused]] bool thing1,
1107 [[maybe_unused]] bool thing2) {
1108 [[maybe_unused]] bool b = thing1 && thing2;
1114 def WarnUnusedResultsDocs : Documentation {
1115 let Category = DocCatFunction;
1116 let Heading = "nodiscard, warn_unused_result, clang::warn_unused_result, gnu::warn_unused_result";
1118 Clang supports the ability to diagnose when the results of a function call
1119 expression are discarded under suspicious circumstances. A diagnostic is
1120 generated when a function or its return type is marked with ``[[nodiscard]]``
1121 (or ``__attribute__((warn_unused_result))``) and the function call appears as a
1122 potentially-evaluated discarded-value expression that is not explicitly cast to
1126 struct [[nodiscard]] error_info { /*...*/ };
1127 error_info enable_missile_safety_mode();
1129 void launch_missiles();
1130 void test_missiles() {
1131 enable_missile_safety_mode(); // diagnoses
1135 void f() { foo(); } // Does not diagnose, error_info is a reference.
1139 def FallthroughDocs : Documentation {
1140 let Category = DocCatStmt;
1141 let Heading = "fallthrough, clang::fallthrough";
1143 The ``fallthrough`` (or ``clang::fallthrough``) attribute is used
1144 to annotate intentional fall-through
1145 between switch labels. It can only be applied to a null statement placed at a
1146 point of execution between any statement and the next switch label. It is
1147 common to mark these places with a specific comment, but this attribute is
1148 meant to replace comments with a more strict annotation, which can be checked
1149 by the compiler. This attribute doesn't change semantics of the code and can
1150 be used wherever an intended fall-through occurs. It is designed to mimic
1151 control-flow statements like ``break;``, so it can be placed in most places
1152 where ``break;`` can, but only if there are no statements on the execution path
1153 between it and the next switch label.
1155 By default, Clang does not warn on unannotated fallthrough from one ``switch``
1156 case to another. Diagnostics on fallthrough without a corresponding annotation
1157 can be enabled with the ``-Wimplicit-fallthrough`` argument.
1163 // compile with -Wimplicit-fallthrough
1166 case 33: // no warning: no statements between case labels
1168 case 44: // warning: unannotated fall-through
1170 [[clang::fallthrough]];
1171 case 55: // no warning
1178 [[clang::fallthrough]];
1180 case 66: // no warning
1182 [[clang::fallthrough]]; // warning: fallthrough annotation does not
1183 // directly precede case label
1185 case 77: // warning: unannotated fall-through
1191 def ARMInterruptDocs : Documentation {
1192 let Category = DocCatFunction;
1194 Clang supports the GNU style ``__attribute__((interrupt("TYPE")))`` attribute on
1195 ARM targets. This attribute may be attached to a function definition and
1196 instructs the backend to generate appropriate function entry/exit code so that
1197 it can be used directly as an interrupt service routine.
1199 The parameter passed to the interrupt attribute is optional, but if
1200 provided it must be a string literal with one of the following values: "IRQ",
1201 "FIQ", "SWI", "ABORT", "UNDEF".
1203 The semantics are as follows:
1205 - If the function is AAPCS, Clang instructs the backend to realign the stack to
1206 8 bytes on entry. This is a general requirement of the AAPCS at public
1207 interfaces, but may not hold when an exception is taken. Doing this allows
1208 other AAPCS functions to be called.
1209 - If the CPU is M-class this is all that needs to be done since the architecture
1210 itself is designed in such a way that functions obeying the normal AAPCS ABI
1211 constraints are valid exception handlers.
1212 - If the CPU is not M-class, the prologue and epilogue are modified to save all
1213 non-banked registers that are used, so that upon return the user-mode state
1214 will not be corrupted. Note that to avoid unnecessary overhead, only
1215 general-purpose (integer) registers are saved in this way. If VFP operations
1216 are needed, that state must be saved manually.
1218 Specifically, interrupt kinds other than "FIQ" will save all core registers
1219 except "lr" and "sp". "FIQ" interrupts will save r0-r7.
1220 - If the CPU is not M-class, the return instruction is changed to one of the
1221 canonical sequences permitted by the architecture for exception return. Where
1222 possible the function itself will make the necessary "lr" adjustments so that
1223 the "preferred return address" is selected.
1225 Unfortunately the compiler is unable to make this guarantee for an "UNDEF"
1226 handler, where the offset from "lr" to the preferred return address depends on
1227 the execution state of the code which generated the exception. In this case
1228 a sequence equivalent to "movs pc, lr" will be used.
1232 def MipsInterruptDocs : Documentation {
1233 let Category = DocCatFunction;
1235 Clang supports the GNU style ``__attribute__((interrupt("ARGUMENT")))`` attribute on
1236 MIPS targets. This attribute may be attached to a function definition and instructs
1237 the backend to generate appropriate function entry/exit code so that it can be used
1238 directly as an interrupt service routine.
1240 By default, the compiler will produce a function prologue and epilogue suitable for
1241 an interrupt service routine that handles an External Interrupt Controller (eic)
1242 generated interrupt. This behaviour can be explicitly requested with the "eic"
1245 Otherwise, for use with vectored interrupt mode, the argument passed should be
1246 of the form "vector=LEVEL" where LEVEL is one of the following values:
1247 "sw0", "sw1", "hw0", "hw1", "hw2", "hw3", "hw4", "hw5". The compiler will
1248 then set the interrupt mask to the corresponding level which will mask all
1249 interrupts up to and including the argument.
1251 The semantics are as follows:
1253 - The prologue is modified so that the Exception Program Counter (EPC) and
1254 Status coprocessor registers are saved to the stack. The interrupt mask is
1255 set so that the function can only be interrupted by a higher priority
1256 interrupt. The epilogue will restore the previous values of EPC and Status.
1258 - The prologue and epilogue are modified to save and restore all non-kernel
1259 registers as necessary.
1261 - The FPU is disabled in the prologue, as the floating pointer registers are not
1262 spilled to the stack.
1264 - The function return sequence is changed to use an exception return instruction.
1266 - The parameter sets the interrupt mask for the function corresponding to the
1267 interrupt level specified. If no mask is specified the interrupt mask
1272 def MicroMipsDocs : Documentation {
1273 let Category = DocCatFunction;
1275 Clang supports the GNU style ``__attribute__((micromips))`` and
1276 ``__attribute__((nomicromips))`` attributes on MIPS targets. These attributes
1277 may be attached to a function definition and instructs the backend to generate
1278 or not to generate microMIPS code for that function.
1280 These attributes override the `-mmicromips` and `-mno-micromips` options
1281 on the command line.
1285 def AVRInterruptDocs : Documentation {
1286 let Category = DocCatFunction;
1288 Clang supports the GNU style ``__attribute__((interrupt))`` attribute on
1289 AVR targets. This attribute may be attached to a function definition and instructs
1290 the backend to generate appropriate function entry/exit code so that it can be used
1291 directly as an interrupt service routine.
1293 On the AVR, the hardware globally disables interrupts when an interrupt is executed.
1294 The first instruction of an interrupt handler declared with this attribute is a SEI
1295 instruction to re-enable interrupts. See also the signal attribute that
1296 does not insert a SEI instruction.
1300 def AVRSignalDocs : Documentation {
1301 let Category = DocCatFunction;
1303 Clang supports the GNU style ``__attribute__((signal))`` attribute on
1304 AVR targets. This attribute may be attached to a function definition and instructs
1305 the backend to generate appropriate function entry/exit code so that it can be used
1306 directly as an interrupt service routine.
1308 Interrupt handler functions defined with the signal attribute do not re-enable interrupts.
1312 def TargetDocs : Documentation {
1313 let Category = DocCatFunction;
1315 Clang supports the GNU style ``__attribute__((target("OPTIONS")))`` attribute.
1316 This attribute may be attached to a function definition and instructs
1317 the backend to use different code generation options than were passed on the
1320 The current set of options correspond to the existing "subtarget features" for
1321 the target with or without a "-mno-" in front corresponding to the absence
1322 of the feature, as well as ``arch="CPU"`` which will change the default "CPU"
1325 Example "subtarget features" from the x86 backend include: "mmx", "sse", "sse4.2",
1326 "avx", "xop" and largely correspond to the machine specific options handled by
1331 def DocCatAMDGPUAttributes : DocumentationCategory<"AMD GPU Attributes">;
1333 def AMDGPUFlatWorkGroupSizeDocs : Documentation {
1334 let Category = DocCatAMDGPUAttributes;
1336 The flat work-group size is the number of work-items in the work-group size
1337 specified when the kernel is dispatched. It is the product of the sizes of the
1338 x, y, and z dimension of the work-group.
1341 ``__attribute__((amdgpu_flat_work_group_size(<min>, <max>)))`` attribute for the
1342 AMDGPU target. This attribute may be attached to a kernel function definition
1343 and is an optimization hint.
1345 ``<min>`` parameter specifies the minimum flat work-group size, and ``<max>``
1346 parameter specifies the maximum flat work-group size (must be greater than
1347 ``<min>``) to which all dispatches of the kernel will conform. Passing ``0, 0``
1348 as ``<min>, <max>`` implies the default behavior (``128, 256``).
1350 If specified, the AMDGPU target backend might be able to produce better machine
1351 code for barriers and perform scratch promotion by estimating available group
1354 An error will be given if:
1355 - Specified values violate subtarget specifications;
1356 - Specified values are not compatible with values provided through other
1361 def AMDGPUWavesPerEUDocs : Documentation {
1362 let Category = DocCatAMDGPUAttributes;
1364 A compute unit (CU) is responsible for executing the wavefronts of a work-group.
1365 It is composed of one or more execution units (EU), which are responsible for
1366 executing the wavefronts. An EU can have enough resources to maintain the state
1367 of more than one executing wavefront. This allows an EU to hide latency by
1368 switching between wavefronts in a similar way to symmetric multithreading on a
1369 CPU. In order to allow the state for multiple wavefronts to fit on an EU, the
1370 resources used by a single wavefront have to be limited. For example, the number
1371 of SGPRs and VGPRs. Limiting such resources can allow greater latency hiding,
1372 but can result in having to spill some register state to memory.
1374 Clang supports the ``__attribute__((amdgpu_waves_per_eu(<min>[, <max>])))``
1375 attribute for the AMDGPU target. This attribute may be attached to a kernel
1376 function definition and is an optimization hint.
1378 ``<min>`` parameter specifies the requested minimum number of waves per EU, and
1379 *optional* ``<max>`` parameter specifies the requested maximum number of waves
1380 per EU (must be greater than ``<min>`` if specified). If ``<max>`` is omitted,
1381 then there is no restriction on the maximum number of waves per EU other than
1382 the one dictated by the hardware for which the kernel is compiled. Passing
1383 ``0, 0`` as ``<min>, <max>`` implies the default behavior (no limits).
1385 If specified, this attribute allows an advanced developer to tune the number of
1386 wavefronts that are capable of fitting within the resources of an EU. The AMDGPU
1387 target backend can use this information to limit resources, such as number of
1388 SGPRs, number of VGPRs, size of available group and private memory segments, in
1389 such a way that guarantees that at least ``<min>`` wavefronts and at most
1390 ``<max>`` wavefronts are able to fit within the resources of an EU. Requesting
1391 more wavefronts can hide memory latency but limits available registers which
1392 can result in spilling. Requesting fewer wavefronts can help reduce cache
1393 thrashing, but can reduce memory latency hiding.
1395 This attribute controls the machine code generated by the AMDGPU target backend
1396 to ensure it is capable of meeting the requested values. However, when the
1397 kernel is executed, there may be other reasons that prevent meeting the request,
1398 for example, there may be wavefronts from other kernels executing on the EU.
1400 An error will be given if:
1401 - Specified values violate subtarget specifications;
1402 - Specified values are not compatible with values provided through other
1404 - The AMDGPU target backend is unable to create machine code that can meet the
1409 def AMDGPUNumSGPRNumVGPRDocs : Documentation {
1410 let Category = DocCatAMDGPUAttributes;
1412 Clang supports the ``__attribute__((amdgpu_num_sgpr(<num_sgpr>)))`` and
1413 ``__attribute__((amdgpu_num_vgpr(<num_vgpr>)))`` attributes for the AMDGPU
1414 target. These attributes may be attached to a kernel function definition and are
1415 an optimization hint.
1417 If these attributes are specified, then the AMDGPU target backend will attempt
1418 to limit the number of SGPRs and/or VGPRs used to the specified value(s). The
1419 number of used SGPRs and/or VGPRs may further be rounded up to satisfy the
1420 allocation requirements or constraints of the subtarget. Passing ``0`` as
1421 ``num_sgpr`` and/or ``num_vgpr`` implies the default behavior (no limits).
1423 These attributes can be used to test the AMDGPU target backend. It is
1424 recommended that the ``amdgpu_waves_per_eu`` attribute be used to control
1425 resources such as SGPRs and VGPRs since it is aware of the limits for different
1428 An error will be given if:
1429 - Specified values violate subtarget specifications;
1430 - Specified values are not compatible with values provided through other
1432 - The AMDGPU target backend is unable to create machine code that can meet the
1437 def DocCatCallingConvs : DocumentationCategory<"Calling Conventions"> {
1439 Clang supports several different calling conventions, depending on the target
1440 platform and architecture. The calling convention used for a function determines
1441 how parameters are passed, how results are returned to the caller, and other
1442 low-level details of calling a function.
1446 def PcsDocs : Documentation {
1447 let Category = DocCatCallingConvs;
1449 On ARM targets, this attribute can be used to select calling conventions
1450 similar to ``stdcall`` on x86. Valid parameter values are "aapcs" and
1455 def RegparmDocs : Documentation {
1456 let Category = DocCatCallingConvs;
1458 On 32-bit x86 targets, the regparm attribute causes the compiler to pass
1459 the first three integer parameters in EAX, EDX, and ECX instead of on the
1460 stack. This attribute has no effect on variadic functions, and all parameters
1461 are passed via the stack as normal.
1465 def SysVABIDocs : Documentation {
1466 let Category = DocCatCallingConvs;
1468 On Windows x86_64 targets, this attribute changes the calling convention of a
1469 function to match the default convention used on Sys V targets such as Linux,
1470 Mac, and BSD. This attribute has no effect on other targets.
1474 def MSABIDocs : Documentation {
1475 let Category = DocCatCallingConvs;
1477 On non-Windows x86_64 targets, this attribute changes the calling convention of
1478 a function to match the default convention used on Windows x86_64. This
1479 attribute has no effect on Windows targets or non-x86_64 targets.
1483 def StdCallDocs : Documentation {
1484 let Category = DocCatCallingConvs;
1486 On 32-bit x86 targets, this attribute changes the calling convention of a
1487 function to clear parameters off of the stack on return. This convention does
1488 not support variadic calls or unprototyped functions in C, and has no effect on
1489 x86_64 targets. This calling convention is used widely by the Windows API and
1490 COM applications. See the documentation for `__stdcall`_ on MSDN.
1492 .. _`__stdcall`: http://msdn.microsoft.com/en-us/library/zxk0tw93.aspx
1496 def FastCallDocs : Documentation {
1497 let Category = DocCatCallingConvs;
1499 On 32-bit x86 targets, this attribute changes the calling convention of a
1500 function to use ECX and EDX as register parameters and clear parameters off of
1501 the stack on return. This convention does not support variadic calls or
1502 unprototyped functions in C, and has no effect on x86_64 targets. This calling
1503 convention is supported primarily for compatibility with existing code. Users
1504 seeking register parameters should use the ``regparm`` attribute, which does
1505 not require callee-cleanup. See the documentation for `__fastcall`_ on MSDN.
1507 .. _`__fastcall`: http://msdn.microsoft.com/en-us/library/6xa169sk.aspx
1511 def RegCallDocs : Documentation {
1512 let Category = DocCatCallingConvs;
1514 On x86 targets, this attribute changes the calling convention to
1515 `__regcall`_ convention. This convention aims to pass as many arguments
1516 as possible in registers. It also tries to utilize registers for the
1517 return value whenever it is possible.
1519 .. _`__regcall`: https://software.intel.com/en-us/node/693069
1523 def ThisCallDocs : Documentation {
1524 let Category = DocCatCallingConvs;
1526 On 32-bit x86 targets, this attribute changes the calling convention of a
1527 function to use ECX for the first parameter (typically the implicit ``this``
1528 parameter of C++ methods) and clear parameters off of the stack on return. This
1529 convention does not support variadic calls or unprototyped functions in C, and
1530 has no effect on x86_64 targets. See the documentation for `__thiscall`_ on
1533 .. _`__thiscall`: http://msdn.microsoft.com/en-us/library/ek8tkfbw.aspx
1537 def VectorCallDocs : Documentation {
1538 let Category = DocCatCallingConvs;
1540 On 32-bit x86 *and* x86_64 targets, this attribute changes the calling
1541 convention of a function to pass vector parameters in SSE registers.
1543 On 32-bit x86 targets, this calling convention is similar to ``__fastcall``.
1544 The first two integer parameters are passed in ECX and EDX. Subsequent integer
1545 parameters are passed in memory, and callee clears the stack. On x86_64
1546 targets, the callee does *not* clear the stack, and integer parameters are
1547 passed in RCX, RDX, R8, and R9 as is done for the default Windows x64 calling
1550 On both 32-bit x86 and x86_64 targets, vector and floating point arguments are
1551 passed in XMM0-XMM5. Homogeneous vector aggregates of up to four elements are
1552 passed in sequential SSE registers if enough are available. If AVX is enabled,
1553 256 bit vectors are passed in YMM0-YMM5. Any vector or aggregate type that
1554 cannot be passed in registers for any reason is passed by reference, which
1555 allows the caller to align the parameter memory.
1557 See the documentation for `__vectorcall`_ on MSDN for more details.
1559 .. _`__vectorcall`: http://msdn.microsoft.com/en-us/library/dn375768.aspx
1563 def DocCatConsumed : DocumentationCategory<"Consumed Annotation Checking"> {
1565 Clang supports additional attributes for checking basic resource management
1566 properties, specifically for unique objects that have a single owning reference.
1567 The following attributes are currently supported, although **the implementation
1568 for these annotations is currently in development and are subject to change.**
1572 def SetTypestateDocs : Documentation {
1573 let Category = DocCatConsumed;
1575 Annotate methods that transition an object into a new state with
1576 ``__attribute__((set_typestate(new_state)))``. The new state must be
1577 unconsumed, consumed, or unknown.
1581 def CallableWhenDocs : Documentation {
1582 let Category = DocCatConsumed;
1584 Use ``__attribute__((callable_when(...)))`` to indicate what states a method
1585 may be called in. Valid states are unconsumed, consumed, or unknown. Each
1586 argument to this attribute must be a quoted string. E.g.:
1588 ``__attribute__((callable_when("unconsumed", "unknown")))``
1592 def TestTypestateDocs : Documentation {
1593 let Category = DocCatConsumed;
1595 Use ``__attribute__((test_typestate(tested_state)))`` to indicate that a method
1596 returns true if the object is in the specified state..
1600 def ParamTypestateDocs : Documentation {
1601 let Category = DocCatConsumed;
1603 This attribute specifies expectations about function parameters. Calls to an
1604 function with annotated parameters will issue a warning if the corresponding
1605 argument isn't in the expected state. The attribute is also used to set the
1606 initial state of the parameter when analyzing the function's body.
1610 def ReturnTypestateDocs : Documentation {
1611 let Category = DocCatConsumed;
1613 The ``return_typestate`` attribute can be applied to functions or parameters.
1614 When applied to a function the attribute specifies the state of the returned
1615 value. The function's body is checked to ensure that it always returns a value
1616 in the specified state. On the caller side, values returned by the annotated
1617 function are initialized to the given state.
1619 When applied to a function parameter it modifies the state of an argument after
1620 a call to the function returns. The function's body is checked to ensure that
1621 the parameter is in the expected state before returning.
1625 def ConsumableDocs : Documentation {
1626 let Category = DocCatConsumed;
1628 Each ``class`` that uses any of the typestate annotations must first be marked
1629 using the ``consumable`` attribute. Failure to do so will result in a warning.
1631 This attribute accepts a single parameter that must be one of the following:
1632 ``unknown``, ``consumed``, or ``unconsumed``.
1636 def NoSanitizeDocs : Documentation {
1637 let Category = DocCatFunction;
1639 Use the ``no_sanitize`` attribute on a function declaration to specify
1640 that a particular instrumentation or set of instrumentations should not be
1641 applied to that function. The attribute takes a list of string literals,
1642 which have the same meaning as values accepted by the ``-fno-sanitize=``
1643 flag. For example, ``__attribute__((no_sanitize("address", "thread")))``
1644 specifies that AddressSanitizer and ThreadSanitizer should not be applied
1647 See :ref:`Controlling Code Generation <controlling-code-generation>` for a
1648 full list of supported sanitizer flags.
1652 def NoSanitizeAddressDocs : Documentation {
1653 let Category = DocCatFunction;
1654 // This function has multiple distinct spellings, and so it requires a custom
1655 // heading to be specified. The most common spelling is sufficient.
1656 let Heading = "no_sanitize_address (no_address_safety_analysis, gnu::no_address_safety_analysis, gnu::no_sanitize_address)";
1658 .. _langext-address_sanitizer:
1660 Use ``__attribute__((no_sanitize_address))`` on a function declaration to
1661 specify that address safety instrumentation (e.g. AddressSanitizer) should
1662 not be applied to that function.
1666 def NoSanitizeThreadDocs : Documentation {
1667 let Category = DocCatFunction;
1668 let Heading = "no_sanitize_thread";
1670 .. _langext-thread_sanitizer:
1672 Use ``__attribute__((no_sanitize_thread))`` on a function declaration to
1673 specify that checks for data races on plain (non-atomic) memory accesses should
1674 not be inserted by ThreadSanitizer. The function is still instrumented by the
1675 tool to avoid false positives and provide meaningful stack traces.
1679 def NoSanitizeMemoryDocs : Documentation {
1680 let Category = DocCatFunction;
1681 let Heading = "no_sanitize_memory";
1683 .. _langext-memory_sanitizer:
1685 Use ``__attribute__((no_sanitize_memory))`` on a function declaration to
1686 specify that checks for uninitialized memory should not be inserted
1687 (e.g. by MemorySanitizer). The function may still be instrumented by the tool
1688 to avoid false positives in other places.
1692 def DocCatTypeSafety : DocumentationCategory<"Type Safety Checking"> {
1694 Clang supports additional attributes to enable checking type safety properties
1695 that can't be enforced by the C type system. To see warnings produced by these
1696 checks, ensure that -Wtype-safety is enabled. Use cases include:
1698 * MPI library implementations, where these attributes enable checking that
1699 the buffer type matches the passed ``MPI_Datatype``;
1700 * for HDF5 library there is a similar use case to MPI;
1701 * checking types of variadic functions' arguments for functions like
1702 ``fcntl()`` and ``ioctl()``.
1704 You can detect support for these attributes with ``__has_attribute()``. For
1709 #if defined(__has_attribute)
1710 # if __has_attribute(argument_with_type_tag) && \
1711 __has_attribute(pointer_with_type_tag) && \
1712 __has_attribute(type_tag_for_datatype)
1713 # define ATTR_MPI_PWT(buffer_idx, type_idx) __attribute__((pointer_with_type_tag(mpi,buffer_idx,type_idx)))
1714 /* ... other macros ... */
1718 #if !defined(ATTR_MPI_PWT)
1719 # define ATTR_MPI_PWT(buffer_idx, type_idx)
1722 int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
1727 def ArgumentWithTypeTagDocs : Documentation {
1728 let Category = DocCatTypeSafety;
1729 let Heading = "argument_with_type_tag";
1731 Use ``__attribute__((argument_with_type_tag(arg_kind, arg_idx,
1732 type_tag_idx)))`` on a function declaration to specify that the function
1733 accepts a type tag that determines the type of some other argument.
1735 This attribute is primarily useful for checking arguments of variadic functions
1736 (``pointer_with_type_tag`` can be used in most non-variadic cases).
1738 In the attribute prototype above:
1739 * ``arg_kind`` is an identifier that should be used when annotating all
1740 applicable type tags.
1741 * ``arg_idx`` provides the position of a function argument. The expected type of
1742 this function argument will be determined by the function argument specified
1743 by ``type_tag_idx``. In the code example below, "3" means that the type of the
1744 function's third argument will be determined by ``type_tag_idx``.
1745 * ``type_tag_idx`` provides the position of a function argument. This function
1746 argument will be a type tag. The type tag will determine the expected type of
1747 the argument specified by ``arg_idx``. In the code example below, "2" means
1748 that the type tag associated with the function's second argument should agree
1749 with the type of the argument specified by ``arg_idx``.
1755 int fcntl(int fd, int cmd, ...)
1756 __attribute__(( argument_with_type_tag(fcntl,3,2) ));
1757 // The function's second argument will be a type tag; this type tag will
1758 // determine the expected type of the function's third argument.
1762 def PointerWithTypeTagDocs : Documentation {
1763 let Category = DocCatTypeSafety;
1764 let Heading = "pointer_with_type_tag";
1766 Use ``__attribute__((pointer_with_type_tag(ptr_kind, ptr_idx, type_tag_idx)))``
1767 on a function declaration to specify that the function accepts a type tag that
1768 determines the pointee type of some other pointer argument.
1770 In the attribute prototype above:
1771 * ``ptr_kind`` is an identifier that should be used when annotating all
1772 applicable type tags.
1773 * ``ptr_idx`` provides the position of a function argument; this function
1774 argument will have a pointer type. The expected pointee type of this pointer
1775 type will be determined by the function argument specified by
1776 ``type_tag_idx``. In the code example below, "1" means that the pointee type
1777 of the function's first argument will be determined by ``type_tag_idx``.
1778 * ``type_tag_idx`` provides the position of a function argument; this function
1779 argument will be a type tag. The type tag will determine the expected pointee
1780 type of the pointer argument specified by ``ptr_idx``. In the code example
1781 below, "3" means that the type tag associated with the function's third
1782 argument should agree with the pointee type of the pointer argument specified
1789 typedef int MPI_Datatype;
1790 int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
1791 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1792 // The function's 3rd argument will be a type tag; this type tag will
1793 // determine the expected pointee type of the function's 1st argument.
1797 def TypeTagForDatatypeDocs : Documentation {
1798 let Category = DocCatTypeSafety;
1800 When declaring a variable, use
1801 ``__attribute__((type_tag_for_datatype(kind, type)))`` to create a type tag that
1802 is tied to the ``type`` argument given to the attribute.
1804 In the attribute prototype above:
1805 * ``kind`` is an identifier that should be used when annotating all applicable
1807 * ``type`` indicates the name of the type.
1809 Clang supports annotating type tags of two forms.
1811 * **Type tag that is a reference to a declared identifier.**
1812 Use ``__attribute__((type_tag_for_datatype(kind, type)))`` when declaring that
1817 typedef int MPI_Datatype;
1818 extern struct mpi_datatype mpi_datatype_int
1819 __attribute__(( type_tag_for_datatype(mpi,int) ));
1820 #define MPI_INT ((MPI_Datatype) &mpi_datatype_int)
1821 // &mpi_datatype_int is a type tag. It is tied to type "int".
1823 * **Type tag that is an integral literal.**
1824 Declare a ``static const`` variable with an initializer value and attach
1825 ``__attribute__((type_tag_for_datatype(kind, type)))`` on that declaration:
1829 typedef int MPI_Datatype;
1830 static const MPI_Datatype mpi_datatype_int
1831 __attribute__(( type_tag_for_datatype(mpi,int) )) = 42;
1832 #define MPI_INT ((MPI_Datatype) 42)
1833 // The number 42 is a type tag. It is tied to type "int".
1836 The ``type_tag_for_datatype`` attribute also accepts an optional third argument
1837 that determines how the type of the function argument specified by either
1838 ``arg_idx`` or ``ptr_idx`` is compared against the type associated with the type
1839 tag. (Recall that for the ``argument_with_type_tag`` attribute, the type of the
1840 function argument specified by ``arg_idx`` is compared against the type
1841 associated with the type tag. Also recall that for the ``pointer_with_type_tag``
1842 attribute, the pointee type of the function argument specified by ``ptr_idx`` is
1843 compared against the type associated with the type tag.) There are two supported
1844 values for this optional third argument:
1846 * ``layout_compatible`` will cause types to be compared according to
1847 layout-compatibility rules (In C++11 [class.mem] p 17, 18, see the
1848 layout-compatibility rules for two standard-layout struct types and for two
1849 standard-layout union types). This is useful when creating a type tag
1850 associated with a struct or union type. For example:
1855 typedef int MPI_Datatype;
1856 struct internal_mpi_double_int { double d; int i; };
1857 extern struct mpi_datatype mpi_datatype_double_int
1858 __attribute__(( type_tag_for_datatype(mpi,
1859 struct internal_mpi_double_int, layout_compatible) ));
1861 #define MPI_DOUBLE_INT ((MPI_Datatype) &mpi_datatype_double_int)
1863 int MPI_Send(void *buf, int count, MPI_Datatype datatype, ...)
1864 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1867 struct my_pair { double a; int b; };
1868 struct my_pair *buffer;
1869 MPI_Send(buffer, 1, MPI_DOUBLE_INT /*, ... */); // no warning because the
1870 // layout of my_pair is
1871 // compatible with that of
1872 // internal_mpi_double_int
1874 struct my_int_pair { int a; int b; }
1875 struct my_int_pair *buffer2;
1876 MPI_Send(buffer2, 1, MPI_DOUBLE_INT /*, ... */); // warning because the
1877 // layout of my_int_pair
1878 // does not match that of
1879 // internal_mpi_double_int
1881 * ``must_be_null`` specifies that the function argument specified by either
1882 ``arg_idx`` (for the ``argument_with_type_tag`` attribute) or ``ptr_idx`` (for
1883 the ``pointer_with_type_tag`` attribute) should be a null pointer constant.
1884 The second argument to the ``type_tag_for_datatype`` attribute is ignored. For
1890 typedef int MPI_Datatype;
1891 extern struct mpi_datatype mpi_datatype_null
1892 __attribute__(( type_tag_for_datatype(mpi, void, must_be_null) ));
1894 #define MPI_DATATYPE_NULL ((MPI_Datatype) &mpi_datatype_null)
1895 int MPI_Send(void *buf, int count, MPI_Datatype datatype, ...)
1896 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1899 struct my_pair { double a; int b; };
1900 struct my_pair *buffer;
1901 MPI_Send(buffer, 1, MPI_DATATYPE_NULL /*, ... */); // warning: MPI_DATATYPE_NULL
1902 // was specified but buffer
1903 // is not a null pointer
1907 def FlattenDocs : Documentation {
1908 let Category = DocCatFunction;
1910 The ``flatten`` attribute causes calls within the attributed function to
1911 be inlined unless it is impossible to do so, for example if the body of the
1912 callee is unavailable or if the callee has the ``noinline`` attribute.
1916 def FormatDocs : Documentation {
1917 let Category = DocCatFunction;
1920 Clang supports the ``format`` attribute, which indicates that the function
1921 accepts a ``printf`` or ``scanf``-like format string and corresponding
1922 arguments or a ``va_list`` that contains these arguments.
1924 Please see `GCC documentation about format attribute
1925 <http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html>`_ to find details
1926 about attribute syntax.
1928 Clang implements two kinds of checks with this attribute.
1930 #. Clang checks that the function with the ``format`` attribute is called with
1931 a format string that uses format specifiers that are allowed, and that
1932 arguments match the format string. This is the ``-Wformat`` warning, it is
1935 #. Clang checks that the format string argument is a literal string. This is
1936 the ``-Wformat-nonliteral`` warning, it is off by default.
1938 Clang implements this mostly the same way as GCC, but there is a difference
1939 for functions that accept a ``va_list`` argument (for example, ``vprintf``).
1940 GCC does not emit ``-Wformat-nonliteral`` warning for calls to such
1941 functions. Clang does not warn if the format string comes from a function
1942 parameter, where the function is annotated with a compatible attribute,
1943 otherwise it warns. For example:
1947 __attribute__((__format__ (__scanf__, 1, 3)))
1948 void foo(const char* s, char *buf, ...) {
1952 vprintf(s, ap); // warning: format string is not a string literal
1955 In this case we warn because ``s`` contains a format string for a
1956 ``scanf``-like function, but it is passed to a ``printf``-like function.
1958 If the attribute is removed, clang still warns, because the format string is
1959 not a string literal.
1965 __attribute__((__format__ (__printf__, 1, 3)))
1966 void foo(const char* s, char *buf, ...) {
1970 vprintf(s, ap); // warning
1973 In this case Clang does not warn because the format string ``s`` and
1974 the corresponding arguments are annotated. If the arguments are
1975 incorrect, the caller of ``foo`` will receive a warning.
1979 def AlignValueDocs : Documentation {
1980 let Category = DocCatType;
1982 The align_value attribute can be added to the typedef of a pointer type or the
1983 declaration of a variable of pointer or reference type. It specifies that the
1984 pointer will point to, or the reference will bind to, only objects with at
1985 least the provided alignment. This alignment value must be some positive power
1990 typedef double * aligned_double_ptr __attribute__((align_value(64)));
1991 void foo(double & x __attribute__((align_value(128)),
1992 aligned_double_ptr y) { ... }
1994 If the pointer value does not have the specified alignment at runtime, the
1995 behavior of the program is undefined.
1999 def FlagEnumDocs : Documentation {
2000 let Category = DocCatType;
2002 This attribute can be added to an enumerator to signal to the compiler that it
2003 is intended to be used as a flag type. This will cause the compiler to assume
2004 that the range of the type includes all of the values that you can get by
2005 manipulating bits of the enumerator when issuing warnings.
2009 def EnumExtensibilityDocs : Documentation {
2010 let Category = DocCatType;
2012 Attribute ``enum_extensibility`` is used to distinguish between enum definitions
2013 that are extensible and those that are not. The attribute can take either
2014 ``closed`` or ``open`` as an argument. ``closed`` indicates a variable of the
2015 enum type takes a value that corresponds to one of the enumerators listed in the
2016 enum definition or, when the enum is annotated with ``flag_enum``, a value that
2017 can be constructed using values corresponding to the enumerators. ``open``
2018 indicates a variable of the enum type can take any values allowed by the
2019 standard and instructs clang to be more lenient when issuing warnings.
2023 enum __attribute__((enum_extensibility(closed))) ClosedEnum {
2027 enum __attribute__((enum_extensibility(open))) OpenEnum {
2031 enum __attribute__((enum_extensibility(closed),flag_enum)) ClosedFlagEnum {
2032 C0 = 1 << 0, C1 = 1 << 1
2035 enum __attribute__((enum_extensibility(open),flag_enum)) OpenFlagEnum {
2036 D0 = 1 << 0, D1 = 1 << 1
2042 enum ClosedFlagEnum cfe;
2043 enum OpenFlagEnum ofe;
2045 ce = A1; // no warnings
2046 ce = 100; // warning issued
2047 oe = B1; // no warnings
2048 oe = 100; // no warnings
2049 cfe = C0 | C1; // no warnings
2050 cfe = C0 | C1 | 4; // warning issued
2051 ofe = D0 | D1; // no warnings
2052 ofe = D0 | D1 | 4; // no warnings
2058 def EmptyBasesDocs : Documentation {
2059 let Category = DocCatType;
2061 The empty_bases attribute permits the compiler to utilize the
2062 empty-base-optimization more frequently.
2063 This attribute only applies to struct, class, and union types.
2064 It is only supported when using the Microsoft C++ ABI.
2068 def LayoutVersionDocs : Documentation {
2069 let Category = DocCatType;
2071 The layout_version attribute requests that the compiler utilize the class
2072 layout rules of a particular compiler version.
2073 This attribute only applies to struct, class, and union types.
2074 It is only supported when using the Microsoft C++ ABI.
2078 def MSInheritanceDocs : Documentation {
2079 let Category = DocCatType;
2080 let Heading = "__single_inhertiance, __multiple_inheritance, __virtual_inheritance";
2082 This collection of keywords is enabled under ``-fms-extensions`` and controls
2083 the pointer-to-member representation used on ``*-*-win32`` targets.
2085 The ``*-*-win32`` targets utilize a pointer-to-member representation which
2086 varies in size and alignment depending on the definition of the underlying
2089 However, this is problematic when a forward declaration is only available and
2090 no definition has been made yet. In such cases, Clang is forced to utilize the
2091 most general representation that is available to it.
2093 These keywords make it possible to use a pointer-to-member representation other
2094 than the most general one regardless of whether or not the definition will ever
2095 be present in the current translation unit.
2097 This family of keywords belong between the ``class-key`` and ``class-name``:
2101 struct __single_inheritance S;
2105 This keyword can be applied to class templates but only has an effect when used
2106 on full specializations:
2110 template <typename T, typename U> struct __single_inheritance A; // warning: inheritance model ignored on primary template
2111 template <typename T> struct __multiple_inheritance A<T, T>; // warning: inheritance model ignored on partial specialization
2112 template <> struct __single_inheritance A<int, float>;
2114 Note that choosing an inheritance model less general than strictly necessary is
2119 struct __multiple_inheritance S; // error: inheritance model does not match definition
2125 def MSNoVTableDocs : Documentation {
2126 let Category = DocCatType;
2128 This attribute can be added to a class declaration or definition to signal to
2129 the compiler that constructors and destructors will not reference the virtual
2130 function table. It is only supported when using the Microsoft C++ ABI.
2134 def OptnoneDocs : Documentation {
2135 let Category = DocCatFunction;
2137 The ``optnone`` attribute suppresses essentially all optimizations
2138 on a function or method, regardless of the optimization level applied to
2139 the compilation unit as a whole. This is particularly useful when you
2140 need to debug a particular function, but it is infeasible to build the
2141 entire application without optimization. Avoiding optimization on the
2142 specified function can improve the quality of the debugging information
2145 This attribute is incompatible with the ``always_inline`` and ``minsize``
2150 def LoopHintDocs : Documentation {
2151 let Category = DocCatStmt;
2152 let Heading = "#pragma clang loop";
2154 The ``#pragma clang loop`` directive allows loop optimization hints to be
2155 specified for the subsequent loop. The directive allows vectorization,
2156 interleaving, and unrolling to be enabled or disabled. Vector width as well
2157 as interleave and unrolling count can be manually specified. See
2158 `language extensions
2159 <http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
2164 def UnrollHintDocs : Documentation {
2165 let Category = DocCatStmt;
2166 let Heading = "#pragma unroll, #pragma nounroll";
2168 Loop unrolling optimization hints can be specified with ``#pragma unroll`` and
2169 ``#pragma nounroll``. The pragma is placed immediately before a for, while,
2170 do-while, or c++11 range-based for loop.
2172 Specifying ``#pragma unroll`` without a parameter directs the loop unroller to
2173 attempt to fully unroll the loop if the trip count is known at compile time and
2174 attempt to partially unroll the loop if the trip count is not known at compile
2184 Specifying the optional parameter, ``#pragma unroll _value_``, directs the
2185 unroller to unroll the loop ``_value_`` times. The parameter may optionally be
2186 enclosed in parentheses:
2200 Specifying ``#pragma nounroll`` indicates that the loop should not be unrolled:
2209 ``#pragma unroll`` and ``#pragma unroll _value_`` have identical semantics to
2210 ``#pragma clang loop unroll(full)`` and
2211 ``#pragma clang loop unroll_count(_value_)`` respectively. ``#pragma nounroll``
2212 is equivalent to ``#pragma clang loop unroll(disable)``. See
2213 `language extensions
2214 <http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
2215 for further details including limitations of the unroll hints.
2219 def OpenCLUnrollHintDocs : Documentation {
2220 let Category = DocCatStmt;
2221 let Heading = "__attribute__((opencl_unroll_hint))";
2223 The opencl_unroll_hint attribute qualifier can be used to specify that a loop
2224 (for, while and do loops) can be unrolled. This attribute qualifier can be
2225 used to specify full unrolling or partial unrolling by a specified amount.
2226 This is a compiler hint and the compiler may ignore this directive. See
2227 `OpenCL v2.0 <https://www.khronos.org/registry/cl/specs/opencl-2.0.pdf>`_
2228 s6.11.5 for details.
2232 def OpenCLIntelReqdSubGroupSizeDocs : Documentation {
2233 let Category = DocCatStmt;
2234 let Heading = "__attribute__((intel_reqd_sub_group_size))";
2236 The optional attribute intel_reqd_sub_group_size can be used to indicate that
2237 the kernel must be compiled and executed with the specified subgroup size. When
2238 this attribute is present, get_max_sub_group_size() is guaranteed to return the
2239 specified integer value. This is important for the correctness of many subgroup
2240 algorithms, and in some cases may be used by the compiler to generate more optimal
2241 code. See `cl_intel_required_subgroup_size
2242 <https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_required_subgroup_size.txt>`
2247 def OpenCLAccessDocs : Documentation {
2248 let Category = DocCatStmt;
2249 let Heading = "__read_only, __write_only, __read_write (read_only, write_only, read_write)";
2251 The access qualifiers must be used with image object arguments or pipe arguments
2252 to declare if they are being read or written by a kernel or function.
2254 The read_only/__read_only, write_only/__write_only and read_write/__read_write
2255 names are reserved for use as access qualifiers and shall not be used otherwise.
2260 foo (read_only image2d_t imageA,
2261 write_only image2d_t imageB) {
2265 In the above example imageA is a read-only 2D image object, and imageB is a
2266 write-only 2D image object.
2268 The read_write (or __read_write) qualifier can not be used with pipe.
2270 More details can be found in the OpenCL C language Spec v2.0, Section 6.6.
2274 def DocOpenCLAddressSpaces : DocumentationCategory<"OpenCL Address Spaces"> {
2276 The address space qualifier may be used to specify the region of memory that is
2277 used to allocate the object. OpenCL supports the following address spaces:
2278 __generic(generic), __global(global), __local(local), __private(private),
2279 __constant(constant).
2283 __constant int c = ...;
2285 __generic int* foo(global int* g) {
2292 More details can be found in the OpenCL C language Spec v2.0, Section 6.5.
2296 def OpenCLAddressSpaceGenericDocs : Documentation {
2297 let Category = DocOpenCLAddressSpaces;
2299 The generic address space attribute is only available with OpenCL v2.0 and later.
2300 It can be used with pointer types. Variables in global and local scope and
2301 function parameters in non-kernel functions can have the generic address space
2302 type attribute. It is intended to be a placeholder for any other address space
2303 except for '__constant' in OpenCL code which can be used with multiple address
2308 def OpenCLAddressSpaceConstantDocs : Documentation {
2309 let Category = DocOpenCLAddressSpaces;
2311 The constant address space attribute signals that an object is located in
2312 a constant (non-modifiable) memory region. It is available to all work items.
2313 Any type can be annotated with the constant address space attribute. Objects
2314 with the constant address space qualifier can be declared in any scope and must
2315 have an initializer.
2319 def OpenCLAddressSpaceGlobalDocs : Documentation {
2320 let Category = DocOpenCLAddressSpaces;
2322 The global address space attribute specifies that an object is allocated in
2323 global memory, which is accessible by all work items. The content stored in this
2324 memory area persists between kernel executions. Pointer types to the global
2325 address space are allowed as function parameters or local variables. Starting
2326 with OpenCL v2.0, the global address space can be used with global (program
2327 scope) variables and static local variable as well.
2331 def OpenCLAddressSpaceLocalDocs : Documentation {
2332 let Category = DocOpenCLAddressSpaces;
2334 The local address space specifies that an object is allocated in the local (work
2335 group) memory area, which is accessible to all work items in the same work
2336 group. The content stored in this memory region is not accessible after
2337 the kernel execution ends. In a kernel function scope, any variable can be in
2338 the local address space. In other scopes, only pointer types to the local address
2339 space are allowed. Local address space variables cannot have an initializer.
2343 def OpenCLAddressSpacePrivateDocs : Documentation {
2344 let Category = DocOpenCLAddressSpaces;
2346 The private address space specifies that an object is allocated in the private
2347 (work item) memory. Other work items cannot access the same memory area and its
2348 content is destroyed after work item execution ends. Local variables can be
2349 declared in the private address space. Function arguments are always in the
2350 private address space. Kernel function arguments of a pointer or an array type
2351 cannot point to the private address space.
2355 def OpenCLNoSVMDocs : Documentation {
2356 let Category = DocCatVariable;
2358 OpenCL 2.0 supports the optional ``__attribute__((nosvm))`` qualifier for
2359 pointer variable. It informs the compiler that the pointer does not refer
2360 to a shared virtual memory region. See OpenCL v2.0 s6.7.2 for details.
2362 Since it is not widely used and has been removed from OpenCL 2.1, it is ignored
2366 def NullabilityDocs : DocumentationCategory<"Nullability Attributes"> {
2368 Whether a particular pointer may be "null" is an important concern when working with pointers in the C family of languages. The various nullability attributes indicate whether a particular pointer can be null or not, which makes APIs more expressive and can help static analysis tools identify bugs involving null pointers. Clang supports several kinds of nullability attributes: the ``nonnull`` and ``returns_nonnull`` attributes indicate which function or method parameters and result types can never be null, while nullability type qualifiers indicate which pointer types can be null (``_Nullable``) or cannot be null (``_Nonnull``).
2370 The nullability (type) qualifiers express whether a value of a given pointer type can be null (the ``_Nullable`` qualifier), doesn't have a defined meaning for null (the ``_Nonnull`` qualifier), or for which the purpose of null is unclear (the ``_Null_unspecified`` qualifier). Because nullability qualifiers are expressed within the type system, they are more general than the ``nonnull`` and ``returns_nonnull`` attributes, allowing one to express (for example) a nullable pointer to an array of nonnull pointers. Nullability qualifiers are written to the right of the pointer to which they apply. For example:
2374 // No meaningful result when 'ptr' is null (here, it happens to be undefined behavior).
2375 int fetch(int * _Nonnull ptr) { return *ptr; }
2377 // 'ptr' may be null.
2378 int fetch_or_zero(int * _Nullable ptr) {
2379 return ptr ? *ptr : 0;
2382 // A nullable pointer to non-null pointers to const characters.
2383 const char *join_strings(const char * _Nonnull * _Nullable strings, unsigned n);
2385 In Objective-C, there is an alternate spelling for the nullability qualifiers that can be used in Objective-C methods and properties using context-sensitive, non-underscored keywords. For example:
2387 .. code-block:: objective-c
2389 @interface NSView : NSResponder
2390 - (nullable NSView *)ancestorSharedWithView:(nonnull NSView *)aView;
2391 @property (assign, nullable) NSView *superview;
2392 @property (readonly, nonnull) NSArray *subviews;
2397 def TypeNonNullDocs : Documentation {
2398 let Category = NullabilityDocs;
2400 The ``_Nonnull`` nullability qualifier indicates that null is not a meaningful value for a value of the ``_Nonnull`` pointer type. For example, given a declaration such as:
2404 int fetch(int * _Nonnull ptr);
2406 a caller of ``fetch`` should not provide a null value, and the compiler will produce a warning if it sees a literal null value passed to ``fetch``. Note that, unlike the declaration attribute ``nonnull``, the presence of ``_Nonnull`` does not imply that passing null is undefined behavior: ``fetch`` is free to consider null undefined behavior or (perhaps for backward-compatibility reasons) defensively handle null.
2410 def TypeNullableDocs : Documentation {
2411 let Category = NullabilityDocs;
2413 The ``_Nullable`` nullability qualifier indicates that a value of the ``_Nullable`` pointer type can be null. For example, given:
2417 int fetch_or_zero(int * _Nullable ptr);
2419 a caller of ``fetch_or_zero`` can provide null.
2423 def TypeNullUnspecifiedDocs : Documentation {
2424 let Category = NullabilityDocs;
2426 The ``_Null_unspecified`` nullability qualifier indicates that neither the ``_Nonnull`` nor ``_Nullable`` qualifiers make sense for a particular pointer type. It is used primarily to indicate that the role of null with specific pointers in a nullability-annotated header is unclear, e.g., due to overly-complex implementations or historical factors with a long-lived API.
2430 def NonNullDocs : Documentation {
2431 let Category = NullabilityDocs;
2433 The ``nonnull`` attribute indicates that some function parameters must not be null, and can be used in several different ways. It's original usage (`from GCC <https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#Common-Function-Attributes>`_) is as a function (or Objective-C method) attribute that specifies which parameters of the function are nonnull in a comma-separated list. For example:
2437 extern void * my_memcpy (void *dest, const void *src, size_t len)
2438 __attribute__((nonnull (1, 2)));
2440 Here, the ``nonnull`` attribute indicates that parameters 1 and 2
2441 cannot have a null value. Omitting the parenthesized list of parameter indices means that all parameters of pointer type cannot be null:
2445 extern void * my_memcpy (void *dest, const void *src, size_t len)
2446 __attribute__((nonnull));
2448 Clang also allows the ``nonnull`` attribute to be placed directly on a function (or Objective-C method) parameter, eliminating the need to specify the parameter index ahead of type. For example:
2452 extern void * my_memcpy (void *dest __attribute__((nonnull)),
2453 const void *src __attribute__((nonnull)), size_t len);
2455 Note that the ``nonnull`` attribute indicates that passing null to a non-null parameter is undefined behavior, which the optimizer may take advantage of to, e.g., remove null checks. The ``_Nonnull`` type qualifier indicates that a pointer cannot be null in a more general manner (because it is part of the type system) and does not imply undefined behavior, making it more widely applicable.
2459 def ReturnsNonNullDocs : Documentation {
2460 let Category = NullabilityDocs;
2462 The ``returns_nonnull`` attribute indicates that a particular function (or Objective-C method) always returns a non-null pointer. For example, a particular system ``malloc`` might be defined to terminate a process when memory is not available rather than returning a null pointer:
2466 extern void * malloc (size_t size) __attribute__((returns_nonnull));
2468 The ``returns_nonnull`` attribute implies that returning a null pointer is undefined behavior, which the optimizer may take advantage of. The ``_Nonnull`` type qualifier indicates that a pointer cannot be null in a more general manner (because it is part of the type system) and does not imply undefined behavior, making it more widely applicable
2472 def NoAliasDocs : Documentation {
2473 let Category = DocCatFunction;
2475 The ``noalias`` attribute indicates that the only memory accesses inside
2476 function are loads and stores from objects pointed to by its pointer-typed
2477 arguments, with arbitrary offsets.
2481 def OMPDeclareSimdDocs : Documentation {
2482 let Category = DocCatFunction;
2483 let Heading = "#pragma omp declare simd";
2485 The `declare simd` construct can be applied to a function to enable the creation
2486 of one or more versions that can process multiple arguments using SIMD
2487 instructions from a single invocation in a SIMD loop. The `declare simd`
2488 directive is a declarative directive. There may be multiple `declare simd`
2489 directives for a function. The use of a `declare simd` construct on a function
2490 enables the creation of SIMD versions of the associated function that can be
2491 used to process multiple arguments from a single invocation from a SIMD loop
2493 The syntax of the `declare simd` construct is as follows:
2497 #pragma omp declare simd [clause[[,] clause] ...] new-line
2498 [#pragma omp declare simd [clause[[,] clause] ...] new-line]
2500 function definition or declaration
2502 where clause is one of the following:
2507 linear(argument-list[:constant-linear-step])
2508 aligned(argument-list[:alignment])
2509 uniform(argument-list)
2516 def OMPDeclareTargetDocs : Documentation {
2517 let Category = DocCatFunction;
2518 let Heading = "#pragma omp declare target";
2520 The `declare target` directive specifies that variables and functions are mapped
2521 to a device for OpenMP offload mechanism.
2523 The syntax of the declare target directive is as follows:
2527 #pragma omp declare target new-line
2528 declarations-definition-seq
2529 #pragma omp end declare target new-line
2533 def NotTailCalledDocs : Documentation {
2534 let Category = DocCatFunction;
2536 The ``not_tail_called`` attribute prevents tail-call optimization on statically bound calls. It has no effect on indirect calls. Virtual functions, objective-c methods, and functions marked as ``always_inline`` cannot be marked as ``not_tail_called``.
2538 For example, it prevents tail-call optimization in the following case:
2542 int __attribute__((not_tail_called)) foo1(int);
2545 return foo1(a); // No tail-call optimization on direct calls.
2548 However, it doesn't prevent tail-call optimization in this case:
2552 int __attribute__((not_tail_called)) foo1(int);
2555 int (*fn)(int) = &foo1;
2557 // not_tail_called has no effect on an indirect call even if the call can be
2558 // resolved at compile time.
2562 Marking virtual functions as ``not_tail_called`` is an error:
2568 // not_tail_called on a virtual function is an error.
2569 [[clang::not_tail_called]] virtual int foo1();
2573 // Non-virtual functions can be marked ``not_tail_called``.
2574 [[clang::not_tail_called]] int foo3();
2577 class Derived1 : public Base {
2579 int foo1() override;
2581 // not_tail_called on a virtual function is an error.
2582 [[clang::not_tail_called]] int foo2() override;
2587 def InternalLinkageDocs : Documentation {
2588 let Category = DocCatFunction;
2590 The ``internal_linkage`` attribute changes the linkage type of the declaration to internal.
2591 This is similar to C-style ``static``, but can be used on classes and class methods. When applied to a class definition,
2592 this attribute affects all methods and static data members of that class.
2593 This can be used to contain the ABI of a C++ library by excluding unwanted class methods from the export tables.
2597 def DisableTailCallsDocs : Documentation {
2598 let Category = DocCatFunction;
2600 The ``disable_tail_calls`` attribute instructs the backend to not perform tail call optimization inside the marked function.
2608 int foo(int a) __attribute__((disable_tail_calls)) {
2609 return callee(a); // This call is not tail-call optimized.
2612 Marking virtual functions as ``disable_tail_calls`` is legal.
2620 [[clang::disable_tail_calls]] virtual int foo1() {
2621 return callee(); // This call is not tail-call optimized.
2625 class Derived1 : public Base {
2627 int foo1() override {
2628 return callee(); // This call is tail-call optimized.
2635 def AnyX86InterruptDocs : Documentation {
2636 let Category = DocCatFunction;
2638 Clang supports the GNU style ``__attribute__((interrupt))`` attribute on
2639 x86/x86-64 targets.The compiler generates function entry and exit sequences
2640 suitable for use in an interrupt handler when this attribute is present.
2641 The 'IRET' instruction, instead of the 'RET' instruction, is used to return
2642 from interrupt or exception handlers. All registers, except for the EFLAGS
2643 register which is restored by the 'IRET' instruction, are preserved by the
2646 Any interruptible-without-stack-switch code must be compiled with
2647 -mno-red-zone since interrupt handlers can and will, because of the
2648 hardware design, touch the red zone.
2650 1. interrupt handler must be declared with a mandatory pointer argument:
2654 struct interrupt_frame
2663 __attribute__ ((interrupt))
2664 void f (struct interrupt_frame *frame) {
2668 2. exception handler:
2670 The exception handler is very similar to the interrupt handler with
2671 a different mandatory function signature:
2675 __attribute__ ((interrupt))
2676 void f (struct interrupt_frame *frame, uword_t error_code) {
2680 and compiler pops 'ERROR_CODE' off stack before the 'IRET' instruction.
2682 The exception handler should only be used for exceptions which push an
2683 error code and all other exceptions must use the interrupt handler.
2684 The system will crash if the wrong handler is used.
2688 def AnyX86NoCallerSavedRegistersDocs : Documentation {
2689 let Category = DocCatFunction;
2691 Use this attribute to indicate that the specified function has no
2692 caller-saved registers. That is, all registers are callee-saved except for
2693 registers used for passing parameters to the function or returning parameters
2695 The compiler saves and restores any modified registers that were not used for
2696 passing or returning arguments to the function.
2698 The user can call functions specified with the 'no_caller_saved_registers'
2699 attribute from an interrupt handler without saving and restoring all
2700 call-clobbered registers.
2702 Note that 'no_caller_saved_registers' attribute is not a calling convention.
2703 In fact, it only overrides the decision of which registers should be saved by
2704 the caller, but not how the parameters are passed from the caller to the callee.
2710 __attribute__ ((no_caller_saved_registers, fastcall))
2711 void f (int arg1, int arg2) {
2715 In this case parameters 'arg1' and 'arg2' will be passed in registers.
2716 In this case, on 32-bit x86 targets, the function 'f' will use ECX and EDX as
2717 register parameters. However, it will not assume any scratch registers and
2718 should save and restore any modified registers except for ECX and EDX.
2722 def SwiftCallDocs : Documentation {
2723 let Category = DocCatVariable;
2725 The ``swiftcall`` attribute indicates that a function should be called
2726 using the Swift calling convention for a function or function pointer.
2728 The lowering for the Swift calling convention, as described by the Swift
2729 ABI documentation, occurs in multiple phases. The first, "high-level"
2730 phase breaks down the formal parameters and results into innately direct
2731 and indirect components, adds implicit paraameters for the generic
2732 signature, and assigns the context and error ABI treatments to parameters
2733 where applicable. The second phase breaks down the direct parameters
2734 and results from the first phase and assigns them to registers or the
2735 stack. The ``swiftcall`` convention only handles this second phase of
2736 lowering; the C function type must accurately reflect the results
2737 of the first phase, as follows:
2739 - Results classified as indirect by high-level lowering should be
2740 represented as parameters with the ``swift_indirect_result`` attribute.
2742 - Results classified as direct by high-level lowering should be represented
2745 - First, remove any empty direct results.
2747 - If there are no direct results, the C result type should be ``void``.
2749 - If there is one direct result, the C result type should be a type with
2750 the exact layout of that result type.
2752 - If there are a multiple direct results, the C result type should be
2753 a struct type with the exact layout of a tuple of those results.
2755 - Parameters classified as indirect by high-level lowering should be
2756 represented as parameters of pointer type.
2758 - Parameters classified as direct by high-level lowering should be
2759 omitted if they are empty types; otherwise, they should be represented
2760 as a parameter type with a layout exactly matching the layout of the
2761 Swift parameter type.
2763 - The context parameter, if present, should be represented as a trailing
2764 parameter with the ``swift_context`` attribute.
2766 - The error result parameter, if present, should be represented as a
2767 trailing parameter (always following a context parameter) with the
2768 ``swift_error_result`` attribute.
2770 ``swiftcall`` does not support variadic arguments or unprototyped functions.
2772 The parameter ABI treatment attributes are aspects of the function type.
2773 A function type which which applies an ABI treatment attribute to a
2774 parameter is a different type from an otherwise-identical function type
2775 that does not. A single parameter may not have multiple ABI treatment
2778 Support for this feature is target-dependent, although it should be
2779 supported on every target that Swift supports. Query for this support
2780 with ``__has_attribute(swiftcall)``. This implies support for the
2781 ``swift_context``, ``swift_error_result``, and ``swift_indirect_result``
2786 def SwiftContextDocs : Documentation {
2787 let Category = DocCatVariable;
2789 The ``swift_context`` attribute marks a parameter of a ``swiftcall``
2790 function as having the special context-parameter ABI treatment.
2792 This treatment generally passes the context value in a special register
2793 which is normally callee-preserved.
2795 A ``swift_context`` parameter must either be the last parameter or must be
2796 followed by a ``swift_error_result`` parameter (which itself must always be
2797 the last parameter).
2799 A context parameter must have pointer or reference type.
2803 def SwiftErrorResultDocs : Documentation {
2804 let Category = DocCatVariable;
2806 The ``swift_error_result`` attribute marks a parameter of a ``swiftcall``
2807 function as having the special error-result ABI treatment.
2809 This treatment generally passes the underlying error value in and out of
2810 the function through a special register which is normally callee-preserved.
2811 This is modeled in C by pretending that the register is addressable memory:
2813 - The caller appears to pass the address of a variable of pointer type.
2814 The current value of this variable is copied into the register before
2815 the call; if the call returns normally, the value is copied back into the
2818 - The callee appears to receive the address of a variable. This address
2819 is actually a hidden location in its own stack, initialized with the
2820 value of the register upon entry. When the function returns normally,
2821 the value in that hidden location is written back to the register.
2823 A ``swift_error_result`` parameter must be the last parameter, and it must be
2824 preceded by a ``swift_context`` parameter.
2826 A ``swift_error_result`` parameter must have type ``T**`` or ``T*&`` for some
2827 type T. Note that no qualifiers are permitted on the intermediate level.
2829 It is undefined behavior if the caller does not pass a pointer or
2830 reference to a valid object.
2832 The standard convention is that the error value itself (that is, the
2833 value stored in the apparent argument) will be null upon function entry,
2834 but this is not enforced by the ABI.
2838 def SwiftIndirectResultDocs : Documentation {
2839 let Category = DocCatVariable;
2841 The ``swift_indirect_result`` attribute marks a parameter of a ``swiftcall``
2842 function as having the special indirect-result ABI treatment.
2844 This treatment gives the parameter the target's normal indirect-result
2845 ABI treatment, which may involve passing it differently from an ordinary
2846 parameter. However, only the first indirect result will receive this
2847 treatment. Furthermore, low-level lowering may decide that a direct result
2848 must be returned indirectly; if so, this will take priority over the
2849 ``swift_indirect_result`` parameters.
2851 A ``swift_indirect_result`` parameter must either be the first parameter or
2852 follow another ``swift_indirect_result`` parameter.
2854 A ``swift_indirect_result`` parameter must have type ``T*`` or ``T&`` for
2855 some object type ``T``. If ``T`` is a complete type at the point of
2856 definition of a function, it is undefined behavior if the argument
2857 value does not point to storage of adequate size and alignment for a
2858 value of type ``T``.
2860 Making indirect results explicit in the signature allows C functions to
2861 directly construct objects into them without relying on language
2862 optimizations like C++'s named return value optimization (NRVO).
2866 def SuppressDocs : Documentation {
2867 let Category = DocCatStmt;
2869 The ``[[gsl::suppress]]`` attribute suppresses specific
2870 clang-tidy diagnostics for rules of the `C++ Core Guidelines`_ in a portable
2871 way. The attribute can be attached to declarations, statements, and at
2876 [[gsl::suppress("Rh-public")]]
2879 [[gsl::suppress("type")]] {
2880 p = reinterpret_cast<int*>(7);
2884 [[clang::suppress("type", "bounds")]];
2888 .. _`C++ Core Guidelines`: https://github.com/isocpp/CppCoreGuidelines/blob/master/CppCoreGuidelines.md#inforce-enforcement
2892 def AbiTagsDocs : Documentation {
2893 let Category = DocCatFunction;
2895 The ``abi_tag`` attribute can be applied to a function, variable, class or
2896 inline namespace declaration to modify the mangled name of the entity. It gives
2897 the ability to distinguish between different versions of the same entity but
2898 with different ABI versions supported. For example, a newer version of a class
2899 could have a different set of data members and thus have a different size. Using
2900 the ``abi_tag`` attribute, it is possible to have different mangled names for
2901 a global variable of the class type. Therefor, the old code could keep using
2902 the old manged name and the new code will use the new mangled name with tags.
2906 def PreserveMostDocs : Documentation {
2907 let Category = DocCatCallingConvs;
2909 On X86-64 and AArch64 targets, this attribute changes the calling convention of
2910 a function. The ``preserve_most`` calling convention attempts to make the code
2911 in the caller as unintrusive as possible. This convention behaves identically
2912 to the ``C`` calling convention on how arguments and return values are passed,
2913 but it uses a different set of caller/callee-saved registers. This alleviates
2914 the burden of saving and recovering a large register set before and after the
2915 call in the caller. If the arguments are passed in callee-saved registers,
2916 then they will be preserved by the callee across the call. This doesn't
2917 apply for values returned in callee-saved registers.
2919 - On X86-64 the callee preserves all general purpose registers, except for
2920 R11. R11 can be used as a scratch register. Floating-point registers
2921 (XMMs/YMMs) are not preserved and need to be saved by the caller.
2923 The idea behind this convention is to support calls to runtime functions
2924 that have a hot path and a cold path. The hot path is usually a small piece
2925 of code that doesn't use many registers. The cold path might need to call out to
2926 another function and therefore only needs to preserve the caller-saved
2927 registers, which haven't already been saved by the caller. The
2928 `preserve_most` calling convention is very similar to the ``cold`` calling
2929 convention in terms of caller/callee-saved registers, but they are used for
2930 different types of function calls. ``coldcc`` is for function calls that are
2931 rarely executed, whereas `preserve_most` function calls are intended to be
2932 on the hot path and definitely executed a lot. Furthermore ``preserve_most``
2933 doesn't prevent the inliner from inlining the function call.
2935 This calling convention will be used by a future version of the Objective-C
2936 runtime and should therefore still be considered experimental at this time.
2937 Although this convention was created to optimize certain runtime calls to
2938 the Objective-C runtime, it is not limited to this runtime and might be used
2939 by other runtimes in the future too. The current implementation only
2940 supports X86-64 and AArch64, but the intention is to support more architectures
2945 def PreserveAllDocs : Documentation {
2946 let Category = DocCatCallingConvs;
2948 On X86-64 and AArch64 targets, this attribute changes the calling convention of
2949 a function. The ``preserve_all`` calling convention attempts to make the code
2950 in the caller even less intrusive than the ``preserve_most`` calling convention.
2951 This calling convention also behaves identical to the ``C`` calling convention
2952 on how arguments and return values are passed, but it uses a different set of
2953 caller/callee-saved registers. This removes the burden of saving and
2954 recovering a large register set before and after the call in the caller. If
2955 the arguments are passed in callee-saved registers, then they will be
2956 preserved by the callee across the call. This doesn't apply for values
2957 returned in callee-saved registers.
2959 - On X86-64 the callee preserves all general purpose registers, except for
2960 R11. R11 can be used as a scratch register. Furthermore it also preserves
2961 all floating-point registers (XMMs/YMMs).
2963 The idea behind this convention is to support calls to runtime functions
2964 that don't need to call out to any other functions.
2966 This calling convention, like the ``preserve_most`` calling convention, will be
2967 used by a future version of the Objective-C runtime and should be considered
2968 experimental at this time.
2972 def DeprecatedDocs : Documentation {
2973 let Category = DocCatFunction;
2975 The ``deprecated`` attribute can be applied to a function, a variable, or a
2976 type. This is useful when identifying functions, variables, or types that are
2977 expected to be removed in a future version of a program.
2979 Consider the function declaration for a hypothetical function ``f``:
2983 void f(void) __attribute__((deprecated("message", "replacement")));
2985 When spelled as `__attribute__((deprecated))`, the deprecated attribute can have
2986 two optional string arguments. The first one is the message to display when
2987 emitting the warning; the second one enables the compiler to provide a Fix-It
2988 to replace the deprecated name with a new name. Otherwise, when spelled as
2989 `[[gnu::deprecated]] or [[deprecated]]`, the attribute can have one optional
2990 string argument which is the message to display when emitting the warning.
2994 def IFuncDocs : Documentation {
2995 let Category = DocCatFunction;
2997 ``__attribute__((ifunc("resolver")))`` is used to mark that the address of a declaration should be resolved at runtime by calling a resolver function.
2999 The symbol name of the resolver function is given in quotes. A function with this name (after mangling) must be defined in the current translation unit; it may be ``static``. The resolver function should take no arguments and return a pointer.
3001 The ``ifunc`` attribute may only be used on a function declaration. A function declaration with an ``ifunc`` attribute is considered to be a definition of the declared entity. The entity must not have weak linkage; for example, in C++, it cannot be applied to a declaration if a definition at that location would be considered inline.
3003 Not all targets support this attribute. ELF targets support this attribute when using binutils v2.20.1 or higher and glibc v2.11.1 or higher. Non-ELF targets currently do not support this attribute.
3007 def LTOVisibilityDocs : Documentation {
3008 let Category = DocCatType;
3010 See :doc:`LTOVisibility`.
3014 def RenderScriptKernelAttributeDocs : Documentation {
3015 let Category = DocCatFunction;
3017 ``__attribute__((kernel))`` is used to mark a ``kernel`` function in
3020 In RenderScript, ``kernel`` functions are used to express data-parallel
3021 computations. The RenderScript runtime efficiently parallelizes ``kernel``
3022 functions to run on computational resources such as multi-core CPUs and GPUs.
3023 See the RenderScript_ documentation for more information.
3025 .. _RenderScript: https://developer.android.com/guide/topics/renderscript/compute.html
3029 def XRayDocs : Documentation {
3030 let Category = DocCatFunction;
3031 let Heading = "xray_always_instrument (clang::xray_always_instrument), xray_never_instrument (clang::xray_never_instrument), xray_log_args (clang::xray_log_args)";
3033 ``__attribute__((xray_always_instrument))`` or ``[[clang::xray_always_instrument]]`` is used to mark member functions (in C++), methods (in Objective C), and free functions (in C, C++, and Objective C) to be instrumented with XRay. This will cause the function to always have space at the beginning and exit points to allow for runtime patching.
3035 Conversely, ``__attribute__((xray_never_instrument))`` or ``[[clang::xray_never_instrument]]`` will inhibit the insertion of these instrumentation points.
3037 If a function has neither of these attributes, they become subject to the XRay heuristics used to determine whether a function should be instrumented or otherwise.
3039 ``__attribute__((xray_log_args(N)))`` or ``[[clang::xray_log_args(N)]]`` is used to preserve N function arguments for the logging function. Currently, only N==1 is supported.
3043 def TransparentUnionDocs : Documentation {
3044 let Category = DocCatType;
3046 This attribute can be applied to a union to change the behaviour of calls to
3047 functions that have an argument with a transparent union type. The compiler
3048 behaviour is changed in the following manner:
3050 - A value whose type is any member of the transparent union can be passed as an
3051 argument without the need to cast that value.
3053 - The argument is passed to the function using the calling convention of the
3054 first member of the transparent union. Consequently, all the members of the
3055 transparent union should have the same calling convention as its first member.
3057 Transparent unions are not supported in C++.
3061 def ObjCSubclassingRestrictedDocs : Documentation {
3062 let Category = DocCatType;
3064 This attribute can be added to an Objective-C ``@interface`` declaration to
3065 ensure that this class cannot be subclassed.