AttrDocs.td

//==--- AttrDocs.td - Attribute documentation ----------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===---------------------------------------------------------------------===//

// To test that the documentation builds cleanly, you must run clang-tblgen to
// convert the .td file into a .md file, and then run sphinx to convert the
// .md file into an HTML file. After completing testing, you should revert the
// generated .md file so that the modified version does not get checked in to
// version control.
//
// To run clang-tblgen to generate the .md file:
// clang-tblgen -gen-attr-docs -I <root>/clang/include
//   <root>/clang/include/clang/Basic/Attr.td -o
//   <root>/clang/docs/AttributeReference.md
//
// To run sphinx to generate the .html files (note that sphinx-build must be
// available on the PATH):
// Windows (from within the clang\docs directory):
//   make.bat html
// Non-Windows (from within the clang/docs directory):
//   sphinx-build -b html . _build/html

def GlobalDocumentation {
  code Intro =[{% -------------------------------------------------------------------
% NOTE: This file is automatically generated by running clang-tblgen
% -gen-attr-docs. Do not edit this file by hand!!
% -------------------------------------------------------------------

# Attributes in Clang


```{program} clang
```

## Introduction

This page lists the attributes currently supported by Clang.
}];
}

def SectionDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `section` attribute allows you to specify a specific section a
global variable or function should be in after translation.
  }];
  let Heading = "section, __declspec(allocate)";
}

def CodeModelDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `model` attribute allows overriding the translation unit's
code model (specified by `-mcmodel`) for a specific global variable.

On LoongArch, allowed values are "normal", "medium", "extreme".

On x86-64, allowed values are `"small"` and `"large"`. `"small"` is
roughly equivalent to `-mcmodel=small`, meaning the global is considered
"small" placed closer to the `.text` section relative to "large" globals, and
to prefer using 32-bit relocations to access the global. `"large"` is roughly
equivalent to `-mcmodel=large`, meaning the global is considered "large" and
placed further from the `.text` section relative to "small" globals, and
64-bit relocations must be used to access the global.
  }];
  let Heading = "model";
}

def UsedDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute, when attached to a function or variable definition, indicates
that there may be references to the entity which are not apparent in the source
code. For example, it may be referenced from inline `asm`, or it may be
found through a dynamic symbol or section lookup.

The compiler must emit the definition even if it appears to be unused, and it
must not apply optimizations which depend on fully understanding how the entity
is used.

Whether this attribute has any effect on the linker depends on the target and
the linker. Most linkers support the feature of section garbage collection
(`--gc-sections`), also known as "dead stripping" (`ld64 -dead_strip`) or
discarding unreferenced sections (`link.exe /OPT:REF`). On COFF and Mach-O
targets (Windows and Apple platforms), the `used` attribute prevents symbols
from being removed by linker section GC. On ELF targets, it has no effect on its
own, and the linker may remove the definition if it is not otherwise referenced.
This linker GC can be avoided by also adding the `retain` attribute. Note
that `retain` requires special support from the linker; see that attribute's
documentation for further information.
  }];
}

def RetainDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute, when attached to a function or variable definition, prevents
section garbage collection in the linker. It does not prevent other discard
mechanisms, such as archive member selection, and COMDAT group resolution.

If the compiler does not emit the definition, e.g. because it was not used in
the translation unit or the compiler was able to eliminate all of the uses,
this attribute has no effect. This attribute is typically combined with the
`used` attribute to force the definition to be emitted and preserved into the
final linked image.

This attribute is only necessary on ELF targets; other targets prevent section
garbage collection by the linker when using the `used` attribute alone.
Using the attributes together should result in consistent behavior across
targets.

This attribute requires the linker to support the `SHF_GNU_RETAIN` extension.
This support is available in GNU `ld` and `gold` as of binutils 2.36, as
well as in `ld.lld` 13.
  }];
}

def InitPriorityDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
In C++, the order in which global variables are initialized across translation
units is unspecified, unlike the ordering within a single translation unit. The
`init_priority` attribute allows you to specify a relative ordering for the
initialization of objects declared at namespace scope in C++ within a single
linked image on supported platforms. The priority is given as an integer constant
expression between 101 and 65535 (inclusive). Priorities outside of that range are
reserved for use by the implementation. A lower value indicates a higher priority
of initialization. Note that only the relative ordering of values is important.
For example:

```c++
struct SomeType { SomeType(); };
__attribute__((init_priority(200))) SomeType Obj1;
__attribute__((init_priority(101))) SomeType Obj2;
```

`Obj2` will be initialized *before* `Obj1` despite the usual order of
initialization being the opposite.

Note that this attribute does not control the initialization order of objects
across final linked image boundaries like shared objects and executables.

On Windows, `init_seg(compiler)` is represented with a priority of 200 and
`init_seg(library)` is represented with a priority of 400. `init_seg(user)`
uses the default 65535 priority.

On MachO platforms, this attribute also does not control the order of initialization
across translation units, where it only affects the order within a single TU.

This attribute is only supported for C++ and Objective-C++ and is ignored in
other language modes.
  }];
}

def InitSegDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The attribute applied by `pragma init_seg()` controls the section into
which global initialization function pointers are emitted. It is only
available with `-fms-extensions`. Typically, this function pointer is
emitted into `.CRT$XCU` on Windows. The user can change the order of
initialization by using a different section name with the same
`.CRT$XC` prefix and a suffix that sorts lexicographically before or
after the standard `.CRT$XCU` sections. See the [init_seg][init_seg]
documentation on MSDN for more information.

[init_seg]: http://msdn.microsoft.com/en-us/library/7977wcck(v=vs.110).aspx
  }];
}

def TLSModelDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `tls_model` attribute allows you to specify which thread-local storage
model to use. It accepts the following strings:

- global-dynamic
- local-dynamic
- initial-exec
- local-exec

TLS models are mutually exclusive.
  }];
}

def AddressSpaceDocs : Documentation {
  let Category = DocCatType;
  let Heading = "address_space";
  let Label = "langext-address_space_documentation";
  let Content = [{
:::{Note}
This attribute is mainly intended to be used by target headers
provided by the toolchain. End users should prefer the documented, named
address space annotations for their platform, such as the
[OpenCL address spaces], `__global__`, `__local__`, or something else.
:::

The `address_space` attribute functions as a type qualifier that allows the
programmer to specify the address space for a pointer or reference type.
Qualified pointer types are considered distinct types for the purposes of
overload resolution. The attribute takes a single, non-negative integer
constant expression identifying the address space. For example:

```c
int * __attribute__((address_space(1))) ptr;

void foo(__attribute__((address_space(2))) float *buf);
```

Only one address space qualifier may be applied to a given pointer or reference
type. Where address spaces are allowed (e.g., variables, parameters, return
types) and what values are valid depends on the target and language mode.

The meaning of each value is defined by the target; multiple address spaces are
used in environments such as OpenCL, CUDA, HIP, and other GPU programming
models to distinguish global, local, constant, and private memory. See for
example the address spaces defined in the [NVPTX Usage Guide][nvptx usage guide] and the
[AMDGPU Usage Guide][amdgpu usage guide].

Address spaces may partially overlap or be entirely distinct. The compiler may
reject attempts to convert between distinct, incompatible address spaces.
Pointer width may vary between different address spaces, so some explicit casts
may truncate.

For more information, refer to [ISO TR18037][iso tr18037], which covers embedded C language
extensions. Section 5 covers named address spaces.

[amdgpu usage guide]: https://llvm.org/docs/AMDGPUUsage.html#address-spaces
[iso tr18037]: https://standards.iso.org/ittf/PubliclyAvailableStandards/c051126_ISO_IEC_TR_18037_2008.zip
[nvptx usage guide]: https://llvm.org/docs/NVPTXUsage.html#address-spaces
  }];
}

def DLLExportDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `__declspec(dllexport)` attribute declares a variable, function, or
Objective-C interface to be exported from the module. It is available under the
`-fdeclspec` flag for compatibility with various compilers. The primary use
is for COFF object files which explicitly specify what interfaces are available
for external use. See the [dllexport][dllexport] documentation on MSDN for more
information.

[dllexport]: https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx
  }];
}

def DLLImportDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `__declspec(dllimport)` attribute declares a variable, function, or
Objective-C interface to be imported from an external module. It is available
under the `-fdeclspec` flag for compatibility with various compilers. The
primary use is for COFF object files which explicitly specify what interfaces
are imported from external modules. See the [dllimport][dllimport] documentation on MSDN
for more information.

Note that a dllimport function may still be inlined, if its definition is
available and it doesn't reference any non-dllimport functions or global
variables.

[dllimport]: https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx
  }];
}

def ThreadDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `__declspec(thread)` attribute declares a variable with thread local
storage. It is available under the `-fms-extensions` flag for MSVC
compatibility. See the documentation for [\_\_declspec(thread)][__declspec(thread)] on MSDN.

In Clang, `__declspec(thread)` is generally equivalent in functionality to the
GNU `__thread` keyword. The variable must not have a destructor and must have
a constant initializer, if any. The attribute only applies to variables
declared with static storage duration, such as globals, class static data
members, and static locals.

[__declspec(thread)]: http://msdn.microsoft.com/en-us/library/9w1sdazb.aspx
  }];
}

def NoEscapeDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
`noescape` placed on a function parameter of a pointer type is used to inform
the compiler that the pointer cannot escape: that is, no reference to the object
the pointer points to that is derived from the parameter value will survive
after the function returns. Users are responsible for making sure parameters
annotated with `noescape` do not actually escape. The optimizer may make
assumptions based on the fact that it knows that a call to the function does
not escape a certain parameter, so incorrectly annotating a parameter with
`noescape` leads to undefined behavior. The callee is also not allowed to
deallocate memory through a `noescape` parameter: the optimizer does not make
assumptions based on this information at the moment, but may do so in the
future. Some cases of invalid uses of `noescape` can be found with
{ref}`-Wlifetime-safety-noescape <Wlifetime-safety-noescape>`.

For example:

```c
int *gp;

void nonescapingFunc(__attribute__((noescape)) int *p) {
  *p += 100; // OK.
}

void escapingFunc(__attribute__((noescape)) int *p) {
  gp = p; // Not OK.
}

void freeingFunc(__attribute__((noescape)) int *p) {
  free(p); // Not OK.
}
```

Since `noescape` is a parameter attribute and not a type attribute, it only
applies to the outermost pointer level, regardless of where in the parameter
declaration you place it:

```c
int **gp;

void nestingEscapes(__attribute__((noescape)) int **p) {
  gp = p; // Not OK.
  *gp = *p; // OK, p does not escape.
}
```

Additionally, when the parameter is a
{doc}`block pointer <BlockLanguageSpec>`, the same restriction applies to
copies of the block. For example:

```c
typedef void (^BlockTy)();
BlockTy g0, g1;

void nonescapingFunc(__attribute__((noescape)) BlockTy block) {
  block(); // OK.
}

void escapingFunc(__attribute__((noescape)) BlockTy block) {
  g0 = block; // Not OK.
  g1 = Block_copy(block); // Not OK either.
}
```

The function *is* allowed to leak information about the memory address of the
pointer, but not any provenance of the allocation:

```c
bool isNull(__attribute__((noescape)) void *p) {
  return !p; // OK.
}

uintptr_t gi;

void escapingAddress(__attribute__((noescape)) int *p) {
  // OK *if and only if* gi is never casted back to a pointer.
  gi = (uintptr_t)p;
}

bool usingEscapedAddress(int *p) {
  return (uintptr_t)p > gi; // OK.
}

bool usingEscapedPointer(int *p) {
  return p > (int*)gi; // Not OK.
}

int *gp;

void escapingEndFunc(__attribute__((noescape)) int *p, size_t len) {
  gp = p + len; // Not OK.
}
```

  }];
}

def WarnUnusedDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `warn_unused` attribute can be placed on the declaration of a structure or union type.
When the `-Wunused-variable` diagnostic is enabled, local variables of types which have a non-trivial constructor or destructor are considered "used" by virtue of the constructor or destructor invocations involved.
Those constructor or destructor invocations are not considered a use if the type is declared with the `warn_unused` attribute.
The variable is considered used if it is named outside of its declaration.

This attribute is available in both C and C++ language modes but is primarily useful in C++ for classes which have a non-trivial constructor or destructor but act as a value type rather than an RAII type.

```c++
struct [[gnu::warn_unused]] S {
  S();
  ~S();
};

struct T {
  T();
  ~T();
 };

 int func() {
   S s1; // -Wunused-variable warning
   S s2; // No -Wunused-variable warning because of the member access expression below
   S s3; // No -Wunused-variable warning because of the sizeof operand below
   T t;  // No -Wunused-variable warning

   s2.~S();
   return sizeof(s3);
 }
```
  }];
}

def MaybeUndefDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `maybe_undef` attribute can be placed on a function parameter. It indicates
that the parameter is allowed to use undef values. It informs the compiler
to insert a freeze LLVM IR instruction on the function parameter.
Please note that this is an attribute that is used as an internal
implementation detail and not intended to be used by external users.

In languages HIP, CUDA etc., some functions have multi-threaded semantics and
it is enough for only one or some threads to provide defined arguments.
Depending on semantics, undef arguments in some threads don't produce
undefined results in the function call. Since, these functions accept undefined
arguments, `maybe_undef` attribute can be placed.

Sample usage:

```c
void maybeundeffunc(int __attribute__((maybe_undef))param);
```
  }];
}

def CPUSpecificCPUDispatchDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `cpu_specific` and `cpu_dispatch` attributes are used to define and
resolve multiversioned functions. This form of multiversioning provides a
mechanism for declaring versions across translation units and manually
specifying the resolved function list. A specified CPU defines a set of minimum
features that are required for the function to be called. The result of this is
that future processors execute the most restrictive version of the function the
new processor can execute.

In addition, unlike the ICC implementation of this feature, the selection of the
version does not consider the manufacturer or microarchitecture of the processor.
It tests solely the list of features that are both supported by the specified
processor and present in the compiler-rt library. This can be surprising at times,
as the runtime processor may be from a completely different manufacturer, as long
as it supports the same feature set.

This can additionally be surprising, as some processors are indistringuishable from
others based on the list of testable features. When this happens, the variant
is selected in an unspecified manner.

Function versions are defined with `cpu_specific`, which takes one or more CPU
names as a parameter. For example:

```c
// Declares and defines the ivybridge version of single_cpu.
__attribute__((cpu_specific(ivybridge)))
void single_cpu(void){}

// Declares and defines the atom version of single_cpu.
__attribute__((cpu_specific(atom)))
void single_cpu(void){}

// Declares and defines both the ivybridge and atom version of multi_cpu.
__attribute__((cpu_specific(ivybridge, atom)))
void multi_cpu(void){}
```

A dispatching (or resolving) function can be declared anywhere in a project's
source code with `cpu_dispatch`. This attribute takes one or more CPU names
as a parameter (like `cpu_specific`). Functions marked with `cpu_dispatch`
are not expected to be defined, only declared. If such a marked function has a
definition, any side effects of the function are ignored; trivial function
bodies are permissible for ICC compatibility.

```c
// Creates a resolver for single_cpu above.
__attribute__((cpu_dispatch(ivybridge, atom)))
void single_cpu(void){}

// Creates a resolver for multi_cpu, but adds a 3rd version defined in another
// translation unit.
__attribute__((cpu_dispatch(ivybridge, atom, sandybridge)))
void multi_cpu(void){}
```

Note that it is possible to have a resolving function that dispatches based on
more or fewer options than are present in the program. Specifying fewer will
result in the omitted options not being considered during resolution. Specifying
a version for resolution that isn't defined in the program will result in a
linking failure.

It is also possible to specify a CPU name of `generic` which will be resolved
if the executing processor doesn't satisfy the features required in the CPU
name. The behavior of a program executing on a processor that doesn't satisfy
any option of a multiversioned function is undefined.
  }];
}

def SYCLKernelDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "sycl_kernel";
  let Content = [{
The `sycl_kernel` attribute specifies that a function template will be used
to outline device code and to generate an OpenCL kernel.
Here is a code example of the SYCL program, which demonstrates the compiler's
outlining job:

```c++
int foo(int x) { return ++x; }

using namespace cl::sycl;
queue Q;
buffer<int, 1> a(range<1>{1024});
Q.submit([&](handler& cgh) {
  auto A = a.get_access<access::mode::write>(cgh);
  cgh.parallel_for<init_a>(range<1>{1024}, [=](id<1> index) {
    A[index] = index[0] + foo(42);
  });
}
```

A C++ function object passed to the `parallel_for` is called a "SYCL kernel".
A SYCL kernel defines the entry point to the "device part" of the code. The
compiler will emit all symbols accessible from a "kernel". In this code
example, the compiler will emit "foo" function. More details about the
compilation of functions for the device part can be found in the SYCL 1.2.1
specification Section 6.4.
To show to the compiler entry point to the "device part" of the code, the SYCL
runtime can use the `sycl_kernel` attribute in the following way:

```c++
namespace cl {
namespace sycl {
class handler {
  template <typename KernelName, typename KernelType/*, ...*/>
  __attribute__((sycl_kernel)) void sycl_kernel_function(KernelType KernelFuncObj) {
    // ...
    KernelFuncObj();
  }

  template <typename KernelName, typename KernelType, int Dims>
  void parallel_for(range<Dims> NumWorkItems, KernelType KernelFunc) {
#ifdef __SYCL_DEVICE_ONLY__
    sycl_kernel_function<KernelName, KernelType, Dims>(KernelFunc);
#else
    // Host implementation
#endif
  }
};
} // namespace sycl
} // namespace cl
```

The compiler will also generate an OpenCL kernel using the function marked with
the `sycl_kernel` attribute.
Here is the list of SYCL device compiler expectations with regard to the
function marked with the `sycl_kernel` attribute:

- The function must be a template with at least two type template parameters.
  The compiler generates an OpenCL kernel and uses the first template parameter
  as a unique name for the generated OpenCL kernel. The host application uses
  this unique name to invoke the OpenCL kernel generated for the SYCL kernel
  specialized by this name and second template parameter `KernelType` (which
  might be an unnamed function object type).
- The function must have at least one parameter. The first parameter is
  required to be a function object type (named or unnamed i.e. lambda). The
  compiler uses function object type fields to generate OpenCL kernel
  parameters.
- The function must return void. The compiler reuses the body of marked functions to
  generate the OpenCL kernel body, and the OpenCL kernel must return `void`.

The SYCL kernel in the previous code sample meets these expectations.
  }];
}

def DeviceKernelDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "device_kernel, nvptx_kernel, amdgpu_kernel, "
                "kernel, __kernel";
  let Content = [{
These attributes specify that the function represents a kernel for device offloading.
The specific semantics depend on the offloading language, target, and attribute spelling.
Here is a code example using the attribute to mark a function as a kernel:

```c++
[[clang::device_kernel]] int foo(int x) { return ++x; }
```
  }];
}

def SYCLExternalDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "sycl_external";
  let Content = [{
The `sycl_external` attribute indicates that a function defined in another
translation unit may be called by a device function defined in the current
translation unit or, if defined in the current translation unit, the function
may be called by device functions defined in other translation units.
The attribute is intended for use in the implementation of the `SYCL_EXTERNAL`
macro as specified in section 5.10.1, "SYCL functions and member functions
linkage", of the SYCL 2020 specification.

The attribute only appertains to functions and only those that meet the
following requirements:

- Has external linkage
- Is not explicitly defined as deleted (the function may be an explicitly
  defaulted function that is defined as deleted)

The attribute shall be present on the first declaration of a function and
may optionally be present on subsequent declarations.

When compiling for a SYCL device target that does not support the generic
address space, the function shall not specify a raw pointer or reference type
as the return type or as a parameter type.
See section 5.10, "SYCL offline linking", of the SYCL 2020 specification.
The following examples demonstrate the use of this attribute:

```c++
[[clang::sycl_external]] void Foo(); // Ok.

[[clang::sycl_external]] void Bar() { /* ... */ } // Ok.

[[clang::sycl_external]] extern void Baz(); // Ok.

[[clang::sycl_external]] static void Quux() { /* ... */ } // error:  Quux() has internal linkage.
```

  }];
}

def SYCLKernelEntryPointDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `sycl_kernel_entry_point` attribute facilitates the launch of a SYCL
kernel and the generation of an offload kernel entry point, sometimes called
a SYCL kernel caller function, suitable for invoking a SYCL kernel on an
offload device. The attribute is intended for use in the implementation of
SYCL kernel invocation functions like the `single_task` and `parallel_for`
member functions of the `sycl::handler` class specified in section 4.9.4,
"Command group `handler` class", of the SYCL 2020 specification.

The attribute requires a single type argument that meets the requirements for
a SYCL kernel name as described in section 5.2, "Naming of kernels", of the
SYCL 2020 specification. A unique kernel name type is required for each
function declared with the attribute. The attribute may not first appear on a
declaration that follows a definition of the function.

The attribute only appertains to functions and only those that meet the
following requirements.

- Has a non-deduced `void` return type.
- Is not a constructor or destructor.
- Is not a non-static member function with an explicit object parameter.
- Is not a C variadic function.
- Is not a coroutine.
- Is not defined as deleted or as defaulted.
- Is not defined with a function try block.
- Is not declared with the `constexpr` or `consteval` specifiers.
- Is not declared with the `[[noreturn]]` attribute.

Use in the implementation of a SYCL kernel invocation function might look as
follows.

```c++
namespace sycl {
class handler {
  template<typename KernelName, typename... Ts>
  void sycl_kernel_launch(const char* kernelSymbol, Ts&&... kernelArgs) {
    // This code will run on the host and is responsible for calling functions
    // appropriate for the desired offload backend (OpenCL, CUDA, HIP,
    // Level Zero, etc...) to copy the kernel arguments denoted by kernelArgs
    // to a device and to schedule an invocation of the offload kernel entry
    // point denoted by kernelSymbol with the copied arguments.
  }

  template<typename KernelName, typename KernelType>
  [[ clang::sycl_kernel_entry_point(KernelName) ]]
  void kernel_entry_point(KernelType kernelFunc) {
    // This code will run on the device. The call to kernelFunc() invokes
    // the SYCL kernel.
    kernelFunc();
  }

public:
  template<typename KernelName, typename KernelType>
  void single_task(const KernelType& kernelFunc) {
    // This code will run on the host. kernel_entry_point() is called to
    // trigger generation of an offload kernel entry point and to schedule
    // an invocation of it on a device with kernelFunc (a SYCL kernel object)
    // passed as a kernel argument. This call will result in an implicit call
    // to sycl_kernel_launch() with the symbol name for the generated offload
    // kernel entry point passed as the first function argument followed by
    // kernelFunc.
    kernel_entry_point<KernelName>(kernelFunc);
  }
};
} // namespace sycl
```

A SYCL kernel object is a callable object of class type that is constructed on
a host, often via a lambda expression, and then passed to a SYCL kernel
invocation function to be executed on an offload device. The `kernelFunc`
parameters in the example code above correspond to SYCL kernel objects.

A SYCL kernel object type is required to satisfy the device copyability
requirements specified in section 3.13.1, "Device copyable", of the SYCL 2020
specification. Additionally, any data members of the kernel object type are
required to satisfy section 4.12.4, "Rules for parameter passing to kernels".
For most types, these rules require that the type is trivially copyable.
However, the SYCL specification mandates that certain special SYCL types, such
as `sycl::accessor` and `sycl::stream`, be device copyable even if they are
not trivially copyable. These types require special handling because they cannot
necessarily be copied to device memory as if by `memcpy()`.

The SYCL kernel object and its data members constitute the parameters of an
offload kernel. An offload kernel consists of an offload entry point function
and the set of all functions and variables that are directly or indirectly used
by the entry point function.

A SYCL kernel invocation function is responsible for performing the following
tasks (likely with the help of an offload backend like OpenCL):

1. Identifying the offload kernel entry point to be used for the SYCL kernel.
2. Validating that the SYCL kernel object type and its data members meet the
   SYCL device copyability and kernel parameter requirements noted above.
3. Copying the SYCL kernel object and any other kernel arguments to device
   memory including any special handling required for SYCL special types.
4. Initiating execution of the offload kernel entry point.

The offload kernel entry point for a SYCL kernel performs the following tasks:

1. Calling the `operator()` member function of the SYCL kernel object.

The `sycl_kernel_entry_point` attribute facilitates or automates these tasks
by providing generation of an offload kernel entry point with a unique symbol
name, type checking of kernel argument requirements, and initiation of kernel
execution via synthesized calls to a `sycl_kernel_launch` template.

A function declared with the `sycl_kernel_entry_point` attribute specifies
the parameters and body of an offload entry point function. Consider the
following call to the `single_task()` SYCL kernel invocation function assuming
an implementation similar to the one shown above.

```c++
struct S { int i; };
void f(sycl::handler &handler, sycl::stream &sout, S s) {
  handler.single_task<struct KN>([=] {
    sout << "The value of s.i is " << s.i << "\n";
  });
}
```

The SYCL kernel object is the result of the lambda expression. The call to
`kernel_entry_point()` via the call to `single_task()` triggers the
generation of an offload kernel entry point function that looks approximately
as follows.

```c++
void sycl-kernel-caller-for-KN(kernel-type kernelFunc) {
  kernelFunc();
}
```

There are a few items worthy of note:

1. `sycl-kernel-caller-for-KN` is an exposition only name; the actual name
   generated for an entry point is an implementation detail and subject to
   change. However, the name will incorporate the SYCL kernel name, `KN`,
   that was passed as the `KernelName` template parameter to
   `single_task()` and eventually provided as the argument to the
   `sycl_kernel_entry_point` attribute in order to ensure that a unique
   name is generated for each entry point. There is a one-to-one correspondence
   between SYCL kernel names and offload kernel entry points.
2. The SYCL kernel is a lambda closure type and therefore has no name;
   `kernel-type` is substituted above and corresponds to the `KernelType`
   template parameter deduced in the call to `single_task()`.
3. The parameter and the call to `kernelFunc()` in the function body
   correspond to the definition of `kernel_entry_point()` as called by
   `single_task()`.
4. The parameter is type checked for conformance with the SYCL device
   copyability and kernel parameter requirements.

Within `single_task()`, the call to `kernel_entry_point()` is effectively
replaced with a synthesized call to a ''sycl_kernel_launch\`\` template that
looks approximately as follows.

```c++
sycl_kernel_launch<KN>("sycl-kernel-caller-for-KN", kernelFunc);
```

There are a few items worthy of note:

1. Lookup for the `sycl_kernel_launch` template is performed as if from the
   body of the (possibly instantiated) definition of `kernel_entry_point()`.
   If name lookup or overload resolution fails, the program is ill-formed.
   If the selected overload is a non-static member function, then `this` is
   passed as the implicit object parameter.
2. Function arguments passed to `sycl_kernel_launch()` are passed
   as if by `std::move(x)`.
3. The `sycl_kernel_launch` template is expected to be provided by the SYCL
   library implementation. It is responsible for copying the kernel arguments
   to device memory and for scheduling execution of the generated offload
   kernel entry point identified by the symbol name passed as the first
   function argument. `sycl-kernel-caller-for-KN` is substituted above for
   the actual symbol name that would be generated for the offload kernel entry
   point.

It is not necessary for a function declared with the `sycl_kernel_entry_point`
attribute to be called for the offload kernel entry point to be emitted. For
inline functions and function templates, any ODR-use will suffice. For other
functions, an ODR-use is not required; the offload kernel entry point will be
emitted if the function is defined. In any case, a call to the function is
required for the synthesized call to `sycl_kernel_launch()` to occur.

A function declared with the `sycl_kernel_entry_point` attribute may include
an exception specification. If a non-throwing exception specification is
present, an exception propagating from the implicit call to the
`sycl_kernel_launch` template will result in a call to `std::terminate()`.
Otherwise, such an exception will propagate normally.

Functions declared with the `sycl_kernel_entry_point` attribute are not
limited to the simple example shown above. They may have additional template
parameters, declare additional function parameters, and have complex control
flow in the function body. The function must abide by the language feature
restrictions described in section 5.4, "Language restrictions for device
functions" in the SYCL 2020 specification. If the function is a non-static
member function, `this` shall not be used in a potentially evaluated
expression.
  }];
}

def SYCLSpecialClassDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
SYCL defines some special classes (accessor, sampler, and stream) which require
specific handling during the generation of the SPIR entry point.
The `__attribute__((sycl_special_class))` attribute is used in SYCL
headers to indicate that a class or a struct needs a specific handling when
it is passed from host to device.
Special classes will have a mandatory `__init` method and an optional
`__finalize` method (the `__finalize` method is used only with the
`stream` type). Kernel parameters types are extract from the `__init` method
parameters. The kernel function arguments list is derived from the
arguments of the `__init` method. The arguments of the `__init` method are
copied into the kernel function argument list and the `__init` and
`__finalize` methods are called at the beginning and the end of the kernel,
respectively.
The `__init` and `__finalize` methods must be defined inside the
special class.
Please note that this is an attribute that is used as an internal
implementation detail and not intended to be used by external users.

The syntax of the attribute is as follows:

```text
class __attribute__((sycl_special_class)) accessor {};
class [[clang::sycl_special_class]] accessor {};
```

This is a code example that illustrates the use of the attribute:

```c++
class __attribute__((sycl_special_class)) SpecialType {
  int F1;
  int F2;
  void __init(int f1) {
    F1 = f1;
    F2 = f1;
  }
  void __finalize() {}
public:
  SpecialType() = default;
  int getF2() const { return F2; }
};

int main () {
  SpecialType T;
  cgh.single_task([=] {
    T.getF2();
  });
}
```

This would trigger the following kernel entry point in the AST:

```c++
void __sycl_kernel(int f1) {
  SpecialType T;
  T.__init(f1);
  ...
  T.__finalize()
}
```
  }];
}

def C11NoReturnDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
A function declared as `_Noreturn` shall not return to its caller. The
compiler will generate a diagnostic for a function declared as `_Noreturn`
that appears to be capable of returning to its caller. Despite being a type
specifier, the `_Noreturn` attribute cannot be specified on a function
pointer type.
  }];
}

def CXX11NoReturnDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "noreturn, _Noreturn";
  let Content = [{
A function declared as `[[noreturn]]` shall not return to its caller. The
compiler will generate a diagnostic for a function declared as `[[noreturn]]`
that appears to be capable of returning to its caller.

The `[[_Noreturn]]` spelling is deprecated and only exists to ease code
migration for code using `[[noreturn]]` after including `<stdnoreturn.h>`.
  }];
}

def NoMergeDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
If a statement is marked `nomerge` and contains call expressions, those call
expressions inside the statement will not be merged during optimization. This
attribute can be used to prevent the optimizer from obscuring the source
location of certain calls. For example, it will prevent tail merging otherwise
identical code sequences that raise an exception or terminate the program. Tail
merging normally reduces the precision of source location information, making
stack traces less useful for debugging. This attribute gives the user control
over the tradeoff between code size and debug information precision.

`nomerge` attribute can also be used as function attribute to prevent all
calls to the specified function from merging. It has no effect on indirect
calls to such functions. For example:

```c++
[[clang::nomerge]] void foo(int) {}

void bar(int x) {
  auto *ptr = foo;
  if (x) foo(1); else foo(2); // will not be merged
  if (x) ptr(1); else ptr(2); // indirect call, can be merged
}
```

`nomerge` attribute can also be used for pointers to functions to
prevent calls through such pointer from merging. In such case the
effect applies only to a specific function pointer. For example:

```c++
[[clang::nomerge]] void (*foo)(int);

void bar(int x) {
  auto *ptr = foo;
  if (x) foo(1); else foo(2); // will not be merged
  if (x) ptr(1); else ptr(2); // 'ptr' has no 'nomerge' attribute, can be merged
}
```
  }];
}

def NoInlineDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This function attribute suppresses the inlining of a function at the call sites
of the function.

`[[clang::noinline]]` spelling can be used as a statement attribute; other
spellings of the attribute are not supported on statements. If a statement is
marked `[[clang::noinline]]` and contains calls, those calls inside the
statement will not be inlined by the compiler.

`__noinline__` can be used as a keyword in CUDA/HIP languages. This is to
avoid diagnostics due to usage of `__attribute__((__noinline__))`
with `__noinline__` defined as a macro as `__attribute__((noinline))`.

```c
int example(void) {
  int r;
  [[clang::noinline]] foo();
  [[clang::noinline]] r = bar();
  return r;
}
```

  }];
}

def NoOutlineDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This function attribute suppresses outlining from the annotated function.

Outlining is the process where common parts of separate functions are extracted
into a separate function (or assembly snippet), and calls to that function or
snippet are inserted in the original functions. In this way, it can be seen as
the opposite of inlining. It can help to reduce code size.
  }];
}

def MustTailDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
If a `return` statement is marked `musttail`, this indicates that the
compiler must generate a tail call for the program to be correct, even when
optimizations are disabled. This guarantees that the call will not cause
unbounded stack growth if it is part of a recursive cycle in the call graph.

If the callee is a virtual function that is implemented by a thunk, there is
no guarantee in general that the thunk tail-calls the implementation of the
virtual function, so such a call in a recursive cycle can still result in
unbounded stack growth.

`clang::musttail` can only be applied to a `return` statement whose value
is the result of a function call (even functions returning void must use
`return`, although no value is returned). The target function must have the
same number of arguments as the caller. The types of the return value and all
arguments must be similar according to C++ rules (differing only in cv
qualifiers or array size), including the implicit "this" argument, if any.
Any variables in scope, including all arguments to the function and the
return value must be trivially destructible. The calling convention of the
caller and callee must match, and they must not be variadic functions or have
old style K&R C function declarations.

The lifetimes of all local variables and function parameters end immediately
before the call to the function. This means that it is undefined behaviour to
pass a pointer or reference to a local variable to the called function, which
is not the case without the attribute. Clang will emit a warning in common
cases where this happens.

`clang::musttail` provides assurances that the tail call can be optimized on
all targets, not just one.
  }];
}

def AssertCapabilityDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "assert_capability, assert_shared_capability";
  let Content = [{
Marks a function that dynamically tests whether a capability is held, and halts
the program if it is not held.
  }];
}

def AcquireCapabilityDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "acquire_capability, acquire_shared_capability";
  let Content = [{
Marks a function as acquiring a capability.
  }];
}

def TryAcquireCapabilityDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "try_acquire_capability, try_acquire_shared_capability";
  let Content = [{
Marks a function that attempts to acquire a capability. This function may fail to
actually acquire the capability; they accept a Boolean value determining
whether acquiring the capability means success (true), or failing to acquire
the capability means success (false).
  }];
}

def ReleaseCapabilityDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "release_capability, release_shared_capability";
  let Content = [{
Marks a function as releasing a capability.
  }];
}

def AssumeAlignedDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use `__attribute__((assume_aligned(<alignment>[,<offset>]))` on a function
declaration to specify that the return value of the function (which must be a
pointer type) has the specified offset, in bytes, from an address with the
specified alignment. The offset is taken to be zero if omitted.

```c++
// The returned pointer value has 32-byte alignment.
void *a() __attribute__((assume_aligned (32)));

// The returned pointer value is 4 bytes greater than an address having
// 32-byte alignment.
void *b() __attribute__((assume_aligned (32, 4)));
```

Note that this attribute provides information to the compiler regarding a
condition that the code already ensures is true. It does not cause the compiler
to enforce the provided alignment assumption.
  }];
}

def AllocSizeDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `alloc_size` attribute can be placed on functions that return pointers in
order to hint to the compiler how many bytes of memory will be available at the
returned pointer. `alloc_size` takes one or two arguments.

- `alloc_size(N)` implies that argument number N equals the number of
  available bytes at the returned pointer.
- `alloc_size(N, M)` implies that the product of argument number N and
  argument number M equals the number of available bytes at the returned
  pointer.

Argument numbers are 1-based.

An example of how to use `alloc_size`

```c
void *my_malloc(int a) __attribute__((alloc_size(1)));
void *my_calloc(int a, int b) __attribute__((alloc_size(1, 2)));

int main() {
  void *const p = my_malloc(100);
  assert(__builtin_object_size(p, 0) == 100);
  void *const a = my_calloc(20, 5);
  assert(__builtin_object_size(a, 0) == 100);
}
```

When `-Walloc-size` is enabled, this attribute allows the compiler to
diagnose cases when the allocated memory is insufficient for the size of the
type the returned pointer is cast to.

```c
void *my_malloc(int a) __attribute__((alloc_size(1)));
void consumer_func(int *);

int main() {
  int *ptr = my_malloc(sizeof(int)); // no warning
  int *w = my_malloc(1); // warning: allocation of insufficient size '1' for type 'int' with size '4'
  consumer_func(my_malloc(1)); // warning: allocation of insufficient size '1' for type 'int' with size '4'
}
```

:::{Note}
This attribute works differently in clang than it does in GCC.
Specifically, clang will only trace `const` pointers (as above); we give up
on pointers that are not marked as `const`. In the vast majority of cases,
this is unimportant, because LLVM has support for the `alloc_size`
attribute. However, this may cause mildly unintuitive behavior when used with
other attributes, such as `enable_if`.
:::
  }];
}

def CodeSegDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `__declspec(code_seg)` attribute enables the placement of code into separate
named segments that can be paged or locked in memory individually. This attribute
is used to control the placement of instantiated templates and compiler-generated
code. See the documentation for [\_\_declspec(code_seg)][__declspec(code_seg)] on MSDN.

[__declspec(code_seg)]: http://msdn.microsoft.com/en-us/library/dn636922.aspx
  }];
}

def AllocAlignDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use `__attribute__((alloc_align(<parameter-index>)))` on a declaration with a
function prototype to specify that the prototype's return value (which must be a
pointer type) is at least as aligned as the value of the indicated parameter.
This includes functions, Objective-C methods, blocks, and declarations of
function pointer, member function pointer, function reference, and block pointer
types. The attribute can also be applied to typedef or type alias declarations
whose underlying type has a function prototype.

The parameter is given by its index in the list of formal parameters; the first
parameter has index 1 unless the function is a C++ non-static member function,
in which case the first parameter has index 2 to account for the implicit `this`
parameter.

```c++
// The returned pointer has the alignment specified by the first parameter.
void *a(size_t align) __attribute__((alloc_align(1)));

// The function pointer's returned pointer has the alignment specified by
// the first parameter of the pointed-to function.
void *(*allocator)(size_t align) __attribute__((alloc_align(1)));

// The returned pointer has the alignment specified by the second parameter.
void *b(void *v, size_t align) __attribute__((alloc_align(2)));

// The returned pointer has the alignment specified by the second visible
// parameter, however it must be adjusted for the implicit 'this' parameter.
void *Foo::b(void *v, size_t align) __attribute__((alloc_align(3)));
```

Note that this attribute merely informs the compiler that a function always
returns a sufficiently aligned pointer. It does not cause the compiler to
emit code to enforce that alignment. The behavior is undefined if the returned
pointer is not sufficiently aligned.
  }];
}

def EnableIfDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
:::{Note}
Some features of this attribute are experimental. The meaning of
multiple enable_if attributes on a single declaration is subject to change in
a future version of clang. Also, the ABI is not standardized and the name
mangling may change in future versions. To avoid that, use asm labels.
:::

The `enable_if` attribute can be placed on function declarations to control
which overload is selected based on the values of the function's arguments.
When combined with the `overloadable` attribute, this feature is also
available in C.

```c++
int isdigit(int c);
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")));

void foo(char c) {
  isdigit(c);
  isdigit(10);
  isdigit(-10);  // results in a compile-time error.
}
```

The enable_if attribute takes two arguments, the first is an expression written
in terms of the function parameters, the second is a string explaining why this
overload candidate could not be selected to be displayed in diagnostics. The
expression is part of the function signature for the purposes of determining
whether it is a redeclaration (following the rules used when determining
whether a C++ template specialization is ODR-equivalent), but is not part of
the type.

The enable_if expression is evaluated as if it were the body of a
bool-returning constexpr function declared with the arguments of the function
it is being applied to, then called with the parameters at the call site. If the
result is false or could not be determined through constant expression
evaluation, then this overload will not be chosen and the provided string may
be used in a diagnostic if the compile fails as a result.

Because the enable_if expression is an unevaluated context, there are no global
state changes, nor the ability to pass information from the enable_if
expression to the function body. For example, suppose we want calls to
strnlen(strbuf, maxlen) to resolve to strnlen_chk(strbuf, maxlen, size of
strbuf) only if the size of strbuf can be determined:

```c++
__attribute__((always_inline))
static inline size_t strnlen(const char *s, size_t maxlen)
  __attribute__((overloadable))
  __attribute__((enable_if(__builtin_object_size(s, 0) != -1))),
                           "chosen when the buffer size is known but 'maxlen' is not")))
{
  return strnlen_chk(s, maxlen, __builtin_object_size(s, 0));
}
```

Multiple enable_if attributes may be applied to a single declaration. In this
case, the enable_if expressions are evaluated from left to right in the
following manner. First, the candidates whose enable_if expressions evaluate to
false or cannot be evaluated are discarded. If the remaining candidates do not
share ODR-equivalent enable_if expressions, the overload resolution is
ambiguous. Otherwise, enable_if overload resolution continues with the next
enable_if attribute on the candidates that have not been discarded and have
remaining enable_if attributes. In this way, we pick the most specific
overload out of a number of viable overloads using enable_if.

```c++
void f() __attribute__((enable_if(true, "")));  // #1
void f() __attribute__((enable_if(true, ""))) __attribute__((enable_if(true, "")));  // #2

void g(int i, int j) __attribute__((enable_if(i, "")));  // #1
void g(int i, int j) __attribute__((enable_if(j, ""))) __attribute__((enable_if(true)));  // #2
```

In this example, a call to f() is always resolved to #2, as the first enable_if
expression is ODR-equivalent for both declarations, but #1 does not have another
enable_if expression to continue evaluating, so the next round of evaluation has
only a single candidate. In a call to g(1, 1), the call is ambiguous even though
#2 has more enable_if attributes, because the first enable_if expressions are
not ODR-equivalent.

Query for this feature with `__has_attribute(enable_if)`.

Note that functions with one or more `enable_if` attributes may not have
their address taken, unless all of the conditions specified by said
`enable_if` are constants that evaluate to `true`. For example:

```c
const int TrueConstant = 1;
const int FalseConstant = 0;
int f(int a) __attribute__((enable_if(a > 0, "")));
int g(int a) __attribute__((enable_if(a == 0 || a != 0, "")));
int h(int a) __attribute__((enable_if(1, "")));
int i(int a) __attribute__((enable_if(TrueConstant, "")));
int j(int a) __attribute__((enable_if(FalseConstant, "")));

void fn() {
  int (*ptr)(int);
  ptr = &f; // error: 'a > 0' is not always true
  ptr = &g; // error: 'a == 0 || a != 0' is not a truthy constant
  ptr = &h; // OK: 1 is a truthy constant
  ptr = &i; // OK: 'TrueConstant' is a truthy constant
  ptr = &j; // error: 'FalseConstant' is a constant, but not truthy
}
```

Because `enable_if` evaluation happens during overload resolution,
`enable_if` may give unintuitive results when used with templates, depending
on when overloads are resolved. In the example below, clang will emit a
diagnostic about no viable overloads for `foo` in `bar`, but not in `baz`:

```c++
double foo(int i) __attribute__((enable_if(i > 0, "")));
void *foo(int i) __attribute__((enable_if(i <= 0, "")));
template <int I>
auto bar() { return foo(I); }

template <typename T>
auto baz() { return foo(T::number); }

struct WithNumber { constexpr static int number = 1; };
void callThem() {
  bar<sizeof(WithNumber)>();
  baz<WithNumber>();
}
```

This is because, in `bar`, `foo` is resolved prior to template
instantiation, so the value for `I` isn't known (thus, both `enable_if`
conditions for `foo` fail). However, in `baz`, `foo` is resolved during
template instantiation, so the value for `T::number` is known.
  }];
}

def ExtVectorTypeDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `ext_vector_type(N)` attribute specifies that a type is a vector with N
elements, directly mapping to an LLVM vector type. Originally from OpenCL, it
allows element access the array subscript operator `[]`, `sN` where N is
a hexadecimal value, or `x, y, z, w` for graphics-style indexing.
This attribute enables efficient SIMD operations and is usable in
general-purpose code.

```c++
template <typename T, uint32_t N>
constexpr T simd_reduce(T [[clang::ext_vector_type(N)]] v) {
  static_assert((N & (N - 1)) == 0, "N must be a power of two");
  if constexpr (N == 1)
    return v[0];
  else
    return simd_reduce<T, N / 2>(v.hi + v.lo);
}
```

The vector type also supports swizzling up to sixteen elements. This can be done
using the object accessors. The OpenCL documentation lists all of the accepted
values.

```c++
using f16_x16 = _Float16 __attribute__((ext_vector_type(16)));

f16_x16 reverse(f16_x16 v) { return v.sfedcba9876543210; }
```

See the OpenCL documentation for some more complete examples.
  }];
}

def DiagnoseIfDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `diagnose_if` attribute can be placed on function declarations to emit
warnings or errors at compile-time if calls to the attributed function meet
certain user-defined criteria. For example:

```c
int abs(int a)
  __attribute__((diagnose_if(a >= 0, "Redundant abs call", "warning")));
int must_abs(int a)
  __attribute__((diagnose_if(a >= 0, "Redundant abs call", "error")));

int val = abs(1); // warning: Redundant abs call
int val2 = must_abs(1); // error: Redundant abs call
int val3 = abs(val);
int val4 = must_abs(val); // Because run-time checks are not emitted for
                          // diagnose_if attributes, this executes without
                          // issue.
```

`diagnose_if` is closely related to `enable_if`, with a few key differences:

- Overload resolution is not aware of `diagnose_if` attributes: they're
  considered only after we select the best candidate from a given candidate set.
- Function declarations that differ only in their `diagnose_if` attributes are
  considered to be redeclarations of the same function (not overloads).
- If the condition provided to `diagnose_if` cannot be evaluated, no
  diagnostic will be emitted.

Otherwise, `diagnose_if` is essentially the logical negation of `enable_if`.

As a result of bullet number two, `diagnose_if` attributes will stack on the
same function. For example:

```c
int foo() __attribute__((diagnose_if(1, "diag1", "warning")));
int foo() __attribute__((diagnose_if(1, "diag2", "warning")));

int bar = foo(); // warning: diag1
                 // warning: diag2
int (*fooptr)(void) = foo; // warning: diag1
                           // warning: diag2

constexpr int supportsAPILevel(int N) { return N < 5; }
int baz(int a)
  __attribute__((diagnose_if(!supportsAPILevel(10),
                             "Upgrade to API level 10 to use baz", "error")));
int baz(int a)
  __attribute__((diagnose_if(!a, "0 is not recommended.", "warning")));

int (*bazptr)(int) = baz; // error: Upgrade to API level 10 to use baz
int v = baz(0); // error: Upgrade to API level 10 to use baz
```

Query for this feature with `__has_attribute(diagnose_if)`.
  }];
}

def NoSpecializationsDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
``[[clang::no_specializations]]`` can be applied to function, class, or variable
templates for which neither an explicit specialization nor a partial specialization should be declared by users. This is primarily
used to diagnose user specializations of standard library type traits.
  }];
}

def PassObjectSizeDocs : Documentation {
  let Category = DocCatVariable; // Technically it's a parameter doc, but eh.
  let Heading = "pass_object_size, pass_dynamic_object_size";
  let Content = [{
:::{Note}
The mangling of functions with parameters that are annotated with
`pass_object_size` is subject to change. You can get around this by
using `__asm__("foo")` to explicitly name your functions, thus preserving
your ABI; also, non-overloadable C functions with `pass_object_size` are
not mangled.
:::

The `pass_object_size(Type)` attribute can be placed on function parameters to
instruct clang to call `__builtin_object_size(param, Type)` at each callsite
of said function, and implicitly pass the result of this call in as an invisible
argument of type `size_t` directly after the parameter annotated with
`pass_object_size`. Clang will also replace any calls to
`__builtin_object_size(param, Type)` in the function by said implicit
parameter.

Example usage:

```c
int bzero1(char *const p __attribute__((pass_object_size(0))))
    __attribute__((noinline)) {
  int i = 0;
  for (/**/; i < (int)__builtin_object_size(p, 0); ++i) {
    p[i] = 0;
  }
  return i;
}

int main() {
  char chars[100];
  int n = bzero1(&chars[0]);
  assert(n == sizeof(chars));
  return 0;
}
```

If successfully evaluating `__builtin_object_size(param, Type)` at the
callsite is not possible, then the "failed" value is passed in. So, using the
definition of `bzero1` from above, the following code would exit cleanly:

```c
int main2(int argc, char *argv[]) {
  int n = bzero1(argv);
  assert(n == -1);
  return 0;
}
```

`pass_object_size` plays a part in overload resolution. If two overload
candidates are otherwise equally good, then the overload with one or more
parameters with `pass_object_size` is preferred. This implies that the choice
between two identical overloads both with `pass_object_size` on one or more
parameters will always be ambiguous; for this reason, having two such overloads
is illegal. For example:

```c++
#define PS(N) __attribute__((pass_object_size(N)))
// OK
void Foo(char *a, char *b); // Overload A
// OK -- overload A has no parameters with pass_object_size.
void Foo(char *a PS(0), char *b PS(0)); // Overload B
// Error -- Same signature (sans pass_object_size) as overload B, and both
// overloads have one or more parameters with the pass_object_size attribute.
void Foo(void *a PS(0), void *b);

// OK
void Bar(void *a PS(0)); // Overload C
// OK
void Bar(char *c PS(1)); // Overload D

void main() {
  char known[10], *unknown;
  Foo(unknown, unknown); // Calls overload B
  Foo(known, unknown); // Calls overload B
  Foo(unknown, known); // Calls overload B
  Foo(known, known); // Calls overload B

  Bar(known); // Calls overload D
  Bar(unknown); // Calls overload D
}
```

Currently, `pass_object_size` is a bit restricted in terms of its usage:

- Only one use of `pass_object_size` is allowed per parameter.
- It is an error to take the address of a function with `pass_object_size` on
  any of its parameters. If you wish to do this, you can create an overload
  without `pass_object_size` on any parameters.
- It is an error to apply the `pass_object_size` attribute to parameters that
  are not pointers. Additionally, any parameter that `pass_object_size` is
  applied to must be marked `const` at its function's definition.

Clang also supports the `pass_dynamic_object_size` attribute, which behaves
identically to `pass_object_size`, but evaluates a call to
`__builtin_dynamic_object_size` at the callee instead of
`__builtin_object_size`. `__builtin_dynamic_object_size` provides some extra
runtime checks when the object size can't be determined at compile-time. You can
read more about `__builtin_dynamic_object_size` in
{ref}`Evaluating Object Size <langext-evaluating-object-size>`.

  }];
}

def OverloadableDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang provides support for C++ function overloading in C. Function overloading
in C is introduced using the `overloadable` attribute. For example, one
might provide several overloaded versions of a `tgsin` function that invokes
the appropriate standard function computing the sine of a value with `float`,
`double`, or `long double` precision:

```c
#include <math.h>
float __attribute__((overloadable)) tgsin(float x) { return sinf(x); }
double __attribute__((overloadable)) tgsin(double x) { return sin(x); }
long double __attribute__((overloadable)) tgsin(long double x) { return sinl(x); }
```

Given these declarations, one can call `tgsin` with a `float` value to
receive a `float` result, with a `double` to receive a `double` result,
etc. Function overloading in C follows the rules of C++ function overloading
to pick the best overload given the call arguments, with a few C-specific
semantics:

- Conversion from `float` or `double` to `long double` is ranked as a
  floating-point promotion (per C99) rather than as a floating-point conversion
  (as in C++).
- A conversion from a pointer of type `T*` to a pointer of type `U*` is
  considered a pointer conversion (with conversion rank) if `T` and `U` are
  compatible types.
- A conversion from type `T` to a value of type `U` is permitted if `T`
  and `U` are compatible types. This conversion is given "conversion" rank.
- If no viable candidates are otherwise available, we allow a conversion from a
  pointer of type `T*` to a pointer of type `U*`, where `T` and `U` are
  incompatible. This conversion is ranked below all other types of conversions.
  Please note: `U` lacking qualifiers that are present on `T` is sufficient
  for `T` and `U` to be incompatible.

The declaration of `overloadable` functions is restricted to function
declarations and definitions. If a function is marked with the `overloadable`
attribute, then all declarations and definitions of functions with that name,
except for at most one (see the note below about unmarked overloads), must have
the `overloadable` attribute. In addition, redeclarations of a function with
the `overloadable` attribute must have the `overloadable` attribute, and
redeclarations of a function without the `overloadable` attribute must *not*
have the `overloadable` attribute. e.g.,

```c
int f(int) __attribute__((overloadable));
float f(float); // error: declaration of "f" must have the "overloadable" attribute
int f(int); // error: redeclaration of "f" must have the "overloadable" attribute

int g(int) __attribute__((overloadable));
int g(int) { } // error: redeclaration of "g" must also have the "overloadable" attribute

int h(int);
int h(int) __attribute__((overloadable)); // error: declaration of "h" must not
                                          // have the "overloadable" attribute
```

Functions marked `overloadable` must have prototypes. Therefore, the
following code is ill-formed:

```c
int h() __attribute__((overloadable)); // error: h does not have a prototype
```

However, `overloadable` functions are allowed to use a ellipsis even if there
are no named parameters (as is permitted in C++). This feature is particularly
useful when combined with the `unavailable` attribute:

```c++
void honeypot(...) __attribute__((overloadable, unavailable)); // calling me is an error
```

Functions declared with the `overloadable` attribute have their names mangled
according to the same rules as C++ function names. For example, the three
`tgsin` functions in our motivating example get the mangled names
`_Z5tgsinf`, `_Z5tgsind`, and `_Z5tgsine`, respectively. There are two
caveats to this use of name mangling:

- Future versions of Clang may change the name mangling of functions overloaded
  in C, so you should not depend on an specific mangling. To be completely
  safe, we strongly urge the use of `static inline` with `overloadable`
  functions.
- The `overloadable` attribute has almost no meaning when used in C++,
  because names will already be mangled and functions are already overloadable.
  However, when an `overloadable` function occurs within an `extern "C"`
  linkage specification, its name *will* be mangled in the same way as it
  would in C.

For the purpose of backwards compatibility, at most one function with the same
name as other `overloadable` functions may omit the `overloadable`
attribute. In this case, the function without the `overloadable` attribute
will not have its name mangled.

For example:

```c
// Notes with mangled names assume Itanium mangling.
int f(int);
int f(double) __attribute__((overloadable));
void foo() {
  f(5); // Emits a call to f (not _Z1fi, as it would with an overload that
        // was marked with overloadable).
  f(1.0); // Emits a call to _Z1fd.
}
```

Support for unmarked overloads is not present in some versions of clang. You may
query for it using `__has_extension(overloadable_unmarked)`.

Query for this attribute with `__has_attribute(overloadable)`.
  }];
}

def OwnershipDocs : Documentation {
  let Heading = "ownership_holds, ownership_returns, ownership_takes (Clang "
                "Static Analyzer)";
  let Category = DocCatFunction;
  let Label = "analyzer-ownership-attrs";
  let Content = [{
:::{note}
In order for the Clang Static Analyzer to acknowledge these attributes, the
`Optimistic` config needs to be set to true for the checker
`unix.DynamicMemoryModeling`:

`-Xclang -analyzer-config -Xclang unix.DynamicMemoryModeling:Optimistic=true`
:::

These attributes are used by the Clang Static Analyzer's dynamic memory modeling
facilities to mark custom allocating/deallocating functions.

All 3 attributes' first parameter of type string is the type of the allocation:
`malloc`, `new`, etc. to allow for catching {ref}`mismatched deallocation
<unix-MismatchedDeallocator>` bugs. The allocation type can be any string, e.g.
a function annotated with
returning a piece of memory of type `lasagna` but freed with a function
annotated to release `cheese` typed memory will result in mismatched
deallocation warning.

The (currently) only allocation type having special meaning is `malloc` --
the Clang Static Analyzer makes sure that allocating functions annotated with
`malloc` are treated like they used the standard `malloc()`, and can be
safely deallocated with the standard `free()`.

- Use `ownership_returns` to mark a function as an allocating function.
  It takes 1 or 2 arguments.
  The first argument is a user-provided identifier representing the "kind" of the allocation.
  This is basically what is enforced when checking the deallocation. This is mandatory.
  The second argument is optional.
  It represents the index of the parameter that represents the allocation size in bytes (counting from 1).
  The referenced parameter must have some integral type.
  This attribute may appear at most once per declaration.
  If this argument is not set, then tooling, such as the Clang Static Analyzer,
  won't be able to reason about the size of the allocation, thus check potential out-of-bounds accesses.
  However, such tooling could still warn if the wrong deallocation function
  was used for the `ownership_returns` attributed resource.
  If forward declarations have this attribute, those must have the same arguments.
- Use `ownership_takes` to mark a function as a deallocating function. Takes 2
  arguments: the allocation type, and the index of the parameter that is being
  deallocated (counting from 1).
- Use `ownership_holds` to mark that a function takes over the ownership of a
  piece of memory and will free it at some unspecified point in the future. Like
  `ownership_takes`, this takes 2 arguments: the allocation type, and the
  index of the parameter whose ownership will be taken over (counting from 1).

The annotations `ownership_takes` and `ownership_holds` both prevent memory
leak reports (concerning the specified parameter); the difference between them
is that using taken memory is a use-after-free error, while using held memory
is assumed to be legitimate. However, releasing the held memory or passing it
to another holding call is reported by the analyzer as an "attempt to release
non-owned memory".

Example:

```c
// Denotes that my_malloc will return with a dynamically allocated piece of
// memory using malloc().
void __attribute((ownership_returns(malloc))) *my_malloc(size_t sz);

// 'sz' (parameter 1) is the allocation size.
void __attribute((ownership_returns(malloc, 1))) *my_sized_malloc(size_t sz);

// Denotes that my_free will deallocate its argument using free().
void __attribute((ownership_takes(malloc, 1))) my_free(void *);

// Denotes that my_hold will take over the ownership of its argument that was
// allocated via malloc().
void __attribute((ownership_holds(malloc, 1))) my_hold(void *);
```

Further reading about dynamic memory modeling in the Clang Static Analyzer is
found in these checker docs:
{ref}`unix.Malloc <unix-Malloc>`, {ref}`unix.MallocSizeof <unix-MallocSizeof>`,
{ref}`unix.MismatchedDeallocator <unix-MismatchedDeallocator>`,
{ref}`cplusplus.NewDelete <cplusplus-NewDelete>`,
{ref}`cplusplus.NewDeleteLeaks <cplusplus-NewDeleteLeaks>`,
{ref}`optin.taint.TaintedAlloc <optin-taint-TaintedAlloc>`.
Mind that many more checkers are affected by dynamic memory modeling changes to
some extent.

Further reading for other annotations:
{doc}`Static Analyzer source annotations <analyzer/user-docs/Annotations>`.
  }];
}

def ObjCMethodFamilyDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Many methods in Objective-C have conventional meanings determined by their
selectors. It is sometimes useful to be able to mark a method as having a
particular conventional meaning despite not having the right selector, or as
not having the conventional meaning that its selector would suggest. For these
use cases, we provide an attribute to specifically describe the "method family"
that a method belongs to.

**Usage**: `__attribute__((objc_method_family(X)))`, where `X` is one of
`none`, `alloc`, `copy`, `init`, `mutableCopy`, or `new`. This
attribute can only be placed at the end of a method declaration:

```objc
- (NSString *)initMyStringValue __attribute__((objc_method_family(none)));
```

Users who do not wish to change the conventional meaning of a method, and who
merely want to document its non-standard retain and release semantics, should
use the retaining behavior attributes (`ns_returns_retained`,
`ns_returns_not_retained`, etc).

Query for this feature with `__has_attribute(objc_method_family)`.
  }];
}

def RetainBehaviorDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The behavior of a function with respect to reference counting for Foundation
(Objective-C), CoreFoundation (C) and OSObject (C++) is determined by a naming
convention (e.g. functions starting with "get" are assumed to return at
`+0`).

It can be overridden using a family of the following attributes. In
Objective-C, the annotation `__attribute__((ns_returns_retained))` applied to
a function communicates that the object is returned at `+1`, and the caller
is responsible for freeing it.
Similarly, the annotation `__attribute__((ns_returns_not_retained))`
specifies that the object is returned at `+0` and the ownership remains with
the callee.
The annotation `__attribute__((ns_consumes_self))` specifies that
the Objective-C method call consumes the reference to `self`, e.g. by
attaching it to a supplied parameter.
Additionally, parameters can have an annotation
`__attribute__((ns_consumed))`, which specifies that passing an owned object
as that parameter effectively transfers the ownership, and the caller is no
longer responsible for it.
These attributes affect code generation when interacting with ARC code, and
they are used by the Clang Static Analyzer.

In C programs using CoreFoundation, a similar set of attributes:
`__attribute__((cf_returns_not_retained))`,
`__attribute__((cf_returns_retained))` and `__attribute__((cf_consumed))`
have the same respective semantics when applied to CoreFoundation objects.
These attributes affect code generation when interacting with ARC code, and
they are used by the Clang Static Analyzer.

Finally, in C++ interacting with XNU kernel (objects inheriting from OSObject),
the same attribute family is present:
`__attribute__((os_returns_not_retained))`,
`__attribute__((os_returns_retained))` and `__attribute__((os_consumed))`,
with the same respective semantics.
Similar to `__attribute__((ns_consumes_self))`,
`__attribute__((os_consumes_this))` specifies that the method call consumes
the reference to "this" (e.g., when attaching it to a different object supplied
as a parameter).
Out parameters (parameters the function is meant to write into,
either via pointers-to-pointers or references-to-pointers)
may be annotated with `__attribute__((os_returns_retained))`
or `__attribute__((os_returns_not_retained))` which specifies that the object
written into the out parameter should (or respectively should not) be released
after use.
Since often out parameters may or may not be written depending on the exit
code of the function,
annotations `__attribute__((os_returns_retained_on_zero))`
and `__attribute__((os_returns_retained_on_non_zero))` specify that
an out parameter at `+1` is written if and only if the function returns a zero
(respectively non-zero) error code.
Observe that return-code-dependent out parameter annotations are only
available for retained out parameters, as non-retained object do not have to be
released by the callee.
These attributes are only used by the Clang Static Analyzer.

The family of attributes `X_returns_X_retained` can be added to functions,
C++ methods, and Objective-C methods and properties.
Attributes `X_consumed` can be added to parameters of methods, functions,
and Objective-C methods.
  }];
}

def NoDebugDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `nodebug` attribute allows you to suppress debugging information for a
function or method, for a variable that is not a parameter or a non-static
data member, or for a typedef or using declaration.
  }];
}

def StandaloneDebugDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `standalone_debug` attribute causes debug info to be emitted for a record
type regardless of the debug info optimizations that are enabled with
-fno-standalone-debug. This attribute only has an effect when debug info
optimizations are enabled (e.g. with -fno-standalone-debug), and is C++-only.
  }];
}

def NoDuplicateDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `noduplicate` attribute can be placed on function declarations to control
whether function calls to this function can be duplicated or not as a result of
optimizations. This is required for the implementation of functions with
certain special requirements, like the OpenCL "barrier" function, that might
need to be run concurrently by all the threads that are executing in lockstep
on the hardware. For example this attribute applied on the function
"nodupfunc" in the code below avoids that:

```c
void nodupfunc() __attribute__((noduplicate));
// Setting it as a C++11 attribute is also valid
// void nodupfunc() [[clang::noduplicate]];
void foo();
void bar();

nodupfunc();
if (a > n) {
  foo();
} else {
  bar();
}
```

gets possibly modified by some optimizations into code similar to this:

```c
if (a > n) {
  nodupfunc();
  foo();
} else {
  nodupfunc();
  bar();
}
```

where the call to "nodupfunc" is duplicated and sunk into the two branches
of the condition.
  }];
}

def ConvergentDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `convergent` attribute can be placed on a function declaration. It is
translated into the LLVM `convergent` attribute, which indicates that the call
instructions of a function with this attribute cannot be made control-dependent
on any additional values.

This attribute is different from `noduplicate` because it allows duplicating
function calls if it can be proved that the duplicated function calls are
not made control-dependent on any additional values, e.g., unrolling a loop
executed by all work items.

Sample usage:

```c
void convfunc(void) __attribute__((convergent));
// Setting it as a C++11 attribute is also valid in a C++ program.
// void convfunc(void) [[clang::convergent]];
```

  }];
}

def NoConvergentDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute prevents a function from being treated as convergent; when a
function is marked `noconvergent`, calls to that function are not
automatically assumed to be convergent, unless such calls are explicitly marked
as `convergent`. If a statement is marked as `noconvergent`, any calls to
inline `asm` in that statement are no longer treated as convergent.

In languages following SPMD/SIMT programming model, e.g., CUDA/HIP, function
declarations and inline asm calls are treated as convergent by default for
correctness. This `noconvergent` attribute is helpful for developers to
prevent them from being treated as convergent when it's safe.

```c
__device__ float bar(float);
__device__ float foo(float) __attribute__((noconvergent)) {}

__device__ int example(void) {
  float x;
  [[clang::noconvergent]] x = bar(x); // no effect on convergence
  [[clang::noconvergent]] { asm volatile ("nop"); } // the asm call is non-convergent
}
```

  }];
}

def NoSplitStackDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `no_split_stack` attribute disables the emission of the split stack
preamble for a particular function. It has no effect if `-fsplit-stack`
is not specified.
  }];
}

def ExplicitInitDocs : Documentation {
  let Category = DocCatField;
  let Content = [{
The `clang::require_explicit_initialization` attribute indicates that a
field of an aggregate must be initialized explicitly by the user when an object
of the aggregate type is constructed. The attribute supports both C and C++,
but its usage is invalid on non-aggregates.

Note that this attribute is *not* a memory safety feature, and is *not* intended
to guard against use of uninitialized memory.

Rather, it is intended for use in "parameter-objects", used to simulate,
for example, the passing of named parameters.
Except inside unevaluated contexts, the attribute generates a warning when
explicit initializers for such variables are not provided (this occurs
regardless of whether any in-class field initializers exist):

```c++
struct Buffer {
  void *address [[clang::require_explicit_initialization]];
  size_t length [[clang::require_explicit_initialization]] = 0;
};

struct ArrayIOParams {
  size_t count [[clang::require_explicit_initialization]];
  size_t element_size [[clang::require_explicit_initialization]];
  int flags = 0;
};

size_t ReadArray(FILE *file, struct Buffer buffer,
                 struct ArrayIOParams params);

int main() {
  unsigned int buf[512];
  ReadArray(stdin, {
    buf
    // warning: field 'length' is not explicitly initialized
  }, {
    .count = sizeof(buf) / sizeof(*buf),
    // warning: field 'element_size' is not explicitly initialized
    // (Note that a missing initializer for 'flags' is not diagnosed, because
    // the field is not marked as requiring explicit initialization.)
  });
}
```

  }];
}

def NoUniqueAddressDocs : Documentation {
  let Category = DocCatField;
  let Content = [{
The `no_unique_address` attribute allows tail padding in a non-static data
member to overlap other members of the enclosing class (and in the special
case when the type is empty, permits it to fully overlap other members).
The field is laid out as if a base class were encountered at the corresponding
point within the class (except that it does not share a vptr with the enclosing
object).

Example usage:

```c++
template<typename T, typename Alloc> struct my_vector {
  T *p;
  [[no_unique_address]] Alloc alloc;
  // ...
};
static_assert(sizeof(my_vector<int, std::allocator<int>>) == sizeof(int*));
```

`[[no_unique_address]]` is a standard C++20 attribute. Clang supports its use
in C++11 onwards.

On MSVC targets, `[[no_unique_address]]` is ignored; use
`[[msvc::no_unique_address]]` instead. Currently there is no guarantee of ABI
compatibility or stability with MSVC.
  }];
}

def ObjCRequiresSuperDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Some Objective-C classes allow a subclass to override a particular method in a
parent class but expect that the overriding method also calls the overridden
method in the parent class. For these cases, we provide an attribute to
designate that a method requires a "call to `super`" in the overriding
method in the subclass.

**Usage**: `__attribute__((objc_requires_super))`. This attribute can only
be placed at the end of a method declaration:

```objc
- (void)foo __attribute__((objc_requires_super));
```

This attribute can only be applied the method declarations within a class, and
not a protocol. Currently this attribute does not enforce any placement of
where the call occurs in the overriding method (such as in the case of
`-dealloc` where the call must appear at the end). It checks only that it
exists.

Note that on both OS X and iOS that the Foundation framework provides a
convenience macro `NS_REQUIRES_SUPER` that provides syntactic sugar for this
attribute:

```objc
- (void)foo NS_REQUIRES_SUPER;
```

This macro is conditionally defined depending on the compiler's support for
this attribute. If the compiler does not support the attribute the macro
expands to nothing.

Operationally, when a method has this annotation the compiler will warn if the
implementation of an override in a subclass does not call super. For example:

```objc
warning: method possibly missing a [super AnnotMeth] call
- (void) AnnotMeth{};
                   ^
```
  }];
}

def ObjCRuntimeNameDocs : Documentation {
    let Category = DocCatDecl;
    let Content = [{
By default, the Objective-C interface or protocol identifier is used
in the metadata name for that object. The `objc_runtime_name`
attribute allows annotated interfaces or protocols to use the
specified string argument in the object's metadata name instead of the
default name.

**Usage**: `__attribute__((objc_runtime_name("MyLocalName")))`. This attribute
can only be placed before an @protocol or @interface declaration:

```objc
__attribute__((objc_runtime_name("MyLocalName")))
@interface Message
@end
```

    }];
}

def ObjCRuntimeVisibleDocs : Documentation {
    let Category = DocCatDecl;
    let Content = [{
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.
    }];
}

def ObjCClassStubDocs : Documentation {
    let Category = DocCatType;
    let Content = [{
This attribute specifies that the Objective-C class to which it applies is
instantiated at runtime.

Unlike `__attribute__((objc_runtime_visible))`, a class having this attribute
still has a "class stub" that is visible to the linker. This allows categories
to be defined. Static message sends with the class as a receiver use a special
access pattern to ensure the class is lazily instantiated from the class stub.

Classes annotated with this attribute cannot be subclassed and cannot have
implementations defined for them. This attribute is intended for use in
Swift-generated headers for classes defined in Swift.

Adding or removing this attribute to a class is an ABI-breaking change.
    }];
}

def ObjCBoxableDocs : Documentation {
    let Category = DocCatDecl;
    let Content = [{
Structs and unions marked with the `objc_boxable` attribute can be used
with the Objective-C boxed expression syntax, `@(...)`.

**Usage**: `__attribute__((objc_boxable))`. This attribute
can only be placed on a declaration of a trivially-copyable struct or union:

```objc
struct __attribute__((objc_boxable)) some_struct {
  int i;
};
union __attribute__((objc_boxable)) some_union {
  int i;
  float f;
};
typedef struct __attribute__((objc_boxable)) _some_struct some_struct;

// ...

some_struct ss;
NSValue *boxed = @(ss);
```

    }];
}

def AvailabilityDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `availability` attribute can be placed on declarations to describe the
lifecycle of that declaration relative to operating system versions. Consider
the function declaration for a hypothetical function `f`:

```c++
void f(void) __attribute__((availability(macos,introduced=10.4,deprecated=10.6,obsoleted=10.7)));
```

The availability attribute states that `f` was introduced in macOS 10.4,
deprecated in macOS 10.6, and obsoleted in macOS 10.7. This information
is used by Clang to determine when it is safe to use `f`: for example, if
Clang is instructed to compile code for macOS 10.5, a call to `f()`
succeeds. If Clang is instructed to compile code for macOS 10.6, the call
succeeds but Clang emits a warning specifying that the function is deprecated.
Finally, if Clang is instructed to compile code for macOS 10.7, the call
fails because `f()` is no longer available.

Clang is instructed to compile code for a minimum deployment version using
the `-target` or `-mtargetos` command line arguments. For example,
macOS 10.7 would be specified as `-target x86_64-apple-macos10.7` or
`-mtargetos=macos10.7`. Variants like Mac Catalyst are specified as
`-target arm64-apple-ios15.0-macabi` or `-mtargetos=ios15.0-macabi`

The availability attribute is a comma-separated list starting with the
platform name and then including clauses specifying important milestones in the
declaration's lifetime (in any order) along with additional information. Those
clauses can be:

introduced=*version*

: The first version in which this declaration was introduced.

deprecated=*version*

: The first version in which this declaration was deprecated, meaning that
  users should migrate away from this API.

obsoleted=*version*

: The first version in which this declaration was obsoleted, meaning that it
  was removed completely and can no longer be used.

unavailable

: This declaration is never available on this platform.

message=*string-literal*

: Additional message text that Clang will provide when emitting a warning or
  error about use of a deprecated or obsoleted declaration. Useful to direct
  users to replacement APIs.

replacement=*string-literal*

: Additional message text that Clang will use to provide Fix-It when emitting
  a warning about use of a deprecated declaration. The Fix-It will replace
  the deprecated declaration with the new declaration specified.

environment=*identifier*

: Target environment in which this declaration is available. If present,
  the availability attribute applies only to targets with the same platform
  and environment. The parameter is currently supported only in HLSL.

Multiple availability attributes can be placed on a declaration, which may
correspond to different platforms. For most platforms, the availability
attribute with the platform corresponding to the target platform will be used;
any others will be ignored. However, the availability for `watchOS` and
`tvOS` can be implicitly inferred from an `iOS` availability attribute.
Any explicit availability attributes for those platforms are still preferred over
the implicitly inferred availability attributes. If no availability attribute
specifies availability for the current target platform, the availability
attributes are ignored. Supported platforms are:

`iOS`
`macOS`
`tvOS`
`watchOS`
`iOSApplicationExtension`
`macOSApplicationExtension`
`tvOSApplicationExtension`
`watchOSApplicationExtension`
`macCatalyst`
`macCatalystApplicationExtension`
`visionOS`
`visionOSApplicationExtension`
`driverkit`
`anyAppleOS`
`swift`
`android`
`fuchsia`
`ohos`
`zos`
`ShaderModel`

Some platforms have alias names:

`ios`
`macos`
`macosx (deprecated)`
`tvos`
`watchos`
`ios_app_extension`
`macos_app_extension`
`macosx_app_extension (deprecated)`
`tvos_app_extension`
`watchos_app_extension`
`maccatalyst`
`maccatalyst_app_extension`
`visionos`
`visionos_app_extension`
`anyappleos`
`shadermodel`

Supported environment names for the ShaderModel platform:

`pixel`
`vertex`
`geometry`
`hull`
`domain`
`compute`
`raygeneration`
`intersection`
`anyhit`
`closesthit`
`miss`
`callable`
`mesh`
`amplification`
`library`

The special platform `anyAppleOS` (alias: `anyappleos`) is a shorthand that
applies the availability attribute to all Apple Darwin platforms. An explicit
platform-specific availability attribute takes precedence over an `anyAppleOS`
attribute for that platform. Versions specified with `anyAppleOS` must be at
least 26.0, which is the first OS release where all supported Apple platforms
share a unified version number.

A declaration can typically be used even when deploying back to a platform
version prior to when the declaration was introduced. When this happens, the
declaration is [weakly linked](https://developer.apple.com/library/mac/#documentation/MacOSX/Conceptual/BPFrameworks/Concepts/WeakLinking.html),
as if the `weak_import` attribute were added to the declaration. A
weakly-linked declaration may or may not be present a run-time, and a program
can determine whether the declaration is present by checking whether the
address of that declaration is non-NULL.

The flag `strict` disallows using API when deploying back to a
platform version prior to when the declaration was introduced. An
attempt to use such API before its introduction causes a hard error.
Weakly-linking is almost always a better API choice, since it allows
users to query availability at runtime.

If there are multiple declarations of the same entity, the availability
attributes must either match on a per-platform basis or later
declarations must not have availability attributes for that
platform. For example:

```c
void g(void) __attribute__((availability(macos,introduced=10.4)));
void g(void) __attribute__((availability(macos,introduced=10.4))); // okay, matches
void g(void) __attribute__((availability(ios,introduced=4.0))); // okay, adds a new platform
void g(void); // okay, inherits both macos and ios availability from above.
void g(void) __attribute__((availability(macos,introduced=10.5))); // error: mismatch
```

When one method overrides another, the overriding method can be more widely available than the overridden method, e.g.,:

```objc
@interface A
- (id)method __attribute__((availability(macos,introduced=10.4)));
- (id)method2 __attribute__((availability(macos,introduced=10.4)));
@end

@interface B : A
- (id)method __attribute__((availability(macos,introduced=10.3))); // okay: method moved into base class later
- (id)method __attribute__((availability(macos,introduced=10.5))); // error: this method was available via the base class in 10.4
@end
```

Starting with the macOS 10.12 SDK, the `API_AVAILABLE` macro from
`<os/availability.h>` can simplify the spelling:

```objc
@interface A
- (id)method API_AVAILABLE(macos(10.11)));
- (id)otherMethod API_AVAILABLE(macos(10.11), ios(11.0));
@end
```

Availability attributes can also be applied using a `#pragma clang attribute`.
Any explicit availability attribute whose platform corresponds to the target
platform is applied to a declaration regardless of the availability attributes
specified in the pragma. For example, in the code below,
`hasExplicitAvailabilityAttribute` will use the `macOS` availability
attribute that is specified with the declaration, whereas
`getsThePragmaAvailabilityAttribute` will use the `macOS` availability
attribute that is applied by the pragma.

```c
#pragma clang attribute push (__attribute__((availability(macOS, introduced=10.12))), apply_to=function)
void getsThePragmaAvailabilityAttribute(void);
void hasExplicitAvailabilityAttribute(void) __attribute__((availability(macos,introduced=10.4)));
#pragma clang attribute pop
```

For platforms like `watchOS` and `tvOS`, whose availability attributes can
be implicitly inferred from an `iOS` availability attribute, the logic is
slightly more complex. The explicit and the pragma-applied availability
attributes whose platform corresponds to the target platform are applied as
described in the previous paragraph. However, the implicitly inferred attributes
are applied to a declaration only when there is no explicit or pragma-applied
availability attribute whose platform corresponds to the target platform. For
example, the function below will receive the `tvOS` availability from the
pragma rather than using the inferred `iOS` availability from the declaration:

```c
#pragma clang attribute push (__attribute__((availability(tvOS, introduced=12.0))), apply_to=function)
void getsThePragmaTVOSAvailabilityAttribute(void) __attribute__((availability(iOS,introduced=11.0)));
#pragma clang attribute pop
```

The compiler is also able to apply implicitly inferred attributes from a pragma
as well. For example, when targeting `tvOS`, the function below will receive
a `tvOS` availability attribute that is implicitly inferred from the `iOS`
availability attribute applied by the pragma:

```c
#pragma clang attribute push (__attribute__((availability(iOS, introduced=12.0))), apply_to=function)
void infersTVOSAvailabilityFromPragma(void);
#pragma clang attribute pop
```

The implicit attributes that are inferred from explicitly specified attributes
whose platform corresponds to the target platform are applied to the declaration
even if there is an availability attribute that can be inferred from a pragma.
For example, the function below will receive the `tvOS, introduced=11.0`
availability that is inferred from the attribute on the declaration rather than
inferring availability from the pragma:

```c
#pragma clang attribute push (__attribute__((availability(iOS, unavailable))), apply_to=function)
void infersTVOSAvailabilityFromAttributeNextToDeclaration(void)
  __attribute__((availability(iOS,introduced=11.0)));
#pragma clang attribute pop
```

Also see the documentation for
{ref}`@available <langext-objective-c-available>`
  }];
}

def PtrAuthDocs : Documentation {
  let Category = DocCatVariable;
  let Heading = "__ptrauth";
  let Content = [{
The `__ptrauth` qualifier allows the programmer to directly control
how pointers are signed when they are stored in a particular variable.
This can be used to strengthen the default protections of pointer
authentication and make it more difficult for an attacker to escalate
an ability to alter memory into full control of a process.

```c
#include <ptrauth.h>

typedef void (*my_callback)(const void*);
my_callback __ptrauth(ptrauth_key_process_dependent_code, 1, 0xe27a) callback;
```

The first argument to `__ptrauth` is the name of the signing key.
Valid key names for the target are defined in `<ptrauth.h>`.

The second argument to `__ptrauth` is a flag (0 or 1) specifying whether
the object should use address discrimination.

The third argument to `__ptrauth` is a 16-bit non-negative integer which
allows additional discrimination between objects.
  }];
}

def ExternalSourceSymbolDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `external_source_symbol` attribute specifies that a declaration originates
from an external source and describes the nature of that source.

The fact that Clang is capable of recognizing declarations that were defined
externally can be used to provide better tooling support for mixed-language
projects or projects that rely on auto-generated code. For instance, an IDE that
uses Clang and that supports mixed-language projects can use this attribute to
provide a correct 'jump-to-definition' feature. For a concrete example,
consider a protocol that's defined in a Swift file:

```swift
@objc public protocol SwiftProtocol {
  func method()
}
```

This protocol can be used from Objective-C code by including a header file that
was generated by the Swift compiler. The declarations in that header can use
the `external_source_symbol` attribute to make Clang aware of the fact
that `SwiftProtocol` actually originates from a Swift module:

```objc
__attribute__((external_source_symbol(language="Swift",defined_in="module")))
@protocol SwiftProtocol
@required
- (void) method;
@end
```

Consequently, when 'jump-to-definition' is performed at a location that
references `SwiftProtocol`, the IDE can jump to the original definition in
the Swift source file rather than jumping to the Objective-C declaration in the
auto-generated header file.

The `external_source_symbol` attribute is a comma-separated list that includes
clauses that describe the origin and the nature of the particular declaration.
Those clauses can be:

language=*string-literal*

: The name of the source language in which this declaration was defined.

defined_in=*string-literal*

: The name of the source container in which the declaration was defined. The
  exact definition of source container is language-specific, e.g. Swift's
  source containers are modules, so `defined_in` should specify the Swift
  module name.

USR=*string-literal*

: String that specifies a unified symbol resolution (USR) value for this
  declaration. USR string uniquely identifies this particular declaration, and
  is typically used when constructing an index of a codebase.
  The USR value in this attribute is expected to be generated by an external
  compiler that compiled the native declaration using its original source
  language. The exact format of the USR string and its other attributes
  are determined by the specification of this declaration's source language.
  When not specified, Clang's indexer will use the Clang USR for this symbol.
  User can query to see if Clang supports the use of the `USR` clause in
  the `external_source_symbol` attribute with
  `__has_attribute(external_source_symbol) >= 20230206`.

generated_declaration

: This declaration was automatically generated by some tool.

The clauses can be specified in any order. The clauses that are listed above are
all optional, but the attribute has to have at least one clause.
  }];
}

def ConstInitDocs : Documentation {
  let Category = DocCatVariable;
  let Heading = "require_constant_initialization, constinit (C++20)";
  let Content = [{
This attribute specifies that the variable to which it is attached is intended
to have a [constant initializer](http://en.cppreference.com/w/cpp/language/constant_initialization)
according to the rules of [basic.start.static]. The variable is required to
have static or thread storage duration. If the initialization of the variable
is not a constant initializer an error will be produced. This attribute may
only be used in C++; the `constinit` spelling is only accepted in C++20
onwards.

Note that in C++03 strict constant expression checking is not done. Instead
the attribute reports if Clang can emit the variable as a constant, even if it's
not technically a 'constant initializer'. This behavior is non-portable.

Static storage duration variables with constant initializers avoid hard-to-find
bugs caused by the indeterminate order of dynamic initialization. They can also
be safely used during dynamic initialization across translation units.

This attribute acts as a compile time assertion that the requirements
for constant initialization have been met. Since these requirements change
between dialects and have subtle pitfalls it's important to fail fast instead
of silently falling back on dynamic initialization.

The first use of the attribute on a variable must be part of, or precede, the
initializing declaration of the variable. C++20 requires the `constinit`
spelling of the attribute to be present on the initializing declaration if it
is used anywhere. The other spellings can be specified on a forward declaration
and omitted on a later initializing declaration.

```c++
// -std=c++14
#define SAFE_STATIC [[clang::require_constant_initialization]]
struct T {
  constexpr T(int) {}
  ~T(); // non-trivial
};
SAFE_STATIC T x = {42}; // Initialization OK. Doesn't check destructor.
SAFE_STATIC T y = 42; // error: variable does not have a constant initializer
// copy initialization is not a constant expression on a non-literal type.
```
  }];
}

def WarnMaybeUnusedDocs : Documentation {
  let Category = DocCatVariable;
  let Heading = "maybe_unused, unused";
  let Content = [{
When passing the `-Wunused` flag to Clang, entities that are unused by the
program may be diagnosed. The `[[maybe_unused]]` (or
`__attribute__((unused))`) attribute can be used to silence such diagnostics
when the entity cannot be removed. For instance, a local variable may exist
solely for use in an `assert()` statement, which makes the local variable
unused when `NDEBUG` is defined.

The attribute may be applied to the declaration of a class, a typedef, a
variable, a function or method, a function parameter, an enumeration, an
enumerator, a non-static data member, or a label.

```c++
#include <cassert>

[[maybe_unused]] void f([[maybe_unused]] bool thing1,
                        [[maybe_unused]] bool thing2) {
  [[maybe_unused]] bool b = thing1 && thing2;
  assert(b);
}
```
  }];
}

def WarnUnusedResultsDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "nodiscard, warn_unused_result";
  let Content  = [{
Clang supports the ability to diagnose when the results of a function call
expression are discarded under suspicious circumstances. A diagnostic is
generated when a function or its return type is marked with `[[nodiscard]]`
(or `__attribute__((warn_unused_result))`) and the function call appears as a
potentially-evaluated discarded-value expression that is not explicitly cast to
`void`.

A string literal may optionally be provided to the attribute, which will be
reproduced in any resulting diagnostics. Redeclarations using different forms
of the attribute (with or without the string literal or with different string
literal contents) are allowed. If there are redeclarations of the entity with
differing string literals, it is unspecified which one will be used by Clang
in any resulting diagnostics.

```c++
struct [[nodiscard]] error_info { /*...*/ };
error_info enable_missile_safety_mode();

void launch_missiles();
void test_missiles() {
  enable_missile_safety_mode(); // diagnoses
  launch_missiles();
}
error_info &foo();
void f() { foo(); } // Does not diagnose, error_info is a reference.
```

Additionally, discarded temporaries resulting from a call to a constructor
marked with `[[nodiscard]]` or a constructor of a type marked
`[[nodiscard]]` will also diagnose. This also applies to type conversions that
use the annotated `[[nodiscard]]` constructor or result in an annotated type.

```c++
struct [[nodiscard]] marked_type {/*..*/ };
struct marked_ctor {
  [[nodiscard]] marked_ctor();
  marked_ctor(int);
};

struct S {
  operator marked_type() const;
  [[nodiscard]] operator int() const;
};

void usages() {
  marked_type(); // diagnoses.
  marked_ctor(); // diagnoses.
  marked_ctor(3); // Does not diagnose, int constructor isn't marked nodiscard.

  S s;
  static_cast<marked_type>(s); // diagnoses
  (int)s; // diagnoses
}
```
  }];
}

def FallthroughDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "fallthrough";
  let Content = [{
The `fallthrough` (or `clang::fallthrough`) attribute is used
to annotate intentional fall-through
between switch labels. It can only be applied to a null statement placed at a
point of execution between any statement and the next switch label. It is
common to mark these places with a specific comment, but this attribute is
meant to replace comments with a more strict annotation, which can be checked
by the compiler. This attribute doesn't change semantics of the code and can
be used wherever an intended fall-through occurs. It is designed to mimic
control-flow statements like `break;`, so it can be placed in most places
where `break;` can, but only if there are no statements on the execution path
between it and the next switch label.

By default, Clang does not warn on unannotated fallthrough from one `switch`
case to another. Diagnostics on fallthrough without a corresponding annotation
can be enabled with the `-Wimplicit-fallthrough` argument.

Here is an example:

```c++
// compile with -Wimplicit-fallthrough
switch (n) {
case 22:
case 33:  // no warning: no statements between case labels
  f();
case 44:  // warning: unannotated fall-through
  g();
  [[clang::fallthrough]];
case 55:  // no warning
  if (x) {
    h();
    break;
  }
  else {
    i();
    [[clang::fallthrough]];
  }
case 66:  // no warning
  p();
  [[clang::fallthrough]]; // warning: fallthrough annotation does not
                          //          directly precede case label
  q();
case 77:  // warning: unannotated fall-through
  r();
}
```
  }];
}

def CXXAssumeDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "assume";
  let Content = [{
The `assume` attribute is used to indicate to the optimizer that a
certain condition is assumed to be true at a certain point in the
program. If this condition is violated at runtime, the behavior is
undefined. `assume` can only be applied to a null statement.

Different optimisers are likely to react differently to the presence of
this attribute; in some cases, adding `assume` may affect performance
negatively. It should be used with parsimony and care.

Example:

```c++
int f(int x, int y) {
  [[assume(x == 27)]];
  [[assume(x == y)]];
  return y + 1; // May be optimised to `return 28`.
}
```
  }];
}

def LikelihoodDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "likely and unlikely";
  let Content = [{
The `likely` and `unlikely` attributes are used as compiler hints.
The attributes are used to aid the compiler to determine which branch is
likely or unlikely to be taken. This is done by marking the branch substatement
with one of the two attributes.

It isn't allowed to annotate a single statement with both `likely` and
`unlikely`. Annotating the `true` and `false` branch of an `if`
statement with the same likelihood attribute will result in a diagnostic and
the attributes are ignored on both branches.

In a `switch` statement it's allowed to annotate multiple `case` labels
or the `default` label with the same likelihood attribute. This makes
\* all labels without an attribute have a neutral likelihood,
\* all labels marked `[[likely]]` have an equally positive likelihood, and
\* all labels marked `[[unlikely]]` have an equally negative likelihood.
The neutral likelihood is the more likely of path execution than the negative
likelihood. The positive likelihood is the more likely of path of execution
than the neutral likelihood.

These attributes have no effect on the generated code when using
PGO (Profile-Guided Optimization) or at optimization level 0.

In Clang, the attributes will be ignored if they're not placed on
\* the `case` or `default` label of a `switch` statement,
\* or on the substatement of an `if` or `else` statement,
\* or on the substatement of an `for` or `while` statement.
The C++ Standard recommends to honor them on every statement in the
path of execution, but that can be confusing:

```c++
if (b) {
  [[unlikely]] --b; // Per the standard this is in the path of
                    // execution, so this branch should be considered
                    // unlikely. However, Clang ignores the attribute
                    // here since it is not on the substatement.
}

if (b) {
  --b;
  if(b)
    return;
  [[unlikely]] --b; // Not in the path of execution,
}                   // the branch has no likelihood information.

if (b) {
  --b;
  foo(b);
  // Whether or not the next statement is in the path of execution depends
  // on the declaration of foo():
  // In the path of execution: void foo(int);
  // Not in the path of execution: [[noreturn]] void foo(int);
  // This means the likelihood of the branch depends on the declaration
  // of foo().
  [[unlikely]] --b;
}
```

Below are some example usages of the likelihood attributes and their effects:

```c++
if (b) [[likely]] { // Placement on the first statement in the branch.
  // The compiler will optimize to execute the code here.
} else {
}

if (b)
  [[unlikely]] b++; // Placement on the first statement in the branch.
else {
  // The compiler will optimize to execute the code here.
}

if (b) {
  [[unlikely]] b++; // Placement on the second statement in the branch.
}                   // The attribute will be ignored.

if (b) [[likely]] {
  [[unlikely]] b++; // No contradiction since the second attribute
}                   // is ignored.

if (b)
  ;
else [[likely]] {
  // The compiler will optimize to execute the code here.
}

if (b)
  ;
else
  // The compiler will optimize to execute the next statement.
  [[likely]] b = f();

if (b) [[likely]]; // Both branches are likely. A diagnostic is issued
else [[likely]];   // and the attributes are ignored.

if (b)
  [[likely]] int i = 5; // Issues a diagnostic since the attribute
                        // isn't allowed on a declaration.

switch (i) {
  [[likely]] case 1:    // This value is likely
    ...
    break;

  [[unlikely]] case 2:  // This value is unlikely
    ...
    [[fallthrough]];

  case 3:               // No likelihood attribute
    ...
    [[likely]] break;   // No effect

  case 4: [[likely]] {  // attribute on substatement has no effect
    ...
    break;
    }

  [[unlikely]] default: // All other values are unlikely
    ...
    break;
}

switch (i) {
  [[likely]] case 0:    // This value and code path is likely
    ...
    [[fallthrough]];

  case 1:               // No likelihood attribute, code path is neutral
    break;              // falling through has no effect on the likelihood

  case 2:               // No likelihood attribute, code path is neutral
    [[fallthrough]];

  [[unlikely]] default: // This value and code path are both unlikely
    break;
}

for(int i = 0; i != size; ++i) [[likely]] {
  ...               // The loop is the likely path of execution
}

for(const auto &E : Elements) [[likely]] {
  ...               // The loop is the likely path of execution
}

while(i != size) [[unlikely]] {
  ...               // The loop is the unlikely path of execution
}                   // The generated code will optimize to skip the loop body

while(true) [[unlikely]] {
  ...               // The attribute has no effect
}                   // Clang elides the comparison and generates an infinite
                    // loop
```

  }];
}

def ARMInterruptDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "interrupt (ARM)";
  let Content = [{
Clang supports the GNU style `__attribute__((interrupt("TYPE")))` attribute on
ARM targets. This attribute may be attached to a function definition and
instructs the backend to generate appropriate function entry/exit code so that
it can be used directly as an interrupt service routine.

The parameter passed to the interrupt attribute is optional, but if
provided it must be a string literal with one of the following values: "IRQ",
"FIQ", "SWI", "ABORT", "UNDEF".

The semantics are as follows:

- If the function is AAPCS, Clang instructs the backend to realign the stack to
  8 bytes on entry. This is a general requirement of the AAPCS at public
  interfaces, but may not hold when an exception is taken. Doing this allows
  other AAPCS functions to be called.

- If the CPU is M-class this is all that needs to be done since the architecture
  itself is designed in such a way that functions obeying the normal AAPCS ABI
  constraints are valid exception handlers.

- If the CPU is not M-class, the prologue and epilogue are modified to save all
  non-banked registers that are used, so that upon return the user-mode state
  will not be corrupted. Note that to avoid unnecessary overhead, only
  general-purpose (integer) registers are saved in this way. If VFP operations
  are needed, that state must be saved manually.

  Specifically, interrupt kinds other than "FIQ" will save all core registers
  except "lr" and "sp". "FIQ" interrupts will save r0-r7.

- If the CPU is not M-class, the return instruction is changed to one of the
  canonical sequences permitted by the architecture for exception return. Where
  possible the function itself will make the necessary "lr" adjustments so that
  the "preferred return address" is selected.

  Unfortunately the compiler is unable to make this guarantee for an "UNDEF"
  handler, where the offset from "lr" to the preferred return address depends on
  the execution state of the code which generated the exception. In this case
  a sequence equivalent to "movs pc, lr" will be used.
  }];
}

def ARMInterruptSaveFPDocs : Documentation {
    let Category = DocCatFunction;
  let Heading = "interrupt_save_fp (ARM)";
  let Content = [{
Clang supports the GNU style `__attribute__((interrupt_save_fp("TYPE")))`
on ARM targets. This attribute behaves the same way as the ARM interrupt
attribute, except the general purpose floating point registers are also saved,
along with FPEXC and FPSCR. Note, even on M-class CPUs, where the floating
point context can be automatically saved depending on the FPCCR, the general
purpose floating point registers will be saved.
  }];
}

def BPFPreserveAccessIndexDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((preserve_access_index))`
attribute for the BPF target. This attribute may be attached to a
struct or union declaration, where if -g is specified, it enables
preserving struct or union member access debuginfo indices of this
struct or union, similar to clang `__builtin_preserve_access_index()`.
  }];
}

def BPFPreserveStaticOffsetDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((preserve_static_offset))`
attribute for the BPF target. This attribute may be attached to a
struct or union declaration. Reading or writing fields of types having
such annotation is guaranteed to generate LDX/ST/STX instruction with
offset corresponding to the field.

For example:

```c
struct foo {
  int a;
  int b;
};

struct bar {
  int a;
  struct foo b;
} __attribute__((preserve_static_offset));

void buz(struct bar *g) {
  g->b.a = 42;
}
```

The assignment to `g`'s field would produce an ST instruction with
offset 8: `*(u32)(r1 + 8) = 42;`.

Without this attribute generated instructions might be different,
depending on optimizations behavior. E.g. the example above could be
rewritten as `r1 += 8; *(u32)(r1 + 0) = 42;`.
  }];
}

def BTFDeclTagDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((btf_decl_tag("ARGUMENT")))` attribute for
all targets. This attribute may be attached to a struct/union, struct/union
field, function, function parameter, variable or typedef declaration. If -g is
specified, the `ARGUMENT` info will be preserved in IR and be emitted to
dwarf. For BPF targets, the `ARGUMENT` info will be emitted to .BTF ELF
section too.
  }];
}

def BTFTypeTagDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
Clang supports the `__attribute__((btf_type_tag("ARGUMENT")))` attribute for
all targets. It only has effect when `-g` is specified on the command line.

The attribute can be applied to a pointer type, in which case the tag is
associated with the pointee type, e.g.:

```c
int __attribute__((btf_type_tag("tag"))) *p;
```

It can also be applied to the underlying type of a typedef, in which case the
tag follows the typedef down to its base type, e.g.:

```c
typedef struct foo __attribute__((btf_type_tag("tag"))) foo_t;
```

The following is the corresponding btf:

```
...
[2] TYPE_TAG 'tag' type_id=4
[3] TYPEDEF 'foo_t' type_id=2
[4] STRUCT 'foo' size=4 vlen=1
    'c' type_id=5 bits_offset=0
[5] INT 'int' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
...
```

The attribute is currently silently ignored in any other position (note: this
scenario may be diagnosed in the future).

The `ARGUMENT` string will be preserved in IR and emitted to DWARF for the
types used in variable declarations, function declarations, or typedef
declarations.

For BPF targets, the `ARGUMENT` string will also be emitted to .BTF ELF
section.
  }];
}

def BPFFastCallDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
Functions annotated with this attribute are likely to be inlined by BPF JIT.
It is assumed that inlined implementation uses less caller saved registers,
than a regular function.
Specifically, the following registers are likely to be preserved:
- `R0` if function return value is `void`;
- `R2-R5` if function takes 1 argument;
- `R3-R5` if function takes 2 arguments;
- `R4-R5` if function takes 3 arguments;
- `R5` if function takes 4 arguments;

For such functions Clang generates code pattern that allows BPF JIT
to recognize and remove unnecessary spills and fills of the preserved
registers.
  }];
}

def MipsInterruptDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "interrupt (MIPS)";
  let Content = [{
Clang supports the GNU style `__attribute__((interrupt("ARGUMENT")))` attribute on
MIPS targets. This attribute may be attached to a function definition and instructs
the backend to generate appropriate function entry/exit code so that it can be used
directly as an interrupt service routine.

By default, the compiler will produce a function prologue and epilogue suitable for
an interrupt service routine that handles an External Interrupt Controller (eic)
generated interrupt. This behavior can be explicitly requested with the "eic"
argument.

Otherwise, for use with vectored interrupt mode, the argument passed should be
of the form "vector=LEVEL" where LEVEL is one of the following values:
"sw0", "sw1", "hw0", "hw1", "hw2", "hw3", "hw4", "hw5". The compiler will
then set the interrupt mask to the corresponding level which will mask all
interrupts up to and including the argument.

The semantics are as follows:

- The prologue is modified so that the Exception Program Counter (EPC) and
  Status coprocessor registers are saved to the stack. The interrupt mask is
  set so that the function can only be interrupted by a higher priority
  interrupt. The epilogue will restore the previous values of EPC and Status.
- The prologue and epilogue are modified to save and restore all non-kernel
  registers as necessary.
- The FPU is disabled in the prologue, as the floating pointer registers are not
  spilled to the stack.
- The function return sequence is changed to use an exception return instruction.
- The parameter sets the interrupt mask for the function corresponding to the
  interrupt level specified. If no mask is specified the interrupt mask
  defaults to "eic".
  }];
}

def MicroMipsDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the GNU style `__attribute__((micromips))` and
`__attribute__((nomicromips))` attributes on MIPS targets. These attributes
may be attached to a function definition and instructs the backend to generate
or not to generate microMIPS code for that function.

These attributes override the `-mmicromips` and `-mno-micromips` options
on the command line.
  }];
}

def MipsLongCallStyleDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "long_call, far";
  let Content = [{
Clang supports the `__attribute__((long_call))`, `__attribute__((far))`,
and `__attribute__((near))` attributes on MIPS targets. These attributes may
only be added to function declarations and change the code generated
by the compiler when directly calling the function. The `near` attribute
allows calls to the function to be made using the `jal` instruction, which
requires the function to be located in the same naturally aligned 256MB
segment as the caller. The `long_call` and `far` attributes are synonyms
and require the use of a different call sequence that works regardless
of the distance between the functions.

These attributes have no effect for position-independent code.

These attributes take priority over command line switches such
as `-mlong-calls` and `-mno-long-calls`.
  }];
}

def MipsShortCallStyleDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "short_call, near";
  let Content = [{
Clang supports the `__attribute__((long_call))`, `__attribute__((far))`,
`__attribute__((short__call))`, and `__attribute__((near))` attributes
on MIPS targets. These attributes may only be added to function declarations
and change the code generated by the compiler when directly calling
the function. The `short_call` and `near` attributes are synonyms and
allow calls to the function to be made using the `jal` instruction, which
requires the function to be located in the same naturally aligned 256MB segment
as the caller. The `long_call` and `far` attributes are synonyms and
require the use of a different call sequence that works regardless
of the distance between the functions.

These attributes have no effect for position-independent code.

These attributes take priority over command line switches such
as `-mlong-calls` and `-mno-long-calls`.
  }];
}

def RISCVInterruptDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "interrupt (RISC-V)";
  let Content = [{
Clang supports the GNU style `__attribute__((interrupt))` attribute on RISCV
targets. This attribute may be attached to a function definition and instructs
the backend to generate appropriate function entry/exit code so that it can be
used directly as an interrupt service routine.

Permissible values for this parameter are `machine`, `supervisor`,
`rnmi`, `qci-nest`, `qci-nonest`, `SiFive-CLIC-preemptible`, and
`SiFive-CLIC-stack-swap`. If there is no parameter, then it defaults to
`machine`.

The `rnmi` value is used for resumable non-maskable interrupts. It requires the
standard Smrnmi extension.

The `qci-nest` and `qci-nonest` values require Qualcomm's Xqciint extension
and are used for Machine-mode Interrupts and Machine-mode Non-maskable
interrupts. These use the following instructions from Xqciint to save and
restore interrupt state to the stack -- the `qci-nest` value will use
`qc.c.mienter.nest` and the `qci-nonest` value will use `qc.c.mienter` to
begin the interrupt handler. Both of these will use `qc.c.mileaveret` to
restore the state and return to the previous context.

The `SiFive-CLIC-preemptible` and `SiFive-CLIC-stack-swap` values are used
for machine-mode interrupts. For `SiFive-CLIC-preemptible` interrupts, the
values of `mcause` and `mepc` are saved onto the stack, and interrupts are
re-enabled. For `SiFive-CLIC-stack-swap` interrupts, the stack pointer is
swapped with `mscratch` before its first use and after its last use.

The SiFive CLIC values may be combined with each other and with the `machine`
attribute value. Any other combination of different values is not allowed.

Repeated interrupt attribute on the same declaration will cause a warning
to be emitted. In case of repeated declarations, the last one prevails.

Refer to:
<https://gcc.gnu.org/onlinedocs/gcc/RISC-V-Function-Attributes.html>
<https://riscv.org/specifications/privileged-isa/>
The RISC-V Instruction Set Manual Volume II: Privileged Architecture
Version 1.10.
<https://github.com/quic/riscv-unified-db/releases/tag/Xqci-0.13.0>
<https://sifive.cdn.prismic.io/sifive/d1984d2b-c9b9-4c91-8de0-d68a5e64fa0f_sifive-interrupt-cookbook-v1p2.pdf>
  }];
}

def RISCVRVVVectorBitsDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
On RISC-V targets, the `riscv_rvv_vector_bits(N)` attribute is used to define
fixed-length variants of sizeless types.

For example:

```c
#include <riscv_vector.h>

#if defined(__riscv_v_fixed_vlen)
typedef vint8m1_t fixed_vint8m1_t __attribute__((riscv_rvv_vector_bits(__riscv_v_fixed_vlen)));
#endif
```

Creates a type `fixed_vint8m1_t` that is a fixed-length variant of
`vint8m1_t` that contains exactly 512 bits. Unlike `vint8m1_t`, this type
can be used in globals, structs, unions, and arrays, all of which are
unsupported for sizeless types.

The attribute can be attached to a single RVV vector (such as `vint8m1_t`).
The attribute will be rejected unless
`N==(__riscv_v_fixed_vlen*LMUL)`, the implementation defined feature macro that
is enabled under the `-mrvv-vector-bits` flag. `__riscv_v_fixed_vlen` can
only be a power of 2 between 64 and 65536.

For types where LMUL!=1, `__riscv_v_fixed_vlen` needs to be scaled by the LMUL
of the type before passing to the attribute.

For `vbool*_t` types, `__riscv_v_fixed_vlen` needs to be divided by the
number from the type name. For example, `vbool8_t` needs to use
`__riscv_v_fixed_vlen` / 8. If the resulting value is not a multiple of 8,
the type is not supported for that value of `__riscv_v_fixed_vlen`.
}];
}

def AVRInterruptDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "interrupt (AVR)";
  let Content = [{
Clang supports the GNU style `__attribute__((interrupt))` attribute on
AVR targets. This attribute may be attached to a function definition and instructs
the backend to generate appropriate function entry/exit code so that it can be used
directly as an interrupt service routine.

On the AVR, the hardware globally disables interrupts when an interrupt is executed.
The first instruction of an interrupt handler declared with this attribute is a SEI
instruction to re-enable interrupts. See also the signal attribute that
does not insert a SEI instruction.
  }];
}

def AVRSignalDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the GNU style `__attribute__((signal))` attribute on
AVR targets. This attribute may be attached to a function definition and instructs
the backend to generate appropriate function entry/exit code so that it can be used
directly as an interrupt service routine.

Interrupt handler functions defined with the signal attribute do not re-enable interrupts.
}];
}

def TargetDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the GNU style `__attribute__((target("OPTIONS")))` attribute.
This attribute may be attached to a function definition and instructs
the backend to use different code generation options than were passed on the
command line.

The current set of options correspond to the existing "subtarget features" for
the target with or without a "-mno-" in front corresponding to the absence
of the feature, as well as `arch="CPU"` which will change the default "CPU"
for the function.

For X86, the attribute also allows `tune="CPU"` to optimize the generated
code for the given CPU without changing the available instructions.

For AArch64, `arch="Arch"` will set the architecture, similar to the -march
command line options. `cpu="CPU"` can be used to select a specific cpu,
as per the `-mcpu` option, similarly for `tune=`. The attribute also allows the
"branch-protection=\<args>" option, where the permissible arguments and their
effect on code generation are the same as for the command-line option
`-mbranch-protection`.

Example "subtarget features" from the x86 backend include: "mmx", "sse", "sse4.2",
"avx", "xop" and largely correspond to the machine specific options handled by
the front end.

Note that this attribute does not apply transitively to nested functions such
as blocks or C++ lambdas.

Additionally, this attribute supports function multiversioning for ELF based
x86/x86-64 targets, which can be used to create multiple implementations of the
same function that will be resolved at runtime based on the priority of their
`target` attribute strings. A function is considered a multiversioned function
if either two declarations of the function have different `target` attribute
strings, or if it has a `target` attribute string of `default`. For
example:

```c++
__attribute__((target("arch=atom")))
void foo() {} // will be called on 'atom' processors.
__attribute__((target("default")))
void foo() {} // will be called on any other processors.
```

All multiversioned functions must contain a `default` (fallback)
implementation, otherwise usages of the function are considered invalid.
Additionally, a function may not become multiversioned after its first use.
}];
}

def TargetVersionDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
For AArch64 target clang supports function multiversioning by
`__attribute__((target_version("OPTIONS")))` attribute. When applied to a
function it instructs compiler to emit multiple function versions based on
`target_version` attribute strings, which resolved at runtime depend on their
priority and target features availability. One of the versions is always
( implicitly or explicitly ) the `default` (fallback). Attribute strings can
contain dependent features names joined by the "+" sign.

For targets that support the GNU indirect function (IFUNC) feature, dispatch
is performed by emitting an indirect function that is resolved to the appropriate
target clone at load time. The indirect function is given the name the
multiversioned function would have if it had been declared without the attribute.
For backward compatibility with earlier Clang releases, a function alias with an
`.ifunc` suffix is also emitted. The `.ifunc` suffixed symbol is a deprecated
feature and support for it may be removed in the future.
}];
}

def TargetClonesDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `target_clones("OPTIONS")` attribute. This attribute may be
attached to a function declaration and causes function multiversioning, where
multiple versions of the function will be emitted with different code
generation options. Additionally, these versions will be resolved at runtime
based on the priority of their attribute options. All `target_clone` functions
are considered multiversioned functions.

For AArch64 target:
The attribute contains comma-separated strings of target features joined by "+"
sign. For example:

```c++
__attribute__((target_clones("sha2+memtag", "fcma+sve2-pmull128")))
void foo() {}
```

For every multiversioned function a `default` (fallback) implementation
always generated if not specified directly.

For x86/x86-64 targets:
All multiversioned functions must contain a `default` (fallback)
implementation, otherwise usages of the function are considered invalid.
Additionally, a function may not become multiversioned after its first use.

The options to `target_clones` can either be a target-specific architecture
(specified as `arch=CPU`), or one of a list of subtarget features.

Example "subtarget features" from the x86 backend include: "mmx", "sse", "sse4.2",
"avx", "xop" and largely correspond to the machine specific options handled by
the front end.

The versions can either be listed as a comma-separated sequence of string
literals or as a single string literal containing a comma-separated list of
versions. For compatibility with GCC, the two formats can be mixed. For
example, the following will emit 4 versions of the function:

```c++
__attribute__((target_clones("arch=atom,avx2","arch=ivybridge","default")))
void foo() {}
```

For targets that support the GNU indirect function (IFUNC) feature, dispatch
is performed by emitting an indirect function that is resolved to the appropriate
target clone at load time. The indirect function is given the name the
multiversioned function would have if it had been declared without the attribute.
For backward compatibility with earlier Clang releases, a function alias with an
`.ifunc` suffix is also emitted. The `.ifunc` suffixed symbol is a deprecated
feature and support for it may be removed in the future.

For PowerPC targets, `target_clones` is supported on AIX only. Only CPU
(specified as `cpu=CPU`) and `default` options are allowed. IFUNC is supported
on AIX in Clang, so dispatch is implemented similar to other targets using IFUNC.
An FMV function that is only declared in a translation unit is treated as a
non-FMV. The resolver and the function clones are given internal linkage.
}];
}

def MinVectorWidthDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((min_vector_width(width)))` attribute. This
attribute may be attached to a function and informs the backend that this
function desires vectors of at least this width to be generated. Target-specific
maximum vector widths still apply. This means even if you ask for something
larger than the target supports, you will only get what the target supports.
This attribute is meant to be a hint to control target heuristics that may
generate narrower vectors than what the target hardware supports.

This is currently used by the X86 target to allow some CPUs that support 512-bit
vectors to be limited to using 256-bit vectors to avoid frequency penalties.
This is currently enabled with the `-prefer-vector-width=256` command line
option. The `min_vector_width` attribute can be used to prevent the backend
from trying to split vector operations to match the `prefer-vector-width`. All
X86 vector intrinsics from x86intrin.h already set this attribute. Additionally,
use of any of the X86-specific vector builtins will implicitly set this
attribute on the calling function. The intent is that explicitly writing vector
code using the X86 intrinsics will prevent `prefer-vector-width` from
affecting the code.
}];
}

def DocCatAMDGPUAttributes : DocumentationCategory<"AMD GPU Attributes">;

def AMDGPUFlatWorkGroupSizeDocs : Documentation {
  let Category = DocCatAMDGPUAttributes;
  let Content = [{
The flat work-group size is the number of work-items in the work-group size
specified when the kernel is dispatched. It is the product of the sizes of the
x, y, and z dimension of the work-group.

Clang supports the
`__attribute__((amdgpu_flat_work_group_size(<min>, <max>)))` attribute for the
AMDGPU target. This attribute may be attached to a kernel function definition
and is an optimization hint.

`<min>` parameter specifies the minimum flat work-group size, and `<max>`
parameter specifies the maximum flat work-group size (must be greater than
`<min>`) to which all dispatches of the kernel will conform. Passing `0, 0`
as `<min>, <max>` implies the default behavior (`128, 256`).

If specified, the AMDGPU target backend might be able to produce better machine
code for barriers and perform scratch promotion by estimating available group
segment size.

An error will be given if:
: - Specified values violate subtarget specifications;
  - Specified values are not compatible with values provided through other
    attributes.
  }];
}

def AMDGPUWavesPerEUDocs : Documentation {
  let Category = DocCatAMDGPUAttributes;
  let Content = [{
A compute unit (CU) is responsible for executing the wavefronts of a work-group.
It is composed of one or more execution units (EU), which are responsible for
executing the wavefronts. An EU can have enough resources to maintain the state
of more than one executing wavefront. This allows an EU to hide latency by
switching between wavefronts in a similar way to symmetric multithreading on a
CPU. In order to allow the state for multiple wavefronts to fit on an EU, the
resources used by a single wavefront have to be limited. For example, the number
of SGPRs and VGPRs. Limiting such resources can allow greater latency hiding,
but can result in having to spill some register state to memory.

Clang supports the `__attribute__((amdgpu_waves_per_eu(<min>[, <max>])))`
attribute for the AMDGPU target. This attribute may be attached to a kernel
function definition and is an optimization hint.

`<min>` parameter specifies the requested minimum number of waves per EU, and
*optional* `<max>` parameter specifies the requested maximum number of waves
per EU (must be greater than `<min>` if specified). If `<max>` is omitted,
then there is no restriction on the maximum number of waves per EU other than
the one dictated by the hardware for which the kernel is compiled. Passing
`0, 0` as `<min>, <max>` implies the default behavior (no limits).

If specified, this attribute allows an advanced developer to tune the number of
wavefronts that are capable of fitting within the resources of an EU. The AMDGPU
target backend can use this information to limit resources, such as number of
SGPRs, number of VGPRs, size of available group and private memory segments, in
such a way that guarantees that at least `<min>` wavefronts and at most
`<max>` wavefronts are able to fit within the resources of an EU. Requesting
more wavefronts can hide memory latency but limits available registers which
can result in spilling. Requesting fewer wavefronts can help reduce cache
thrashing, but can reduce memory latency hiding.

This attribute controls the machine code generated by the AMDGPU target backend
to ensure it is capable of meeting the requested values. However, when the
kernel is executed, there may be other reasons that prevent meeting the request,
for example, there may be wavefronts from other kernels executing on the EU.

An error will be given if:
: - Specified values violate subtarget specifications;
  - Specified values are not compatible with values provided through other
    attributes;

The AMDGPU target backend will emit a warning whenever it is unable to
create machine code that meets the request.
  }];
}

def LaunchBoundsDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The ``__launch_bounds__`` attribute (also spelled ``launch_bounds``) originates
in CUDA. It informs the compiler of the launch configuration a kernel will be
dispatched with, allowing it to optimize the kernel accordingly. It takes the
form ``__launch_bounds__(<max-threads-per-block>[,
<min-blocks-per-multiprocessor>[, <max-blocks-per-cluster>]])``. All arguments
are constant expressions.

The attribute only takes effect on ``__global__`` (kernel) functions; like
NVCC, Clang ignores it on any other function.

``<max-threads-per-block>`` specifies the maximum number of threads per block
the kernel will be launched with. ``<min-blocks-per-multiprocessor>`` specifies
the desired minimum number of blocks resident per multiprocessor, and
``<max-blocks-per-cluster>`` the maximum number of blocks per cluster.

For the NVPTX target, ``<max-threads-per-block>`` and
``<min-blocks-per-multiprocessor>`` map to the ``.maxntid`` and ``.minnctapersm``
PTX directives, respectively, and ``<max-blocks-per-cluster>`` (which requires
``sm_90`` or newer) maps to ``.maxclusterrank``.

For the AMDGPU target, the attribute is translated into the equivalent AMDGPU
kernel attributes:

  - ``<max-threads-per-block>`` sets the maximum
    ``amdgpu_flat_work_group_size`` (as ``1, <max-threads-per-block>``).
  - ``<min-blocks-per-multiprocessor>`` sets the minimum
    ``amdgpu_waves_per_eu``. Note that HIP reinterprets this CUDA argument as a
    minimum number of waves per execution unit, so its meaning differs from the
    NVPTX interpretation.
  - ``<max-blocks-per-cluster>`` is currently ignored.

An explicit ``amdgpu_flat_work_group_size`` or ``amdgpu_waves_per_eu`` attribute
takes precedence over the value derived from ``__launch_bounds__``.

When the same kernel is declared multiple times, the launch bounds from the most
recent declaration that specifies them are used; a definition without
``__launch_bounds__`` inherits the bounds from an earlier declaration.
  }];
}

def AMDGPUNumSGPRNumVGPRDocs : Documentation {
  let Category = DocCatAMDGPUAttributes;
  let Content = [{
:::{warning}
These attributes are deprecated. Use the `amdgpu_waves_per_eu` attribute to
control SGPR and VGPR usage instead.
:::

Clang supports the `__attribute__((amdgpu_num_sgpr(<num_sgpr>)))` and
`__attribute__((amdgpu_num_vgpr(<num_vgpr>)))` attributes for the AMDGPU
target. These attributes may be attached to a kernel function definition and are
an optimization hint.

If these attributes are specified, then the AMDGPU target backend will attempt
to limit the number of SGPRs and/or VGPRs used to the specified value(s). The
number of used SGPRs and/or VGPRs may further be rounded up to satisfy the
allocation requirements or constraints of the subtarget. Passing `0` as
`num_sgpr` and/or `num_vgpr` implies the default behavior (no limits).

These attributes can be used to test the AMDGPU target backend. It is
recommended that the `amdgpu_waves_per_eu` attribute be used to control
resources such as SGPRs and VGPRs since it is aware of the limits for different
subtargets.

An error will be given if:
: - Specified values violate subtarget specifications;
  - Specified values are not compatible with values provided through other
    attributes;
  - The AMDGPU target backend is unable to create machine code that can meet the
    request.
  }];
}

def AMDGPUMaxNumWorkGroupsDocs : Documentation {
  let Category = DocCatAMDGPUAttributes;
  let Content = [{
This attribute specifies the max number of work groups when the kernel
is dispatched.

Clang supports the
`__attribute__((amdgpu_max_num_work_groups(<x>, <y>, <z>)))` or
`[[clang::amdgpu_max_num_work_groups(<x>, <y>, <z>)]]` attribute for the
AMDGPU target. This attribute may be attached to HIP or OpenCL kernel function
definitions and is an optimization hint.

The `<x>` parameter specifies the maximum number of work groups in the x dimension.
Similarly `<y>` and `<z>` are for the y and z dimensions respectively.
Each of the three values must be greater than 0 when provided. The `<x>` parameter
is required, while `<y>` and `<z>` are optional with default value of 1.

If specified, the AMDGPU target backend might be able to produce better machine
code.

An error will be given if:
: - Specified values violate subtarget specifications;
  - Specified values are not compatible with values provided through other
    attributes.
  }];
}

def AMDGPUAvailableVisibleDocs : Documentation {
  let Category = DocCatAMDGPUAttributes;
  let Content = [{
This attribute controls availability and visibility as described in the [AMDGPU
Memory Model](https://llvm.org/docs/AMDGPUMemoryModel.html). When placed on
an atomic expression or fence, the resulting atomic or fence instruction carries
the corresponding *AV Metadata*.

The attribute takes a string literal as an argument, which currently has only
one supported value:

- `"none"`: Disable MakeAvailable and MakeVisible semantics on release and
  acquire operations respectively.

```c++
[[clang::amdgpu_av("none")]] __atomic_thread_fence(__ATOMIC_SEQ_CST);
[[clang::amdgpu_av("none")]] __atomic_fetch_add(ptr, 1, __ATOMIC_ACQ_REL);

// Also works with _Atomic type qualifier operations.
_Atomic int *p;
[[clang::amdgpu_av("none")]] *p += 1;
```
  }];
}

def DocCatCallingConvs : DocumentationCategory<"Calling Conventions"> {
  let Content = [{
Clang supports several different calling conventions, depending on the target
platform and architecture. The calling convention used for a function determines
how parameters are passed, how results are returned to the caller, and other
low-level details of calling a function.
  }];
}

def PcsDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On ARM targets, this attribute can be used to select calling conventions
similar to `stdcall` on x86. Valid parameter values are "aapcs" and
"aapcs-vfp".
  }];
}

def AArch64VectorPcsDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On AArch64 targets, this attribute changes the calling convention of a
function to preserve additional floating-point and Advanced SIMD registers
relative to the default calling convention used for AArch64.

This means it is more efficient to call such functions from code that performs
extensive floating-point and vector calculations, because fewer live SIMD and FP
registers need to be saved. This property makes it well-suited for e.g.
floating-point or vector math library functions, which are typically leaf
functions that require a small number of registers.

However, using this attribute also means that it is more expensive to call
a function that adheres to the default calling convention from within such
a function. Therefore, it is recommended that this attribute is only used
for leaf functions.

For more information, see the documentation for [aarch64_vector_pcs][aarch64_vector_pcs] on
the Arm Developer website.

[aarch64_vector_pcs]: https://developer.arm.com/products/software-development-tools/hpc/arm-compiler-for-hpc/vector-function-abi
  }];
}

def AArch64SVEPcsDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On AArch64 targets, this attribute changes the calling convention of a
function to preserve additional Scalable Vector registers and Scalable
Predicate registers relative to the default calling convention used for
AArch64.

This means it is more efficient to call such functions from code that performs
extensive scalable vector and scalable predicate calculations, because fewer
live SVE registers need to be saved. This property makes it well-suited for SVE
math library functions, which are typically leaf functions that require a small
number of registers.

However, using this attribute also means that it is more expensive to call
a function that adheres to the default calling convention from within such
a function. Therefore, it is recommended that this attribute is only used
for leaf functions.

For more information, see the documentation for `aarch64_sve_pcs` in the
ARM C Language Extension (ACLE) documentation.

[aarch64_sve_pcs]: https://github.com/ARM-software/acle/blob/main/main/acle.md#scalable-vector-extension-procedure-call-standard-attribute
  }];
}

def RegparmDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On 32-bit x86 targets, the regparm attribute causes the compiler to pass
the first three integer parameters in EAX, EDX, and ECX instead of on the
stack. This attribute has no effect on variadic functions, and all parameters
are passed via the stack as normal.
  }];
}

def SysVABIDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On Windows x86_64 targets, this attribute changes the calling convention of a
function to match the default convention used on Sys V targets such as Linux,
Mac, and BSD. This attribute has no effect on other targets.
  }];
}

def MSABIDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On non-Windows x86_64 and aarch64 targets, this attribute changes the calling convention of
a function to match the default convention used on Windows. This
attribute has no effect on Windows targets or non-x86_64, non-aarch64 targets.
  }];
}

def StdCallDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On 32-bit x86 targets, this attribute changes the calling convention of a
function to clear parameters off of the stack on return. This convention does
not support variadic calls or unprototyped functions in C, and has no effect on
x86_64 targets. This calling convention is used widely by the Windows API and
COM applications. See the documentation for [\_\_stdcall][__stdcall] on MSDN.

[__stdcall]: http://msdn.microsoft.com/en-us/library/zxk0tw93.aspx
  }];
}

def FastCallDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On 32-bit x86 targets, this attribute changes the calling convention of a
function to use ECX and EDX as register parameters and clear parameters off of
the stack on return. This convention does not support variadic calls or
unprototyped functions in C, and has no effect on x86_64 targets. This calling
convention is supported primarily for compatibility with existing code. Users
seeking register parameters should use the `regparm` attribute, which does
not require callee-cleanup. See the documentation for [\_\_fastcall][__fastcall] on MSDN.

[__fastcall]: http://msdn.microsoft.com/en-us/library/6xa169sk.aspx
  }];
}

def RegCallDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On x86 targets, this attribute changes the calling convention to
[\_\_regcall][__regcall] convention. This convention aims to pass as many arguments
as possible in registers. It also tries to utilize registers for the
return value whenever it is possible.

[__regcall]: https://www.intel.com/content/www/us/en/docs/dpcpp-cpp-compiler/developer-guide-reference/2023-2/c-c-sycl-calling-conventions.html
  }];
}

def ThisCallDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On 32-bit x86 targets, this attribute changes the calling convention of a
function to use ECX for the first parameter (typically the implicit `this`
parameter of C++ methods) and clear parameters off of the stack on return. This
convention does not support variadic calls or unprototyped functions in C, and
has no effect on x86_64 targets. See the documentation for [\_\_thiscall][__thiscall] on
MSDN.

[__thiscall]: http://msdn.microsoft.com/en-us/library/ek8tkfbw.aspx
  }];
}

def VectorCallDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On 32-bit x86 *and* x86_64 targets, this attribute changes the calling
convention of a function to pass vector parameters in SSE registers.

On 32-bit x86 targets, this calling convention is similar to `__fastcall`.
The first two integer parameters are passed in ECX and EDX. Subsequent integer
parameters are passed in memory, and callee clears the stack. On x86_64
targets, the callee does *not* clear the stack, and integer parameters are
passed in RCX, RDX, R8, and R9 as is done for the default Windows x64 calling
convention.

On both 32-bit x86 and x86_64 targets, vector and floating point arguments are
passed in XMM0-XMM5. Homogeneous vector aggregates of up to four elements are
passed in sequential SSE registers if enough are available. If AVX is enabled,
256 bit vectors are passed in YMM0-YMM5. Any vector or aggregate type that
cannot be passed in registers for any reason is passed by reference, which
allows the caller to align the parameter memory.

See the documentation for [\_\_vectorcall][__vectorcall] on MSDN for more details.

[__vectorcall]: http://msdn.microsoft.com/en-us/library/dn375768.aspx
  }];
}

def M68kRTDDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On M68k targets, this attribute changes the calling convention of a function
to clear parameters off the stack on return. In other words, callee is
responsible for cleaning out the stack space allocated for incoming paramters.
This convention does not support variadic calls or unprototyped functions in C.
When targeting M68010 or newer CPUs, this calling convention is implemented
using the `rtd` instruction.
  }];
}

def DocCatConsumed : DocumentationCategory<"Consumed Annotation Checking"> {
  let Content = [{
Clang supports additional attributes for checking basic resource management
properties, specifically for unique objects that have a single owning reference.
The following attributes are currently supported, although **the implementation
for these annotations is currently in development and are subject to change.**
  }];
}

def SetTypestateDocs : Documentation {
  let Category = DocCatConsumed;
  let Content = [{
Annotate methods that transition an object into a new state with
`__attribute__((set_typestate(new_state)))`. The new state must be
unconsumed, consumed, or unknown.
  }];
}

def CallableWhenDocs : Documentation {
  let Category = DocCatConsumed;
  let Content = [{
Use `__attribute__((callable_when(...)))` to indicate what states a method
may be called in. Valid states are unconsumed, consumed, or unknown. Each
argument to this attribute must be a quoted string. E.g.:

`__attribute__((callable_when("unconsumed", "unknown")))`
  }];
}

def TestTypestateDocs : Documentation {
  let Category = DocCatConsumed;
  let Content = [{
Use `__attribute__((test_typestate(tested_state)))` to indicate that a method
returns true if the object is in the specified state..
  }];
}

def ParamTypestateDocs : Documentation {
  let Category = DocCatConsumed;
  let Content = [{
This attribute specifies expectations about function parameters. Calls to an
function with annotated parameters will issue a warning if the corresponding
argument isn't in the expected state. The attribute is also used to set the
initial state of the parameter when analyzing the function's body.
  }];
}

def ReturnTypestateDocs : Documentation {
  let Category = DocCatConsumed;
  let Content = [{
The `return_typestate` attribute can be applied to functions or parameters.
When applied to a function the attribute specifies the state of the returned
value. The function's body is checked to ensure that it always returns a value
in the specified state. On the caller side, values returned by the annotated
function are initialized to the given state.

When applied to a function parameter it modifies the state of an argument after
a call to the function returns. The function's body is checked to ensure that
the parameter is in the expected state before returning.
  }];
}

def ConsumableDocs : Documentation {
  let Category = DocCatConsumed;
  let Content = [{
Each `class` that uses any of the typestate annotations must first be marked
using the `consumable` attribute. Failure to do so will result in a warning.

This attribute accepts a single parameter that must be one of the following:
`unknown`, `consumed`, or `unconsumed`.
  }];
}

def NoProfileInstrumentFunctionDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use the `no_profile_instrument_function` attribute on a function declaration
to denote that the compiler should not instrument the function with
profile-related instrumentation, such as via the
`-fprofile-generate` / `-fprofile-instr-generate` /
`-fcs-profile-generate` / `-fprofile-arcs` flags.
}];
}

def NoSanitizeDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use the `no_sanitize` attribute on a function or a global variable
declaration to specify that a particular instrumentation or set of
instrumentations should not be applied.

The attribute takes a list of string literals with the following accepted
values:

- all values accepted by `-fno-sanitize=`;
- `coverage`, to disable SanitizerCoverage instrumentation.

For example, `__attribute__((no_sanitize("address", "thread")))` specifies
that AddressSanitizer and ThreadSanitizer should not be applied to the function
or variable. Using `__attribute__((no_sanitize("coverage")))` specifies that
SanitizerCoverage should not be applied to the function.

See {ref}`Controlling Code Generation <controlling-code-generation>` for a
full list of supported sanitizer flags.
  }];
}

def DisableSanitizerInstrumentationDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use the `disable_sanitizer_instrumentation` attribute on a function,
Objective-C method, or global variable, to specify that no sanitizer
instrumentation should be applied.

This is not the same as `__attribute__((no_sanitize(...)))`, which depending
on the tool may still insert instrumentation to prevent false positive reports.
  }];
}

def NoSanitizeAddressDocs : Documentation {
  let Category = DocCatFunction;
  // This function has multiple distinct spellings, and so it requires a custom
  // heading to be specified. The most common spelling is sufficient.
  let Heading = "no_sanitize_address, no_address_safety_analysis";
  let Label = "langext-address_sanitizer";
  let Content = [{
Use `__attribute__((no_sanitize_address))` on a function or a global
variable declaration to specify that address safety instrumentation
(e.g. AddressSanitizer) should not be applied.
  }];
}

def NoSanitizeThreadDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "no_sanitize_thread";
  let Label = "langext-thread_sanitizer";
  let Content = [{
Use `__attribute__((no_sanitize_thread))` on a function declaration to
specify that checks for data races on plain (non-atomic) memory accesses should
not be inserted by ThreadSanitizer. The function is still instrumented by the
tool to avoid false positives and provide meaningful stack traces.
  }];
}

def NoSanitizeMemoryDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "no_sanitize_memory";
  let Label = "langext-memory_sanitizer";
  let Content = [{
Use `__attribute__((no_sanitize_memory))` on a function declaration to
specify that checks for uninitialized memory should not be inserted
(e.g. by MemorySanitizer). The function may still be instrumented by the tool
to avoid false positives in other places.
  }];
}

def CFICanonicalJumpTableDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "cfi_canonical_jump_table";
  let Label = "langext-cfi_canonical_jump_table";
  let Content = [{
Use `__attribute__((cfi_canonical_jump_table))` on a function declaration to
make the function's CFI jump table canonical. See {ref}`the CFI documentation
<cfi-canonical-jump-tables>` for more details.
  }];
}

def CFISaltDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "cfi_salt";
  let Label = "langext-cfi_salt";
  let Content = [{
The `cfi_salt` attribute specifies a string literal that is used as a salt
for Control-Flow Integrity (CFI) checks to distinguish between functions with
the same type signature. This attribute can be applied to function declarations,
function definitions, and function pointer typedefs.

The attribute prevents function pointers from being replaced with pointers to
functions that have a compatible type, which can be a CFI bypass vector.

**Syntax:**

- GNU-style: `__attribute__((cfi_salt("<salt_string>")))`
- C++11-style: `[[clang::cfi_salt("<salt_string>")]]`

**Usage:**

The attribute takes a single string literal argument that serves as the salt.
Functions or function types with different salt values will have different CFI
hashes, even if they have identical type signatures.

**Motivation:**

In large codebases like the Linux kernel, there are often hundreds of functions
with identical type signatures that are called indirectly:

```
1662 functions with void (*)(void)
1179 functions with int (*)(void)
 ...
```

By salting the CFI hashes, you can make CFI more robust by ensuring that
functions intended for different purposes have distinct CFI identities.

**Type Compatibility:**

- Functions with different salt values are considered to have incompatible types
- Function pointers with different salt values cannot be assigned to each other
- All declarations of the same function must use the same salt value

**Example:**

```c
// Header file - define convenience macros
#define __cfi_salt(s) __attribute__((cfi_salt(s)))

// Typedef for regular function pointers
typedef int (*fptr_t)(void);

// Typedef for salted function pointers
typedef int (*fptr_salted_t)(void) __cfi_salt("pepper");

struct widget_ops {
  fptr_t init;          // Regular CFI
  fptr_salted_t exec;   // Salted CFI
  fptr_t cleanup;       // Regular CFI
};

// Function implementations
static int widget_init(void) { return 0; }
static int widget_exec(void) __cfi_salt("pepper") { return 1; }
static int widget_cleanup(void) { return 0; }

static struct widget_ops ops = {
  .init = widget_init,      // OK - compatible types
  .exec = widget_exec,      // OK - both use "pepper" salt
  .cleanup = widget_cleanup // OK - compatible types
};

// Using C++11 attribute syntax
void secure_callback(void) [[clang::cfi_salt("secure")]];

// This would cause a compilation error:
// fptr_t bad_ptr = widget_exec;  // Error: incompatible types
```

**Notes:**

- The salt string can contain non-NULL ASCII characters, including spaces and
  quotes
- This attribute only applies to function types; using it on non-function
  types will generate a warning
- All declarations and definitions of the same function must use identical
  salt values
- The attribute affects type compatibility during compilation and CFI hash
  generation during code generation
  }];
}

def DocCatTypeSafety : DocumentationCategory<"Type Safety Checking"> {
  let Content = [{
Clang supports additional attributes to enable checking type safety properties
that can't be enforced by the C type system. To see warnings produced by these
checks, ensure that -Wtype-safety is enabled. Use cases include:

- MPI library implementations, where these attributes enable checking that
  the buffer type matches the passed `MPI_Datatype`;
- for HDF5 library there is a similar use case to MPI;
- checking types of variadic functions' arguments for functions like
  `fcntl()` and `ioctl()`.

You can detect support for these attributes with `__has_attribute()`. For
example:

```c++
#if defined(__has_attribute)
#  if __has_attribute(argument_with_type_tag) && \
      __has_attribute(pointer_with_type_tag) && \
      __has_attribute(type_tag_for_datatype)
#    define ATTR_MPI_PWT(buffer_idx, type_idx) __attribute__((pointer_with_type_tag(mpi,buffer_idx,type_idx)))
/* ... other macros ... */
#  endif
#endif

#if !defined(ATTR_MPI_PWT)
# define ATTR_MPI_PWT(buffer_idx, type_idx)
#endif

int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
    ATTR_MPI_PWT(1,3);
```
  }];
}

def ArgumentWithTypeTagDocs : Documentation {
  let Category = DocCatTypeSafety;
  let Heading = "argument_with_type_tag";
  let Content = [{
Use `__attribute__((argument_with_type_tag(arg_kind, arg_idx,
type_tag_idx)))` on a function declaration to specify that the function
accepts a type tag that determines the type of some other argument.

This attribute is primarily useful for checking arguments of variadic functions
(`pointer_with_type_tag` can be used in most non-variadic cases).

In the attribute prototype above:
: - `arg_kind` is an identifier that should be used when annotating all
    applicable type tags.
  - `arg_idx` provides the position of a function argument. The expected type of
    this function argument will be determined by the function argument specified
    by `type_tag_idx`. In the code example below, "3" means that the type of the
    function's third argument will be determined by `type_tag_idx`.
  - `type_tag_idx` provides the position of a function argument. This function
    argument will be a type tag. The type tag will determine the expected type of
    the argument specified by `arg_idx`. In the code example below, "2" means
    that the type tag associated with the function's second argument should agree
    with the type of the argument specified by `arg_idx`.

For example:

```c++
int fcntl(int fd, int cmd, ...)
    __attribute__(( argument_with_type_tag(fcntl,3,2) ));
// The function's second argument will be a type tag; this type tag will
// determine the expected type of the function's third argument.
```
  }];
}

def PointerWithTypeTagDocs : Documentation {
  let Category = DocCatTypeSafety;
  let Heading = "pointer_with_type_tag";
  let Content = [{
Use `__attribute__((pointer_with_type_tag(ptr_kind, ptr_idx, type_tag_idx)))`
on a function declaration to specify that the function accepts a type tag that
determines the pointee type of some other pointer argument.

In the attribute prototype above:
: - `ptr_kind` is an identifier that should be used when annotating all
    applicable type tags.
  - `ptr_idx` provides the position of a function argument; this function
    argument will have a pointer type. The expected pointee type of this pointer
    type will be determined by the function argument specified by
    `type_tag_idx`. In the code example below, "1" means that the pointee type
    of the function's first argument will be determined by `type_tag_idx`.
  - `type_tag_idx` provides the position of a function argument; this function
    argument will be a type tag. The type tag will determine the expected pointee
    type of the pointer argument specified by `ptr_idx`. In the code example
    below, "3" means that the type tag associated with the function's third
    argument should agree with the pointee type of the pointer argument specified
    by `ptr_idx`.

For example:

```c++
typedef int MPI_Datatype;
int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
    __attribute__(( pointer_with_type_tag(mpi,1,3) ));
// The function's 3rd argument will be a type tag; this type tag will
// determine the expected pointee type of the function's 1st argument.
```
  }];
}

def TypeTagForDatatypeDocs : Documentation {
  let Category = DocCatTypeSafety;
  let Content = [{
When declaring a variable, use
`__attribute__((type_tag_for_datatype(kind, type)))` to create a type tag that
is tied to the `type` argument given to the attribute.

In the attribute prototype above:
: - `kind` is an identifier that should be used when annotating all applicable
    type tags.
  - `type` indicates the name of the type.

Clang supports annotating type tags of two forms.

- **Type tag that is a reference to a declared identifier.**
  Use `__attribute__((type_tag_for_datatype(kind, type)))` when declaring that
  identifier:

  ```c++
  typedef int MPI_Datatype;
  extern struct mpi_datatype mpi_datatype_int
      __attribute__(( type_tag_for_datatype(mpi,int) ));
  #define MPI_INT ((MPI_Datatype) &mpi_datatype_int)
  // &mpi_datatype_int is a type tag. It is tied to type "int".
  ```

- **Type tag that is an integral literal.**
  Declare a `static const` variable with an initializer value and attach
  `__attribute__((type_tag_for_datatype(kind, type)))` on that declaration:

  ```c++
  typedef int MPI_Datatype;
  static const MPI_Datatype mpi_datatype_int
      __attribute__(( type_tag_for_datatype(mpi,int) )) = 42;
  #define MPI_INT ((MPI_Datatype) 42)
  // The number 42 is a type tag. It is tied to type "int".
  ```

The `type_tag_for_datatype` attribute also accepts an optional third argument
that determines how the type of the function argument specified by either
`arg_idx` or `ptr_idx` is compared against the type associated with the type
tag. (Recall that for the `argument_with_type_tag` attribute, the type of the
function argument specified by `arg_idx` is compared against the type
associated with the type tag. Also recall that for the `pointer_with_type_tag`
attribute, the pointee type of the function argument specified by `ptr_idx` is
compared against the type associated with the type tag.) There are two supported
values for this optional third argument:

- `layout_compatible` will cause types to be compared according to
  layout-compatibility rules (In C++11 [class.mem] p 17, 18, see the
  layout-compatibility rules for two standard-layout struct types and for two
  standard-layout union types). This is useful when creating a type tag
  associated with a struct or union type. For example:

  ```c++
  /* In mpi.h */
  typedef int MPI_Datatype;
  struct internal_mpi_double_int { double d; int i; };
  extern struct mpi_datatype mpi_datatype_double_int
      __attribute__(( type_tag_for_datatype(mpi,
                      struct internal_mpi_double_int, layout_compatible) ));

  #define MPI_DOUBLE_INT ((MPI_Datatype) &mpi_datatype_double_int)

  int MPI_Send(void *buf, int count, MPI_Datatype datatype, ...)
      __attribute__(( pointer_with_type_tag(mpi,1,3) ));

  /* In user code */
  struct my_pair { double a; int b; };
  struct my_pair *buffer;
  MPI_Send(buffer, 1, MPI_DOUBLE_INT /*, ... */); // no warning because the
                                                   // layout of my_pair is
                                                   // compatible with that of
                                                   // internal_mpi_double_int

  struct my_int_pair { int a; int b; }
  struct my_int_pair *buffer2;
  MPI_Send(buffer2, 1, MPI_DOUBLE_INT /*, ... */); // warning because the
                                                    // layout of my_int_pair
                                                    // does not match that of
                                                    // internal_mpi_double_int
  ```

- `must_be_null` specifies that the function argument specified by either
  `arg_idx` (for the `argument_with_type_tag` attribute) or `ptr_idx` (for
  the `pointer_with_type_tag` attribute) should be a null pointer constant.
  The second argument to the `type_tag_for_datatype` attribute is ignored. For
  example:

  ```c++
  /* In mpi.h */
  typedef int MPI_Datatype;
  extern struct mpi_datatype mpi_datatype_null
      __attribute__(( type_tag_for_datatype(mpi, void, must_be_null) ));

  #define MPI_DATATYPE_NULL ((MPI_Datatype) &mpi_datatype_null)
  int MPI_Send(void *buf, int count, MPI_Datatype datatype, ...)
      __attribute__(( pointer_with_type_tag(mpi,1,3) ));

  /* In user code */
  struct my_pair { double a; int b; };
  struct my_pair *buffer;
  MPI_Send(buffer, 1, MPI_DATATYPE_NULL /*, ... */); // warning: MPI_DATATYPE_NULL
                                                      // was specified but buffer
                                                      // is not a null pointer
  ```
  }];
}

def FlattenDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `flatten` attribute causes calls within the attributed function to
be inlined unless it is impossible to do so, for example if the body of the
callee is unavailable or if the callee has the `noinline` attribute.
  }];
}

def FormatDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `format` attribute, which indicates that the function
accepts (among other possibilities) a `printf` or `scanf`-like format string
and corresponding arguments or a `va_list` that contains these arguments.

Please see [GCC documentation about format attribute](http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html) to find details
about attribute syntax.

Clang implements two kinds of checks with this attribute.

1. Clang checks that the function with the `format` attribute is called with
   a format string that uses format specifiers that are allowed, and that
   arguments match the format string. This is the `-Wformat` warning, it is
   on by default.

2. Clang checks that the format string argument is a literal string. This is
   the `-Wformat-nonliteral` warning, it is off by default.

   Clang implements this mostly the same way as GCC, but there is a difference
   for functions that accept a `va_list` argument (for example, `vprintf`).
   GCC does not emit `-Wformat-nonliteral` warning for calls to such
   functions. Clang does not warn if the format string comes from a function
   parameter, where the function is annotated with a compatible attribute,
   otherwise it warns. For example:

   ```c
   __attribute__((__format__ (__scanf__, 1, 3)))
   void foo(const char* s, char *buf, ...) {
     va_list ap;
     va_start(ap, buf);

     vprintf(s, ap); // warning: format string is not a string literal
   }
   ```

   In this case we warn because `s` contains a format string for a
   `scanf`-like function, but it is passed to a `printf`-like function.

   If the attribute is removed, clang still warns, because the format string is
   not a string literal.

   Another example:

   ```c
   __attribute__((__format__ (__printf__, 1, 3)))
   void foo(const char* s, char *buf, ...) {
     va_list ap;
     va_start(ap, buf);

     vprintf(s, ap); // warning
   }
   ```

   In this case Clang does not warn because the format string `s` and
   the corresponding arguments are annotated. If the arguments are
   incorrect, the caller of `foo` will receive a warning.

As an extension to GCC's behavior, Clang accepts the `format` attribute on
non-variadic functions. Clang checks non-variadic format functions for the same
classes of issues that can be found on variadic functions, as controlled by the
same warning flags, except that the types of formatted arguments is forced by
the function signature. For example:

```c
__attribute__((__format__(__printf__, 1, 2)))
void fmt(const char *s, const char *a, int b);

void bar(void) {
  fmt("%s %i", "hello", 123); // OK
  fmt("%i %g", "hello", 123); // warning: arguments don't match format
  extern const char *fmt;
  fmt(fmt, "hello", 123); // warning: format string is not a string literal
}
```

When using the format attribute on a variadic function, the first data parameter
\_must\_ be the index of the ellipsis in the parameter list. Clang will generate
a diagnostic otherwise, as it wouldn't be possible to forward that argument list
to `printf`-family functions. For instance, this is an error:

```c
__attribute__((__format__(__printf__, 1, 2)))
void fmt(const char *s, int b, ...);
// ^ error: format attribute parameter 3 is out of bounds
// (must be __printf__, 1, 3)
```

Using the `format` attribute on a non-variadic function emits a GCC
compatibility diagnostic.
  }];
}

def AlignValueDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The align_value attribute can be added to the typedef of a pointer type or the
declaration of a variable of pointer or reference type. It specifies that the
pointer will point to, or the reference will bind to, only objects with at
least the provided alignment. This alignment value must be some positive power
of 2.

```c
typedef double * aligned_double_ptr __attribute__((align_value(64)));
void foo(double & x  __attribute__((align_value(128))),
         aligned_double_ptr y) { ... }
```

If the pointer value does not have the specified alignment at runtime, the
behavior of the program is undefined.
  }];
}

def FormatMatchesDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `format` attribute is the basis for the enforcement of diagnostics in the
`-Wformat` family, but it only handles the case where the format string is
passed along with the arguments it is going to format. It cannot handle the case
where the format string and the format arguments are passed separately from each
other. For instance:

```c
static const char *first_name;
static double todays_temperature;
static int wind_speed;

void say_hi(const char *fmt) {
  printf(fmt, first_name, todays_temperature);
      // ^ warning: format string is not a string literal
  printf(fmt, first_name, wind_speed);
      // ^ warning: format string is not a string literal
}

int main() {
  say_hi("hello %s, it is %g degrees outside");
  say_hi("hello %s, it is %d degrees outside!");
                        // ^ no diagnostic, but %d cannot format doubles
}
```

In this example, `fmt` is expected to format a `const char *` and a
`double`, but these values are not passed to `say_hi`. Without the
`format` attribute (which cannot apply in this case), the -Wformat-nonliteral
diagnostic unnecessarily triggers in the body of `say_hi`, and incorrect
`say_hi` call sites do not trigger a diagnostic.

To complement the `format` attribute, Clang also defines the
`format_matches` attribute. Its syntax is similar to the `format`
attribute's, but instead of taking the index of the first formatted value
argument, it takes a C string literal with the expected specifiers:

```c
static const char *first_name;
static double todays_temperature;
static int wind_speed;

__attribute__((__format_matches__(printf, 1, "%s %g")))
void say_hi(const char *fmt) {
  printf(fmt, first_name, todays_temperature); // no dignostic
  printf(fmt, first_name, wind_speed); // warning: format specifies type 'int' but the argument has type 'double'
}

int main() {
  say_hi("hello %s, it is %g degrees outside");
  say_hi("it is %g degrees outside, have a good day %s!");
  // warning: format specifies 'double' where 'const char *' is required
  // warning: format specifies 'const char *' where 'double' is required
}
```

The third argument to `format_matches` is expected to evaluate to a **C string
literal** even when the format string would normally be a different type for the
given flavor, like a `CFStringRef` or a `NSString *`.

The only requirement on the format string literal is that it has specifiers
that are compatible with the arguments that will be used. It can contain
arbitrary non-format characters. For instance, for the purposes of compile-time
validation, `"%s scored %g%% on her test"` and `"%s%g"` are interchangeable
as the format string argument. As a means of self-documentation, users may
prefer the former when it provides a useful example of an expected format
string.

In the implementation of a function with the `format_matches` attribute,
format verification works as if the format string was identical to the one
specified in the attribute.

```c
__attribute__((__format_matches__(printf, 1, "%s %g")))
void say_hi(const char *fmt) {
  printf(fmt, "person", 546);
                     // ^ warning: format specifies type 'double' but the
                     //   argument has type 'int'
  // note: format string is defined here:
  // __attribute__((__format_matches__(printf, 1, "%s %g")))
  //                                                  ^~
}
```

At the call sites of functions with the `format_matches` attribute, format
verification instead compares the two format strings to evaluate their
equivalence. Each format flavor defines equivalence between format specifiers.
Generally speaking, two specifiers are equivalent if they format the same type.
For instance, in the `printf` flavor, `%2i` and `%-0.5d` are compatible.
When `-Wformat-signedness` is disabled, `%d` and `%u` are compatible. For
a negative example, `%ld` is incompatible with `%d`.

Do note the following un-obvious cases:

- Passing `NULL` as the format string does not trigger format diagnostics.
- When the format string is not NULL, it cannot \_miss\_ specifiers, even in
  trailing positions. For instance, `%d` is not accepted when the required
  format is `%d %d %d`.
- While checks for the `format` attribute tolerate sone size mismatches
  that standard argument promotion renders immaterial (such as formatting an
  `int` with `%hhd`, which specifies a `char`-sized integer), checks for
  `format_matches` require specified argument sizes to match exactly.
- Format strings expecting a variable modifier (such as `%*s`) are
  incompatible with format strings that would itemize the variable modifiers
  (such as `%i %s`), even if the two specify ABI-compatible argument lists.
- All pointer specifiers, modifiers aside, are mutually incompatible. For
  instance, `%s` is not compatible with `%p`, and `%p` is not compatible
  with `%n`, and `%hhn` is incompatible with `%s`, even if the pointers
  are ABI-compatible or identical on the selected platform. However, `%0.5s`
  is compatible with `%s`, since the difference only exists in modifier flags.
  This is not overridable with `-Wformat-pedantic` or its inverse, which
  control similar behavior in `-Wformat`.

At this time, clang implements `format_matches` only for format types in the
`printf` family. This includes variants such as Apple's NSString format and
the FreeBSD `kprintf`, but excludes `scanf`. Using a known but unsupported
format silently fails in order to be compatible with other implementations that
would support these formats.

  }];
}

def FlagEnumDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute can be added to an enumerator to signal to the compiler that it
is intended to be used as a flag type. This will cause the compiler to assume
that the range of the type includes all of the values that you can get by
manipulating bits of the enumerator when issuing warnings.
  }];
}

def AsmLabelDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute can be used on a function or variable to specify its symbol name.

On some targets, all C symbols are prefixed by default with a single character,
typically `_`. This was done historically to distinguish them from symbols
used by other languages. (This prefix is also added to the standard Itanium
C++ ABI prefix on "mangled" symbol names, so that e.g. on such targets the true
symbol name for a C++ variable declared as `int cppvar;` would be
`__Z6cppvar`; note the two underscores.) This prefix is *not* added to the
symbol names specified by the `__asm` attribute; programmers wishing to match
a C symbol name must compensate for this.

For example, consider the following C code:

```c
int var1 __asm("altvar") = 1;  // "altvar" in symbol table.
int var2 = 1; // "_var2" in symbol table.

void func1(void) __asm("altfunc");
void func1(void) {} // "altfunc" in symbol table.
void func2(void) {} // "_func2" in symbol table.
```

Clang's implementation of this attribute is compatible with GCC's, [documented here](https://gcc.gnu.org/onlinedocs/gcc/Asm-Labels.html).

While it is possible to use this attribute to name a special symbol used
internally by the compiler, such as an LLVM intrinsic, this is neither
recommended nor supported and may cause the compiler to crash or miscompile.
Users who wish to gain access to intrinsic behavior are strongly encouraged to
request new builtin functions.
  }];
}

def EnumExtensibilityDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
Attribute `enum_extensibility` is used to distinguish between enum definitions
that are extensible and those that are not. The attribute can take either
`closed` or `open` as an argument. `closed` indicates a variable of the
enum type takes a value that corresponds to one of the enumerators listed in the
enum definition or, when the enum is annotated with `flag_enum`, a value that
can be constructed using values corresponding to the enumerators. `open`
indicates a variable of the enum type can take any values allowed by the
standard and instructs clang to be more lenient when issuing warnings.

```c
enum __attribute__((enum_extensibility(closed))) ClosedEnum {
  A0, A1
};

enum __attribute__((enum_extensibility(open))) OpenEnum {
  B0, B1
};

enum __attribute__((enum_extensibility(closed),flag_enum)) ClosedFlagEnum {
  C0 = 1 << 0, C1 = 1 << 1
};

enum __attribute__((enum_extensibility(open),flag_enum)) OpenFlagEnum {
  D0 = 1 << 0, D1 = 1 << 1
};

void foo1() {
  enum ClosedEnum ce;
  enum OpenEnum oe;
  enum ClosedFlagEnum cfe;
  enum OpenFlagEnum ofe;

  ce = A1;           // no warnings
  ce = 100;          // warning issued
  oe = B1;           // no warnings
  oe = 100;          // no warnings
  cfe = C0 | C1;     // no warnings
  cfe = C0 | C1 | 4; // warning issued
  ofe = D0 | D1;     // no warnings
  ofe = D0 | D1 | 4; // no warnings
}
```

  }];
}

def EmptyBasesDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The empty_bases attribute permits the compiler to utilize the
empty-base-optimization more frequently.
This attribute only applies to struct, class, and union types.
It is only supported when using the Microsoft C++ ABI.
  }];
}

def LayoutVersionDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The layout_version attribute requests that the compiler utilize the class
layout rules of a particular compiler version.
This attribute only applies to struct, class, and union types.
It is only supported when using the Microsoft C++ ABI.
  }];
}

def LifetimeBoundDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `lifetimebound` attribute on a function parameter or implicit object
parameter indicates that objects that are referred to by that parameter may
also be referred to by the return value of the annotated function (or, for a
parameter of a constructor, by the value of the constructed object).

By default, a reference is considered to refer to its referenced object, a
pointer is considered to refer to its pointee, a `std::initializer_list<T>`
is considered to refer to its underlying array, and aggregates (arrays and
simple `struct`s) are considered to refer to all objects that their
transitive subobjects refer to.

Clang warns if it is able to detect that an object or reference refers to
another object with a shorter lifetime. For example, Clang will warn if a
function returns a reference to a local variable, or if a reference is bound to
a temporary object whose lifetime is not extended. By using the
`lifetimebound` attribute, this determination can be extended to look through
user-declared functions. For example:

```c++
#include <map>
#include <string>

using namespace std::literals;

// Returns m[key] if key is present, or default_value if not.
template<typename T, typename U>
const U &get_or_default(const std::map<T, U> &m [[clang::lifetimebound]],
                        const T &key, /* note, not lifetimebound */
                        const U &default_value [[clang::lifetimebound]]) {
  if (auto iter = m.find(key); iter != m.end()) return iter->second;
  else return default_value;
}

int main() {
  std::map<std::string, std::string> m;
  // warning: temporary bound to local reference 'val1' will be destroyed
  // at the end of the full-expression
  const std::string &val1 = get_or_default(m, "foo"s, "bar"s);

  // No warning in this case.
  std::string def_val = "bar"s;
  const std::string &val2 = get_or_default(m, "foo"s, def_val);

  return 0;
}
```

The attribute can be applied to the implicit `this` parameter of a member
function by writing the attribute after the function type:

```c++
struct string {
  // The returned pointer should not outlive ``*this``.
  const char *data() const [[clang::lifetimebound]];
};
```

This attribute is inspired by the C++ committee paper [P0936R0](http://wg21.link/p0936r0), but does not affect whether temporary objects
have their lifetimes extended.
  }];
}

def LifetimeCaptureByDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "lifetime_capture_by, lifetime_capture_by_this, lifetime_capture_by_global, lifetime_capture_by_unknown";
  let Content = [{
Similar to [lifetimebound], the `lifetime_capture_by` attribute family on a
function parameter or implicit object parameter indicates that a capturing
entity may refer to the object referred to by that parameter. The capturing
entity can be named in `lifetime_capture_by(X)` or selected by one of the
standalone special forms listed below.

Below is a list of types of the parameters and what they're considered to refer to:

- A reference param (of non-view type) is considered to refer to its referenced object.
- A pointer param (of non-view type) is considered to refer to its pointee.
- View type param (type annotated with `[[gsl::Pointer()]]`) is considered to refer
  to its pointee (gsl owner). This holds true even if the view type appears as a reference
  in the parameter. For example, both `std::string_view` and
  `const std::string_view &` are considered to refer to a `std::string`.
- A `std::initializer_list<T>` is considered to refer to its underlying array.
- Aggregates (arrays and simple `struct`s) are considered to refer to all
  objects that their transitive subobjects refer to.

Clang would diagnose when a temporary object is used as an argument to such an
annotated parameter.
In this case, the capturing entity `X` could capture a dangling reference to this
temporary object.

```c++
void addToSet(std::string_view a [[clang::lifetime_capture_by(s)]], std::set<std::string_view>& s) {
  s.insert(a);
}
void use() {
  std::set<std::string_view> s;
  addToSet(std::string(), s); // Warning: object whose reference is captured by 's' will be destroyed at the end of the full-expression.
  //       ^^^^^^^^^^^^^
  std::string local;
  addToSet(local, s); // Ok.
}
```

The capturing entity can be one of the following:

- Another (named) function parameter.

  ```c++
  void addToSet(std::string_view a [[clang::lifetime_capture_by(s)]], std::set<std::string_view>& s) {
    s.insert(a);
  }
  ```

- `this` (in case of member functions), written as
  `lifetime_capture_by_this`.

  ```c++
  class S {
    void addToSet(std::string_view a [[clang::lifetime_capture_by_this]]) {
      s.insert(a);
    }
    std::set<std::string_view> s;
  };
  ```

  Note: When applied to a constructor parameter, `[[clang::lifetime_capture_by_this]]` is just an alias of `[[clang::lifetimebound]]`.

- `global` and `unknown`, written as `lifetime_capture_by_global` and
  `lifetime_capture_by_unknown` respectively.

  ```c++
  std::set<std::string_view> s;
  void addToSet(std::string_view a [[clang::lifetime_capture_by_global]]) {
    s.insert(a);
  }
  void addSomewhere(std::string_view a [[clang::lifetime_capture_by_unknown]]);
  ```

The attribute can be applied to the implicit `this` parameter of a member
function by writing the attribute after the function type:

```c++
struct S {
  const char *data(std::set<S*>& s) [[clang::lifetime_capture_by(s)]] {
    s.insert(this);
  }
};
```

The parameter-list form supports specifying more than one capturing entity:

```c++
void addToSets(std::string_view a [[clang::lifetime_capture_by(s1, s2)]],
               std::set<std::string_view>& s1,
               std::set<std::string_view>& s2) {
  s1.insert(a);
  s2.insert(a);
}
```

Distinct `lifetime_capture_by` forms can also be combined on the same
declaration, but each form can appear at most once. For example,
`[[clang::lifetime_capture_by(s), clang::lifetime_capture_by_this]]` is
allowed, but two `[[clang::lifetime_capture_by(...)]]` attributes or two
`[[clang::lifetime_capture_by_this]]` attributes on the same declaration are
rejected.

Limitation: The capturing entity `X` is not used by the analysis and is
used for documentation purposes only. This is because the analysis is
statement-local and only detects use of a temporary as an argument to the
annotated parameter.

```c++
void addToSet(std::string_view a [[clang::lifetime_capture_by(s)]], std::set<std::string_view>& s);
void use() {
  std::set<std::string_view> s;
  if (foo()) {
    std::string str;
    addToSet(str, s); // Not detected.
  }
}
```
  }];
}

def TrivialABIDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `trivial_abi` attribute can be applied to a C++ class, struct, or union.
It instructs the compiler to pass and return the type using the C ABI for the
underlying type when the type would otherwise be considered non-trivial for the
purpose of calls.
A class annotated with `trivial_abi` can have non-trivial destructors or
copy/move constructors without automatically becoming non-trivial for the
purposes of calls. For example:

```c++
// A is trivial for the purposes of calls because `trivial_abi` makes the
// user-provided special functions trivial.
struct __attribute__((trivial_abi)) A {
  ~A();
  A(const A &);
  A(A &&);
  int x;
};

// B's destructor and copy/move constructor are considered trivial for the
// purpose of calls because A is trivial.
struct B {
  A a;
};
```

If a type is trivial for the purposes of calls, has a non-trivial destructor,
and is passed as an argument by value, the convention is that the callee will
destroy the object before returning. The lifetime of the copy of the parameter
in the caller ends without a destructor call when the call begins.

If a type is trivial for the purpose of calls, it is assumed to be trivially
relocatable for the purpose of `__is_trivially_relocatable` and
`__builtin_is_cpp_trivially_relocatable`.
When a type marked with `[[trivial_abi]]` is used as a function argument,
the compiler may omit the call to the copy constructor.
Thus, side effects of the copy constructor are potentially not performed.
For example, objects that contain pointers to themselves or otherwise depend
on their address (or the address or their subobjects) should not be declared
`[[trivial_abi]]`.

Attribute `trivial_abi` has no effect in the following cases:

- The class directly declares a virtual base or virtual methods.

- Copy constructors and move constructors of the class are all deleted.

- The class has a base class that is non-trivial for the purposes of calls.

- The class has a non-static data member whose type is non-trivial for the
  purposes of calls, which includes:

  - classes that are non-trivial for the purposes of calls
  - \_\_weak-qualified types in Objective-C++
  - arrays of any of the above
  }];
}

def MSInheritanceDocs : Documentation {
  let Category = DocCatDecl;
  let Heading = "__single_inheritance, __multiple_inheritance, __virtual_inheritance";
  let Content = [{
This collection of keywords is enabled under `-fms-extensions` and controls
the pointer-to-member representation used on `*-*-win32` targets.

The `*-*-win32` targets utilize a pointer-to-member representation which
varies in size and alignment depending on the definition of the underlying
class.

However, this is problematic when a forward declaration is only available and
no definition has been made yet. In such cases, Clang is forced to utilize the
most general representation that is available to it.

These keywords make it possible to use a pointer-to-member representation other
than the most general one regardless of whether or not the definition will ever
be present in the current translation unit.

This family of keywords belong between the `class-key` and `class-name`:

```c++
struct __single_inheritance S;
int S::*i;
struct S {};
```

This keyword can be applied to class templates but only has an effect when used
on full specializations:

```c++
template <typename T, typename U> struct __single_inheritance A; // warning: inheritance model ignored on primary template
template <typename T> struct __multiple_inheritance A<T, T>; // warning: inheritance model ignored on partial specialization
template <> struct __single_inheritance A<int, float>;
```

Note that choosing an inheritance model less general than strictly necessary is
an error:

```c++
struct __multiple_inheritance S; // error: inheritance model does not match definition
int S::*i;
struct S {};
```
}];
}

def MSConstexprDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
The `[[msvc::constexpr]]` attribute can be applied only to a function
definition or a `return` statement. It does not impact function declarations.
A `[[msvc::constexpr]]` function cannot be `constexpr` or `consteval`.
A `[[msvc::constexpr]]` function is treated as if it were a `constexpr` function
when it is evaluated in a constant context of `[[msvc::constexpr]] return` statement.
Otherwise, it is treated as a regular function.

Semantics of this attribute are enabled only under MSVC compatibility
(`-fms-compatibility-version`) 19.33 and later.
  }];
}

def MSNoVTableDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute can be added to a class declaration or definition to signal to
the compiler that constructors and destructors will not reference the virtual
function table. It is only supported when using the Microsoft C++ ABI.
  }];
}

def OptnoneDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `optnone` attribute suppresses essentially all optimizations
on a function or method, regardless of the optimization level applied to
the compilation unit as a whole. This is particularly useful when you
need to debug a particular function, but it is infeasible to build the
entire application without optimization. Avoiding optimization on the
specified function can improve the quality of the debugging information
for that function.

This attribute is incompatible with the `always_inline` and `minsize`
attributes.

Note that this attribute does not apply recursively to nested functions such as
lambdas or blocks when using declaration-specific attribute syntaxes such as double
square brackets (`[[]]`) or `__attribute__`. The `#pragma` syntax can be
used to apply the attribute to all functions, including nested functions, in a
range of source code.
  }];
}

def LoopHintDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "#pragma clang loop";
  let Content = [{
The `#pragma clang loop` directive allows loop optimization hints to be
specified for the subsequent loop. The directive allows pipelining to be
disabled, or vectorization, vector predication, interleaving, and unrolling to
be enabled or disabled. Vector width, vector predication, interleave count,
unrolling count, and the initiation interval for pipelining can be explicitly
specified. See
{ref}`loop hint optimizations <langext-loop-hint-optimizations>` for details.
  }];
}

def UnrollHintDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "#pragma unroll, #pragma nounroll";
  let Content = [{
Loop unrolling optimization hints can be specified with `#pragma unroll` and
`#pragma nounroll`. The pragma is placed immediately before a for, while,
do-while, or c++11 range-based for loop. GCC's loop unrolling hints
`#pragma GCC unroll` and `#pragma GCC nounroll` are also supported and have
identical semantics to `#pragma unroll` and `#pragma nounroll`.

Specifying `#pragma unroll` without a parameter directs the loop unroller to
attempt to fully unroll the loop if the trip count is known at compile time and
attempt to partially unroll the loop if the trip count is not known at compile
time:

```c++
#pragma unroll
for (...) {
  ...
}
```

Specifying the optional parameter, `#pragma unroll _value_`, directs the
unroller to unroll the loop `_value_` times. The parameter may optionally be
enclosed in parentheses:

```c++
#pragma unroll 16
for (...) {
  ...
}

#pragma unroll(16)
for (...) {
  ...
}
```

Specifying `#pragma nounroll` indicates that the loop should not be unrolled:

```c++
#pragma nounroll
for (...) {
  ...
}
```

`#pragma unroll` and `#pragma unroll _value_` have identical semantics to
`#pragma clang loop unroll(enable)` and
`#pragma clang loop unroll_count(_value_)` respectively. `#pragma nounroll`
is equivalent to `#pragma clang loop unroll(disable)`. See
{ref}`loop hint optimizations <langext-loop-hint-optimizations>` for further
details including limitations of the unroll hints.
  }];
}

def PipelineHintDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "#pragma clang loop pipeline, #pragma clang loop pipeline_initiation_interval";
  let Content = [{
Software Pipelining optimization is a technique used to optimize loops by
utilizing instruction-level parallelism. It reorders loop instructions to
overlap iterations. As a result, the next iteration starts before the previous
iteration has finished. The module scheduling technique creates a schedule for
one iteration such that when repeating at regular intervals, no inter-iteration
dependencies are violated. This constant interval(in cycles) between the start
of iterations is called the initiation interval. i.e. The initiation interval
is the number of cycles between two iterations of an unoptimized loop in the
newly created schedule. A new, optimized loop is created such that a single iteration
of the loop executes in the same number of cycles as the initiation interval.

For further details see \<<https://llvm.org/pubs/2005-06-17-LattnerMSThesis-book.pdf>>.

`#pragma clang loop pipeline and #pragma loop pipeline_initiation_interval`
could be used as hints for the software pipelining optimization. The pragma is
placed immediately before a for, while, do-while, or a C++11 range-based for
loop.

Using `#pragma clang loop pipeline(disable)` avoids the software pipelining
optimization. The disable state can only be specified:

```c++
#pragma clang loop pipeline(disable)
for (...) {
  ...
}
```

Using `#pragma loop pipeline_initiation_interval` instructs
the software pipeliner to try the specified initiation interval.
If a schedule was found then the resulting loop iteration would have
the specified cycle count. If a schedule was not found then loop
remains unchanged. The initiation interval must be a positive number
greater than zero:

```c++
#pragma loop pipeline_initiation_interval(10)
for (...) {
  ...
}
```

  }];
}

def OpenCLUnrollHintDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
The opencl_unroll_hint attribute qualifier can be used to specify that a loop
(for, while and do loops) can be unrolled. This attribute qualifier can be
used to specify full unrolling or partial unrolling by a specified amount.
This is a compiler hint and the compiler may ignore this directive. See
[OpenCL v2.0](https://www.khronos.org/registry/cl/specs/opencl-2.0.pdf)
s6.11.5 for details.
  }];
}

def OpenCLIntelReqdSubGroupSizeDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
The optional attribute intel_reqd_sub_group_size can be used to indicate that
the kernel must be compiled and executed with the specified subgroup size. When
this attribute is present, get_max_sub_group_size() is guaranteed to return the
specified integer value. This is important for the correctness of many subgroup
algorithms, and in some cases may be used by the compiler to generate more optimal
code. See `cl_intel_required_subgroup_size
<https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_required_subgroup_size.txt>`
for details.
  }];
}

def OpenCLAccessDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "__read_only, __write_only, __read_write (read_only, write_only, read_write)";
  let Content = [{
The access qualifiers must be used with image object arguments or pipe arguments
to declare if they are being read or written by a kernel or function.

The read_only/\_\_read_only, write_only/\_\_write_only and read_write/\_\_read_write
names are reserved for use as access qualifiers and shall not be used otherwise.

```c
kernel void
foo (read_only image2d_t imageA,
     write_only image2d_t imageB) {
  ...
}
```

In the above example imageA is a read-only 2D image object, and imageB is a
write-only 2D image object.

The read_write (or \_\_read_write) qualifier can not be used with pipe.

More details can be found in the OpenCL C language Spec v2.0, Section 6.6.
    }];
}

def DocOpenCLAddressSpaces : DocumentationCategory<"OpenCL Address Spaces"> {
  let Content = [{
The address space qualifier may be used to specify the region of memory that is
used to allocate the object. OpenCL supports the following address spaces:
\_\_generic(generic), \_\_global(global), \_\_local(local), \_\_private(private),
\_\_constant(constant).

```c
__constant int c = ...;

__generic int* foo(global int* g) {
  __local int* l;
  private int p;
  ...
  return l;
}
```

More details can be found in the OpenCL C language Spec v2.0, Section 6.5.
  }];
}

def OpenCLAddressSpaceGenericDocs : Documentation {
  let Category = DocOpenCLAddressSpaces;
  let Heading = "__generic, generic, [[clang::opencl_generic]]";
  let Content = [{
The generic address space attribute is only available with OpenCL v2.0 and later.
It can be used with pointer types. Variables in global and local scope and
function parameters in non-kernel functions can have the generic address space
type attribute. It is intended to be a placeholder for any other address space
except for '\_\_constant' in OpenCL code which can be used with multiple address
spaces.
  }];
}

def OpenCLAddressSpaceConstantDocs : Documentation {
  let Category = DocOpenCLAddressSpaces;
  let Heading = "__constant, constant, [[clang::opencl_constant]]";
  let Content = [{
The constant address space attribute signals that an object is located in
a constant (non-modifiable) memory region. It is available to all work items.
Any type can be annotated with the constant address space attribute. Objects
with the constant address space qualifier can be declared in any scope and must
have an initializer.
  }];
}

def OpenCLAddressSpaceGlobalDocs : Documentation {
  let Category = DocOpenCLAddressSpaces;
  let Heading = "__global, global, [[clang::opencl_global]]";
  let Content = [{
The global address space attribute specifies that an object is allocated in
global memory, which is accessible by all work items. The content stored in this
memory area persists between kernel executions. Pointer types to the global
address space are allowed as function parameters or local variables. Starting
with OpenCL v2.0, the global address space can be used with global (program
scope) variables and static local variable as well.
  }];
}

def OpenCLAddressSpaceGlobalExtDocs : Documentation {
  let Category = DocOpenCLAddressSpaces;
  let Heading = "[[clang::opencl_global_device]], [[clang::opencl_global_host]]";
  let Content = [{
The `global_device` and `global_host` address space attributes specify that
an object is allocated in global memory on the device/host. It helps to
distinguish USM (Unified Shared Memory) pointers that access global device
memory from those that access global host memory. These new address spaces are
a subset of the `__global/opencl_global` address space, the full address space
set model for OpenCL 2.0 with the extension looks as follows:

```text
generic->global->host
               ->device
       ->private
       ->local
constant
```

As `global_device` and `global_host` are a subset of
`__global/opencl_global` address spaces it is allowed to convert
`global_device` and `global_host` address spaces to
`__global/opencl_global` address spaces (following ISO/IEC TR 18037 5.1.3
"Address space nesting and rules for pointers").

These attributes are deprecated and may be removed in a future version of Clang.
  }];
}

def OpenCLAddressSpaceLocalDocs : Documentation {
  let Category = DocOpenCLAddressSpaces;
  let Heading = "__local, local, [[clang::opencl_local]]";
  let Content = [{
The local address space specifies that an object is allocated in the local (work
group) memory area, which is accessible to all work items in the same work
group. The content stored in this memory region is not accessible after
the kernel execution ends. In a kernel function scope, any variable can be in
the local address space. In other scopes, only pointer types to the local address
space are allowed. Local address space variables cannot have an initializer.
  }];
}

def OpenCLAddressSpacePrivateDocs : Documentation {
  let Category = DocOpenCLAddressSpaces;
  let Heading = "__private, private, [[clang::opencl_private]]";
  let Content = [{
The private address space specifies that an object is allocated in the private
(work item) memory. Other work items cannot access the same memory area and its
content is destroyed after work item execution ends. Local variables can be
declared in the private address space. Function arguments are always in the
private address space. Kernel function arguments of a pointer or an array type
cannot point to the private address space.
  }];
}

def OpenCLNoSVMDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
OpenCL 2.0 supports the optional `__attribute__((nosvm))` qualifier for
pointer variable. It informs the compiler that the pointer does not refer
to a shared virtual memory region. See OpenCL v2.0 s6.7.2 for details.

Since it is not widely used and has been removed from OpenCL 2.1, it is ignored
by Clang.
  }];
}

def Ptr32Docs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `__ptr32` qualifier represents a native pointer on a 32-bit system. On a
64-bit system, a pointer with `__ptr32` is extended to a 64-bit pointer. The
`__sptr` and `__uptr` qualifiers can be used to specify whether the pointer
is sign extended or zero extended. This qualifier is enabled under
`-fms-extensions`.
  }];
}

def Ptr64Docs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `__ptr64` qualifier represents a native pointer on a 64-bit system. On a
32-bit system, a `__ptr64` pointer is truncated to a 32-bit pointer. This
qualifier is enabled under `-fms-extensions`.
  }];
}

def SPtrDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `__sptr` qualifier specifies that a 32-bit pointer should be sign
extended when converted to a 64-bit pointer.
  }];
}

def UPtrDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `__uptr` qualifier specifies that a 32-bit pointer should be zero
extended when converted to a 64-bit pointer.
  }];
}


def NullabilityDocs : DocumentationCategory<"Nullability Attributes"> {
  let Content = [{
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`).

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:

```c
// No meaningful result when 'ptr' is null (here, it happens to be undefined behavior).
int fetch(int * _Nonnull ptr) { return *ptr; }

// 'ptr' may be null.
int fetch_or_zero(int * _Nullable ptr) {
  return ptr ? *ptr : 0;
}

// A nullable pointer to non-null pointers to const characters.
const char *join_strings(const char * _Nonnull * _Nullable strings, unsigned n);
```

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:

```objective-c
@interface NSView : NSResponder
  - (nullable NSView *)ancestorSharedWithView:(nonnull NSView *)aView;
  @property (assign, nullable) NSView *superview;
  @property (readonly, nonnull) NSArray *subviews;
@end
```

As well as built-in pointer types, the nullability attributes can be attached
to C++ classes marked with the `_Nullable` attribute.

The following C++ standard library types are considered nullable:
`unique_ptr`, `shared_ptr`, `auto_ptr`, `exception_ptr`, `function`,
`move_only_function` and `coroutine_handle`.

Types should be marked nullable only where the type itself leaves nullability
ambiguous. For example, `std::optional` is not marked `_Nullable`, because
`optional<int> _Nullable` is redundant and `optional<int> _Nonnull` is
not a useful type. `std::weak_ptr` is not nullable, because its nullability
can change with no visible modification, so static annotation is unlikely to be
unhelpful.
  }];
}

def TypeNonNullDocs : Documentation {
  let Category = NullabilityDocs;
  let Content = [{
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:

```c
int fetch(int * _Nonnull ptr);
```

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.
  }];
}

def TypeNullableDocs : Documentation {
  let Category = NullabilityDocs;
  let Content = [{
The `_Nullable` nullability qualifier indicates that a value of the
`_Nullable` pointer type can be null. For example, given:

```c
int fetch_or_zero(int * _Nullable ptr);
```

a caller of `fetch_or_zero` can provide null.

The `_Nullable` attribute on classes indicates that the given class can
represent null values, and so the `_Nullable`, `_Nonnull` etc qualifiers
make sense for this type. For example:

```c
class _Nullable ArenaPointer { ... };

ArenaPointer _Nonnull x = ...;
ArenaPointer _Nullable y = nullptr;
```
  }];
}

def TypeNullableResultDocs : Documentation {
  let Category = NullabilityDocs;
  let Content = [{
The `_Nullable_result` nullability qualifier means that a value of the
`_Nullable_result` pointer can be `nil`, just like `_Nullable`. Where this
attribute differs from `_Nullable` is when it's used on a parameter to a
completion handler in a Swift async method. For instance, here:

```objc
-(void)fetchSomeDataWithID:(int)identifier
         completionHandler:(void (^)(Data *_Nullable_result result, NSError *error))completionHandler;
```

This method asynchronously calls `completionHandler` when the data is
available, or calls it with an error. `_Nullable_result` indicates to the
Swift importer that this is the uncommon case where `result` can get `nil`
even if no error has occurred, and will therefore import it as a Swift optional
type. Otherwise, if `result` was annotated with `_Nullable`, the Swift
importer will assume that `result` will always be non-nil unless an error
occurred.
}];
}

def TypeNullUnspecifiedDocs : Documentation {
  let Category = NullabilityDocs;
  let Content = [{
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.
  }];
}

def NonNullDocs : Documentation {
  let Category = NullabilityDocs;
  let Content = [{
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:

```c
extern void * my_memcpy (void *dest, const void *src, size_t len)
                __attribute__((nonnull (1, 2)));
```

Here, the `nonnull` attribute indicates that parameters 1 and 2
cannot have a null value. Omitting the parenthesized list of parameter indices
means that all parameters of pointer type cannot be null:

```c
extern void * my_memcpy (void *dest, const void *src, size_t len)
                __attribute__((nonnull));
```

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:

```c
extern void * my_memcpy (void *dest __attribute__((nonnull)),
                         const void *src __attribute__((nonnull)), size_t len);
```

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.
  }];
}

def RestrictDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "malloc";
  let Content = [{
The `malloc` attribute has two forms with different functionality. The first
is when it is used without arguments, where it marks that a function acts like
a system memory allocation function, returning a pointer to allocated storage
that does not alias storage from any other object accessible to the caller.

The second form is when `malloc` takes one or two arguments. The first
argument names a function that should be associated with this function as its
deallocation function. When this form is used, it enables the compiler to
diagnose when the incorrect deallocation function is used with this variable.
However the associated warning, spelled `-Wmismatched-dealloc` in GCC, is not
yet implemented in clang.
  }];
}

def MallocSpanDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "malloc_span";
  let Content = [{
The `malloc_span` attribute can be used to mark that a function which acts
like a system memory allocation function and returns a span-like structure,
where the returned memory range does not alias storage from any other object
accessible to the caller.

In this context, a span-like structure is assumed to have two non-static data
members, one of which is a pointer to the start of the allocated memory and
the other one is either an integer type containing the size of the actually
allocated memory or a pointer to the end of the allocated region. Note, static
data members do not impact whether a type is span-like or not.
  }];
}

def ReturnsNonNullDocs : Documentation {
  let Category = NullabilityDocs;
  let Content = [{
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:

```c
extern void * malloc (size_t size) __attribute__((returns_nonnull));
```

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
}];
}

def NoAliasDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `noalias` attribute indicates that the only memory accesses inside
function are loads and stores from objects pointed to by its pointer-typed
arguments, with arbitrary offsets.
  }];
}

def NSErrorDomainDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
In Cocoa frameworks in Objective-C, one can group related error codes in enums
and categorize these enums with error domains.

The `ns_error_domain` attribute indicates a global `NSString` or
`CFString` constant representing the error domain that an error code belongs
to. For pointer uniqueness and code size this is a constant symbol, not a
literal.

The domain and error code need to be used together. The `ns_error_domain`
attribute links error codes to their domain at the source level.

This metadata is useful for documentation purposes, for static analysis, and for
improving interoperability between Objective-C and Swift. It is not used for
code generation in Objective-C.

For example:

```objc
#define NS_ERROR_ENUM(_type, _name, _domain)  \
  enum _name : _type _name; enum __attribute__((ns_error_domain(_domain))) _name : _type

extern NSString *const MyErrorDomain;
typedef NS_ERROR_ENUM(unsigned char, MyErrorEnum, MyErrorDomain) {
  MyErrFirst,
  MyErrSecond,
};
```
  }];
}

def SwiftDocs : DocumentationCategory<"Customizing Swift Import"> {
  let Content = [{
Clang supports additional attributes for customizing how APIs are imported into
Swift.
  }];
}

def SwiftAsyncNameDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_async_name";
  let Content = [{
The `swift_async_name` attribute provides the name of the `async` overload for
the given declaration in Swift. If this attribute is absent, the name is
transformed according to the algorithm built into the Swift compiler.

The argument is a string literal that contains the Swift name of the function or
method. The name may be a compound Swift name. The function or method with such
an attribute must have more than zero parameters, as its last parameter is
assumed to be a callback that's eliminated in the Swift `async` name.

```objc
@interface URL
+ (void) loadContentsFrom:(URL *)url callback:(void (^)(NSData *))data __attribute__((__swift_async_name__("URL.loadContentsFrom(_:)")))
@end
```
  }];
}

def SwiftAttrDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_attr";
  let Content = [{
The `swift_attr` provides a Swift-specific annotation for the declaration
or type to which the attribute appertains to. It can be used on any declaration
or type in Clang. This kind of annotation is ignored by Clang as it doesn't have any
semantic meaning in languages supported by Clang. The Swift compiler can
interpret these annotations according to its own rules when importing C or
Objective-C declarations.
}];
}

def SwiftBridgeDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_bridge";
  let Content = [{
The `swift_bridge` attribute indicates that the declaration to which the
attribute appertains is bridged to the named Swift type.

```objc
__attribute__((__objc_root__))
@interface Base
- (instancetype)init;
@end

__attribute__((__swift_bridge__("BridgedI")))
@interface I : Base
@end
```

In this example, the Objective-C interface `I` will be made available to Swift
with the name `BridgedI`. It would be possible for the compiler to refer to
`I` still in order to bridge the type back to Objective-C.
  }];
}

def SwiftBridgedTypedefDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_bridged";
  let Content = [{
The `swift_bridged_typedef` attribute indicates that when the typedef to which
the attribute appertains is imported into Swift, it should refer to the bridged
Swift type (e.g. Swift's `String`) rather than the Objective-C type as written
(e.g. `NSString`).

```objc
@interface NSString;
typedef NSString *AliasedString __attribute__((__swift_bridged_typedef__));

extern void acceptsAliasedString(AliasedString _Nonnull parameter);
```

In this case, the function `acceptsAliasedString` will be imported into Swift
as a function which accepts a `String` type parameter.
  }];
}

def SwiftObjCMembersDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_objc_members";
  let Content = [{
This attribute indicates that Swift subclasses and members of Swift extensions
of this class will be implicitly marked with the `@objcMembers` Swift
attribute, exposing them back to Objective-C.
  }];
}

def SwiftErrorDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_error";
  let Content = [{
The `swift_error` attribute controls whether a particular function (or
Objective-C method) is imported into Swift as a throwing function, and if so,
which dynamic convention it uses.

All of these conventions except `none` require the function to have an error
parameter. Currently, the error parameter is always the last parameter of type
`NSError**` or `CFErrorRef*`. Swift will remove the error parameter from
the imported API. When calling the API, Swift will always pass a valid address
initialized to a null pointer.

- `swift_error(none)` means that the function should not be imported as
  throwing. The error parameter and result type will be imported normally.
- `swift_error(null_result)` means that calls to the function should be
  considered to have thrown if they return a null value. The return type must be
  a pointer type, and it will be imported into Swift with a non-optional type.
  This is the default error convention for Objective-C methods that return
  pointers.
- `swift_error(zero_result)` means that calls to the function should be
  considered to have thrown if they return a zero result. The return type must be
  an integral type. If the return type would have been imported as `Bool`, it
  is instead imported as `Void`. This is the default error convention for
  Objective-C methods that return a type that would be imported as `Bool`.
- `swift_error(nonzero_result)` means that calls to the function should be
  considered to have thrown if they return a non-zero result. The return type must
  be an integral type. If the return type would have been imported as `Bool`,
  it is instead imported as `Void`.
- `swift_error(nonnull_error)` means that calls to the function should be
  considered to have thrown if they leave a non-null error in the error parameter.
  The return type is left unmodified.

  }];
}

def SwiftNameDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_name";
  let Content = [{
The `swift_name` attribute provides the name of the declaration in Swift. If
this attribute is absent, the name is transformed according to the algorithm
built into the Swift compiler.

The argument is a string literal that contains the Swift name of the function,
variable, or type. When renaming a function, the name may be a compound Swift
name. For a type, enum constant, property, or variable declaration, the name
must be a simple or qualified identifier.

```objc
@interface URL
- (void) initWithString:(NSString *)s __attribute__((__swift_name__("URL.init(_:)")))
@end

void __attribute__((__swift_name__("squareRoot()"))) sqrt(double v) {
}
```
  }];
}

def SwiftNewTypeDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_newtype";
  let Content = [{
The `swift_newtype` attribute indicates that the typedef to which the
attribute appertains is imported as a new Swift type of the typedef's name.
Previously, the attribute was spelt `swift_wrapper`. While the behaviour of
the attribute is identical with either spelling, `swift_wrapper` is
deprecated, only exists for compatibility purposes, and should not be used in
new code.

- `swift_newtype(struct)` means that a Swift struct will be created for this
  typedef.

- `swift_newtype(enum)` means that a Swift enum will be created for this
  typedef.

  ```c
  // Import UIFontTextStyle as an enum type, with enumerated values being
  // constants.
  typedef NSString * UIFontTextStyle __attribute__((__swift_newtype__(enum)));

  // Import UIFontDescriptorFeatureKey as a structure type, with enumerated
  // values being members of the type structure.
  typedef NSString * UIFontDescriptorFeatureKey __attribute__((__swift_newtype__(struct)));
  ```

  }];
}

def SwiftPrivateDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_private";
  let Content = [{
Declarations marked with the `swift_private` attribute are hidden from the
framework client but are still made available for use within the framework or
Swift SDK overlay.

The purpose of this attribute is to permit a more idomatic implementation of
declarations in Swift while hiding the non-idiomatic one.
  }];
}

def OMPDeclareSimdDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "#pragma omp declare simd";
  let Content = [{
The `declare simd` construct can be applied to a function to enable the creation
of one or more versions that can process multiple arguments using SIMD
instructions from a single invocation in a SIMD loop. The `declare simd`
directive is a declarative directive. There may be multiple `declare simd`
directives for a function. The use of a `declare simd` construct on a function
enables the creation of SIMD versions of the associated function that can be
used to process multiple arguments from a single invocation from a SIMD loop
concurrently.
The syntax of the `declare simd` construct is as follows:

```none
#pragma omp declare simd [clause[[,] clause] ...] new-line
[#pragma omp declare simd [clause[[,] clause] ...] new-line]
[...]
function definition or declaration
```

where clause is one of the following:

```none
simdlen(length)
linear(argument-list[:constant-linear-step])
aligned(argument-list[:alignment])
uniform(argument-list)
inbranch
notinbranch
```

  }];
}

def OMPDeclareTargetDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "#pragma omp declare target";
  let Content = [{
The `declare target` directive specifies that variables and functions are mapped
to a device for OpenMP offload mechanism.

The syntax of the declare target directive is as follows:

```c
#pragma omp declare target new-line
declarations-definition-seq
#pragma omp end declare target new-line
```

or

```c
#pragma omp declare target (extended-list) new-line
```

or

```c
#pragma omp declare target clause[ [,] clause ... ] new-line
```

where clause is one of the following:

```c
to(extended-list)
link(list)
device_type(host | nohost | any)
```
  }];
}

def OMPDeclareVariantDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "#pragma omp declare variant";
  let Content = [{
The `declare variant` directive declares a specialized variant of a base
function and specifies the context in which that specialized variant is used.
The declare variant directive is a declarative directive.
The syntax of the `declare variant` construct is as follows:

```none
#pragma omp declare variant(variant-func-id) clause new-line
[#pragma omp declare variant(variant-func-id) clause new-line]
[...]
function definition or declaration
```

where clause is one of the following:

```none
match(context-selector-specification)
```

and where `variant-func-id` is the name of a function variant that is either a
base language identifier or, for C++, a template-id.

Clang provides the following context selector extensions, used via
`implementation={extension(EXTENSION)}`:

```none
match_all
match_any
match_none
disable_implicit_base
allow_templates
bind_to_declaration
```

The match extensions change when the *entire* context selector is considered a
match for an OpenMP context. The default is `all`, with `none` no trait in the
selector is allowed to be in the OpenMP context, with `any` a single trait in
both the selector and OpenMP context is sufficient. Only a single match
extension trait is allowed per context selector.
The disable extensions remove default effects of the `begin declare variant`
applied to a definition. If `disable_implicit_base` is given, we will not
introduce an implicit base function for a variant if no base function was
found. The variant is still generated but will never be called, due to the
absence of a base function and consequently calls to a base function.
The allow extensions change when the `begin declare variant` effect is
applied to a definition. If `allow_templates` is given, template function
definitions are considered as specializations of existing or assumed template
declarations with the same name. The template parameters for the base functions
are used to instantiate the specialization. If `bind_to_declaration` is given,
apply the same variant rules to function declarations. This allows the user to
override declarations with only a function declaration.
  }];
}

def LeafDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `leaf` attribute is used as a compiler hint to improve dataflow analysis
in library functions. Functions marked with the `leaf` attribute are not allowed
to jump back into the caller's translation unit, whether through invoking a
callback function, an external function call, use of `longjmp`, or other means.
Therefore, they cannot use or modify any data that does not escape the caller function's
compilation unit.

For more information see
`gcc documentation <https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html>`
}];
}

def OMPAssumeDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "assume";
  let Content = [{
Clang supports the `[[omp::assume("assumption")]]` attribute to
provide additional information to the optimizer. The string-literal, here
"assumption", will be attached to the function declaration such that later
analysis and optimization passes can assume the "assumption" to hold.
This is similar to {ref}`__builtin_assume <langext-__builtin_assume>` but
instead of an expression that can be assumed to be non-zero, the assumption is
expressed as a string and it holds for the entire function.

A function can have multiple assume attributes and they propagate from prior
declarations to later definitions. Multiple assumptions are aggregated into a
single comma separated string. Thus, one can provide multiple assumptions via
a comma separated string, i.a.,
`[[omp::assume("assumption1,assumption2")]]`.

While LLVM plugins might provide more assumption strings, the default LLVM
optimization passes are aware of the following assumptions:

```none
"omp_no_openmp"
"omp_no_openmp_routines"
"omp_no_parallelism"
"omp_no_openmp_constructs"
```

The OpenMP standard defines the meaning of OpenMP assumptions ("omp_XYZ" is
spelled "XYZ" in the [OpenMP 5.1 Standard][openmp 5.1 standard]).

[openmp 5.1 standard]: https://www.openmp.org/spec-html/5.1/openmpsu37.html#x56-560002.5.2

}];
}

def NoStackProtectorDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "no_stack_protector, safebuffers";
  let Content = [{
Clang supports the GNU style `__attribute__((no_stack_protector))` and Microsoft
style `__declspec(safebuffers)` attribute which disables
the stack protector on the specified function. This attribute is useful for
selectively disabling the stack protector on some functions when building with
`-fstack-protector` compiler option.

For example, it disables the stack protector for the function `foo` but function
`bar` will still be built with the stack protector with the `-fstack-protector`
option.

```c
int __attribute__((no_stack_protector))
foo (int x); // stack protection will be disabled for foo.

int bar(int y); // bar can be built with the stack protector.
```

    }];
}

def StackProtectorIgnoreDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `stack_protector_ignore` attribute skips analysis of the given local
variable when determining if a function should use a stack protector.

The `-fstack-protector` option uses a heuristic to only add stack protectors
to functions which contain variables or buffers over some size threshold. This
attribute overrides that heuristic for the attached variable, opting
them out. If this results in no variables or buffers remaining over the stack
protector threshold, then the function will no longer use a stack protector.
  }];
}

def StrictGuardStackCheckDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the Microsoft style `__declspec((strict_gs_check))` attribute
which upgrades the stack protector check from `-fstack-protector` to
`-fstack-protector-strong`.

For example, it upgrades the stack protector for the function `foo` to
`-fstack-protector-strong` but function `bar` will still be built with the
stack protector with the `-fstack-protector` option.

```c
__declspec((strict_gs_check))
int foo(int x); // stack protection will be upgraded for foo.

int bar(int y); // bar can be built with the standard stack protector checks.
```

    }];
}

def NotTailCalledDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `not_tail_called` attribute prevents tail-call optimization on statically
bound calls. Objective-c methods, and functions marked as `always_inline`
cannot be marked as `not_tail_called`.

For example, it prevents tail-call optimization in the following case:

```c
int __attribute__((not_tail_called)) foo1(int);

int foo2(int a) {
  return foo1(a); // No tail-call optimization on direct calls.
}
```

However, it doesn't prevent tail-call optimization in this case:

```c
int __attribute__((not_tail_called)) foo1(int);

int foo2(int a) {
  int (*fn)(int) = &foo1;

  // not_tail_called has no effect on an indirect call even if the call can
  // be resolved at compile time.
  return (*fn)(a);
}
```

Generally, marking an overriding virtual function as `not_tail_called` is
not useful, because this attribute is a property of the static type. Calls
made through a pointer or reference to the base class type will respect
the `not_tail_called` attribute of the base class's member function,
regardless of the runtime destination of the call:

```c++
struct Foo { virtual void f(); };
struct Bar : Foo {
  [[clang::not_tail_called]] void f() override;
};
void callera(Bar& bar) {
  Foo& foo = bar;
  // not_tail_called has no effect on here, even though the
  // underlying method is f from Bar.
  foo.f();
  bar.f(); // No tail-call optimization on here.
}
```
  }];
}

def NoThrowDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the GNU style `__attribute__((nothrow))` and Microsoft style
`__declspec(nothrow)` attribute as an equivalent of `noexcept` on function
declarations. This attribute informs the compiler that the annotated function
does not throw an exception. This prevents exception-unwinding. This attribute
is particularly useful on functions in the C Standard Library that are
guaranteed to not throw an exception.
    }];
}

def NoUwtableDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `nouwtable` attribute which skips emitting
the unwind table entry for the specified function. This attribute is useful for
selectively emitting the unwind table entry on some functions when building with
`-funwind-tables` compiler option.
    }];
}

def InternalLinkageDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `internal_linkage` attribute changes the linkage type of the declaration
to internal. This is similar to C-style `static`, but can be used on classes
and class methods. When applied to a class definition, this attribute affects
all methods and static data members of that class. This can be used to contain
the ABI of a C++ library by excluding unwanted class methods from the export
tables.
  }];
}

def ExcludeFromExplicitInstantiationDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `exclude_from_explicit_instantiation` attribute opts-out a member of a
class template from being part of explicit template instantiations of that
class template. This means that an explicit instantiation will not instantiate
members of the class template marked with the attribute, but also that code
where an extern template declaration of the enclosing class template is visible
will not take for granted that an external instantiation of the class template
would provide those members (which would otherwise be a link error, since the
explicit instantiation won't provide those members). For example, let's say we
don't want the `data()` method to be part of libc++'s ABI. To make sure it
is not exported from the dylib, we give it hidden visibility:

```c++
// in <string>
template <class CharT>
class basic_string {
public:
  __attribute__((__visibility__("hidden")))
  const value_type* data() const noexcept { ... }
};

template class basic_string<char>;
```

Since an explicit template instantiation declaration for `basic_string<char>`
is provided, the compiler is free to assume that `basic_string<char>::data()`
will be provided by another translation unit, and it is free to produce an
external call to this function. However, since `data()` has hidden visibility
and the explicit template instantiation is provided in a shared library (as
opposed to simply another translation unit), `basic_string<char>::data()`
won't be found and a link error will ensue. This happens because the compiler
assumes that `basic_string<char>::data()` is part of the explicit template
instantiation declaration, when it really isn't. To tell the compiler that
`data()` is not part of the explicit template instantiation declaration, the
`exclude_from_explicit_instantiation` attribute can be used:

```c++
// in <string>
template <class CharT>
class basic_string {
public:
  __attribute__((__visibility__("hidden")))
  __attribute__((exclude_from_explicit_instantiation))
  const value_type* data() const noexcept { ... }
};

template class basic_string<char>;
```

Now, the compiler won't assume that `basic_string<char>::data()` is provided
externally despite there being an explicit template instantiation declaration:
the compiler will implicitly instantiate `basic_string<char>::data()` in the
TUs where it is used.

This attribute can be used on static and non-static member functions of class
templates, static data members of class templates and member classes of class
templates.

**Interaction with \_\_declspec(dllexport/dllimport)**

For a DLL platform (i.e., Windows), this attribute also means "this member will
never be exported or imported". Despite its name, this semantics applies to
implicit instantiations and non-template entities as well.

```c++
// in <exception>
class __declspec(dllimport) nested_exception {
  ...
public:
  __attribute__((exclude_from_explicit_instantiation))
  exception_ptr nested_ptr() const noexcept { ... }
};
```

In this case, `nested_exception::nested_ptr` will never be attempted to be
imported.
  }];
}

def DisableTailCallsDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `disable_tail_calls` attribute instructs the backend to not perform tail
call optimization inside the marked function.

For example:

```c
int callee(int);

int foo(int a) __attribute__((disable_tail_calls)) {
  return callee(a); // This call is not tail-call optimized.
}
```

Marking virtual functions as `disable_tail_calls` is legal.

```c++
int callee(int);

class Base {
public:
  [[clang::disable_tail_calls]] virtual int foo1() {
    return callee(); // This call is not tail-call optimized.
  }
};

class Derived1 : public Base {
public:
  int foo1() override {
    return callee(); // This call is tail-call optimized.
  }
};
```

  }];
}

def AnyX86InterruptDocs : Documentation {
    let Category = DocCatFunction;
    let Heading = "interrupt (X86)";
    let Content = [{
Clang supports the GNU style `__attribute__((interrupt))` attribute on X86
targets. This attribute may be attached to a function definition and instructs
the backend to generate appropriate function entry/exit code so that it can be
used directly as an interrupt service routine.

Interrupt handlers have access to the stack frame pushed onto the stack by the processor,
and return using the `IRET` instruction. All registers in an interrupt handler are callee-saved.
Exception handlers also have access to the error code pushed onto the stack by the processor,
when applicable.

An interrupt handler must take the following arguments:

```c
__attribute__ ((interrupt))
void f (struct stack_frame *frame) {
    ...
}
```

Where `struct stack_frame` is a suitable struct matching the stack frame pushed
by the processor.

An exception handler must take the following arguments:

```c
__attribute__ ((interrupt))
void g (struct stack_frame *frame, unsigned long code) {
    ...
}
```

On 32-bit targets, the `code` argument should be of type `unsigned int`.

Exception handlers should only be used when an error code is pushed by the processor.
Using the incorrect handler type will crash the system.

Interrupt and exception handlers cannot be called by other functions and must have return type `void`.

Interrupt and exception handlers should only call functions with the 'no_caller_saved_registers'
attribute, or should be compiled with the '-mgeneral-regs-only' flag to avoid saving unused
non-GPR registers.
    }];
}

def AnyX86NoCallerSavedRegistersDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use this attribute to indicate that the specified function has no
caller-saved registers. That is, all registers are callee-saved except for
registers used for passing parameters to the function or returning parameters
from the function.
The compiler saves and restores any modified registers that were not used for
passing or returning arguments to the function.

The user can call functions specified with the 'no_caller_saved_registers'
attribute from an interrupt handler without saving and restoring all
call-clobbered registers.

Functions specified with the 'no_caller_saved_registers' attribute should only
call other functions with the 'no_caller_saved_registers' attribute, or should be
compiled with the '-mgeneral-regs-only' flag to avoid saving unused non-GPR registers.

Note that 'no_caller_saved_registers' attribute is not a calling convention.
In fact, it only overrides the decision of which registers should be saved by
the caller, but not how the parameters are passed from the caller to the callee.

For example:

```c
__attribute__ ((no_caller_saved_registers, fastcall))
void f (int arg1, int arg2) {
  ...
}
```

In this case parameters 'arg1' and 'arg2' will be passed in registers.
In this case, on 32-bit x86 targets, the function 'f' will use ECX and EDX as
register parameters. However, it will not assume any scratch registers and
should save and restore any modified registers except for ECX and EDX.
  }];
}

def X86ForceAlignArgPointerDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Use this attribute to force stack alignment.

Legacy x86 code uses 4-byte stack alignment. Newer aligned SSE instructions
(like 'movaps') that work with the stack require operands to be 16-byte aligned.
This attribute realigns the stack in the function prologue to make sure the
stack can be used with SSE instructions.

Note that the x86_64 ABI forces 16-byte stack alignment at the call site.
Because of this, 'force_align_arg_pointer' is not needed on x86_64, except in
rare cases where the caller does not align the stack properly (e.g. flow
jumps from i386 arch code).

```c
__attribute__ ((force_align_arg_pointer))
void f () {
  ...
}
```

  }];
}

def AnyX86NoCfCheckDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Jump Oriented Programming attacks rely on tampering with addresses used by
indirect call / jmp, e.g. redirect control-flow to non-programmer
intended bytes in the binary.
X86 Supports Indirect Branch Tracking (IBT) as part of Control-Flow
Enforcement Technology (CET). IBT instruments ENDBR instructions used to
specify valid targets of indirect call / jmp.
The `nocf_check` attribute has two roles:
1\. Appertains to a function - do not add ENDBR instruction at the beginning of
the function.
2\. Appertains to a function pointer - do not track the target function of this
pointer (by adding nocf_check prefix to the indirect-call instruction).
}];
}

def SwiftCallDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `swiftcall` attribute indicates that a function should be called
using the Swift calling convention for a function or function pointer.

The lowering for the Swift calling convention, as described by the Swift
ABI documentation, occurs in multiple phases. The first, "high-level"
phase breaks down the formal parameters and results into innately direct
and indirect components, adds implicit parameters for the generic
signature, and assigns the context and error ABI treatments to parameters
where applicable. The second phase breaks down the direct parameters
and results from the first phase and assigns them to registers or the
stack. The `swiftcall` convention only handles this second phase of
lowering; the C function type must accurately reflect the results
of the first phase, as follows:

- Results classified as indirect by high-level lowering should be
  represented as parameters with the `swift_indirect_result` attribute.

- Results classified as direct by high-level lowering should be represented
  as follows:

  - First, remove any empty direct results.
  - If there are no direct results, the C result type should be `void`.
  - If there is one direct result, the C result type should be a type with
    the exact layout of that result type.
  - If there are a multiple direct results, the C result type should be
    a struct type with the exact layout of a tuple of those results.

- Parameters classified as indirect by high-level lowering should be
  represented as parameters of pointer type.

- Parameters classified as direct by high-level lowering should be
  omitted if they are empty types; otherwise, they should be represented
  as a parameter type with a layout exactly matching the layout of the
  Swift parameter type.

- The context parameter, if present, should be represented as a trailing
  parameter with the `swift_context` attribute.

- The error result parameter, if present, should be represented as a
  trailing parameter (always following a context parameter) with the
  `swift_error_result` attribute.

`swiftcall` does not support variadic arguments or unprototyped functions.

The parameter ABI treatment attributes are aspects of the function type.
A function type which applies an ABI treatment attribute to a
parameter is a different type from an otherwise-identical function type
that does not. A single parameter may not have multiple ABI treatment
attributes.

Support for this feature is target-dependent, although it should be
supported on every target that Swift supports. Query for this attribute
with `__has_attribute(swiftcall)`. Query if the target supports the
calling convention with `__has_extension(swiftcc)`. This implies
support for the `swift_context`, `swift_error_result`, and
`swift_indirect_result` attributes.

Since this attribute follows the Swift calling convention, it is
considered ABI-unstable except on targets where the Swift project
has declared ABI stability. Users are responsible for ensuring that
calls and definitions of functions with this attribute are compiled
with compatible compilers. Note that different operating systems
on the same architecture may use different ABIs and therefore may
have different standards for ABI stability.
  }];
}

def SwiftContextDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `swift_context` attribute marks a parameter of a `swiftcall`
or `swiftasynccall` function as having the special context-parameter
ABI treatment.

This treatment generally passes the context value in a special register
which is normally callee-preserved.

A `swift_context` parameter must either be the last parameter or must be
followed by a `swift_error_result` parameter (which itself must always be
the last parameter).

A context parameter must have pointer or reference type.
  }];
}

def SwiftAsyncCallDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `swiftasynccall` attribute indicates that a function is
compatible with the low-level conventions of Swift async functions,
provided it declares the right formal arguments.

In most respects, this is similar to the `swiftcall` attribute, except for
the following:

- A parameter may be marked `swift_async_context`, `swift_context`
  or `swift_indirect_result` (with the same restrictions on parameter
  ordering as `swiftcall`) but the parameter attribute
  `swift_error_result` is not permitted.
- A `swiftasynccall` function must have return type `void`.
- Within a `swiftasynccall` function, a call to a `swiftasynccall`
  function that is the immediate operand of a `return` statement is
  guaranteed to be performed as a tail call. This syntax is allowed even
  in C as an extension (a call to a void-returning function cannot be a
  return operand in standard C). If something in the calling function would
  semantically be performed after a guaranteed tail call, such as the
  non-trivial destruction of a local variable or temporary,
  then the program is ill-formed.

Query for this attribute with `__has_attribute(swiftasynccall)`. Query if
the target supports the calling convention with
`__has_extension(swiftasynccc)`.

Since this attribute follows the Swift async calling convention, it is
considered ABI-unstable except on targets where the Swift project
has declared ABI stability. Users are responsible for ensuring that
calls and definitions of functions with this attribute are compiled
with compatible compilers. Note that different operating systems
on the same architecture may use different ABIs and therefore may
have different standards for ABI stability.
  }];
}

def SwiftAsyncContextDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `swift_async_context` attribute marks a parameter of a `swiftasynccall`
function as having the special asynchronous context-parameter ABI treatment.

If the function is not `swiftasynccall`, this attribute only generates
extended frame information.

A context parameter must have pointer or reference type.
  }];
}

def SwiftErrorResultDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `swift_error_result` attribute marks a parameter of a `swiftcall`
function as having the special error-result ABI treatment.

This treatment generally passes the underlying error value in and out of
the function through a special register which is normally callee-preserved.
This is modeled in C by pretending that the register is addressable memory:

- The caller appears to pass the address of a variable of pointer type.
  The current value of this variable is copied into the register before
  the call; if the call returns normally, the value is copied back into the
  variable.
- The callee appears to receive the address of a variable. This address
  is actually a hidden location in its own stack, initialized with the
  value of the register upon entry. When the function returns normally,
  the value in that hidden location is written back to the register.

A `swift_error_result` parameter must be the last parameter, and it must be
preceded by a `swift_context` parameter.

A `swift_error_result` parameter must have type `T**` or `T*&` for some
type T. Note that no qualifiers are permitted on the intermediate level.

It is undefined behavior if the caller does not pass a pointer or
reference to a valid object.

The standard convention is that the error value itself (that is, the
value stored in the apparent argument) will be null upon function entry,
but this is not enforced by the ABI.
  }];
}

def SwiftIndirectResultDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `swift_indirect_result` attribute marks a parameter of a `swiftcall`
or `swiftasynccall` function as having the special indirect-result ABI
treatment.

This treatment gives the parameter the target's normal indirect-result
ABI treatment, which may involve passing it differently from an ordinary
parameter. However, only the first indirect result will receive this
treatment. Furthermore, low-level lowering may decide that a direct result
must be returned indirectly; if so, this will take priority over the
`swift_indirect_result` parameters.

A `swift_indirect_result` parameter must either be the first parameter or
follow another `swift_indirect_result` parameter.

A `swift_indirect_result` parameter must have type `T*` or `T&` for
some object type `T`. If `T` is a complete type at the point of
definition of a function, it is undefined behavior if the argument
value does not point to storage of adequate size and alignment for a
value of type `T`.

Making indirect results explicit in the signature allows C functions to
directly construct objects into them without relying on language
optimizations like C++'s named return value optimization (NRVO).
  }];
}

def SwiftAsyncDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_async";
  let Content = [{
The `swift_async` attribute specifies if and how a particular function or
Objective-C method is imported into a swift async method. For instance:

```objc
@interface MyClass : NSObject
-(void)notActuallyAsync:(int)p1 withCompletionHandler:(void (^)())handler
    __attribute__((swift_async(none)));

-(void)actuallyAsync:(int)p1 callThisAsync:(void (^)())fun
    __attribute__((swift_async(swift_private, 1)));
@end
```

Here, `notActuallyAsync:withCompletionHandler` would have been imported as
`async` (because it's last parameter's selector piece is
`withCompletionHandler`) if not for the `swift_async(none)` attribute.
Conversely, `actuallyAsync:callThisAsync` wouldn't have been imported as
`async` if not for the `swift_async` attribute because it doesn't match the
naming convention.

When using `swift_async` to enable importing, the first argument to the
attribute is either `swift_private` or `not_swift_private` to indicate
whether the function/method is private to the current framework, and the second
argument is the index of the completion handler parameter.
  }];
}

def SwiftAsyncErrorDocs : Documentation {
  let Category = SwiftDocs;
  let Heading = "swift_async_error";
  let Content = [{
The `swift_async_error` attribute specifies how an error state will be
represented in a swift async method. It's a bit analogous to the `swift_error`
attribute for the generated async method. The `swift_async_error` attribute
can indicate a variety of different ways of representing an error.

- `__attribute__((swift_async_error(zero_argument, N)))`, specifies that the
  async method is considered to have failed if the Nth argument to the
  completion handler is zero.
- `__attribute__((swift_async_error(nonzero_argument, N)))`, specifies that
  the async method is considered to have failed if the Nth argument to the
  completion handler is non-zero.
- `__attribute__((swift_async_error(nonnull_error)))`, specifies that the
  async method is considered to have failed if the `NSError *` argument to the
  completion handler is non-null.
- `__attribute__((swift_async_error(none)))`, specifies that the async method
  cannot fail.

For instance:

```objc
@interface MyClass : NSObject
-(void)asyncMethod:(void (^)(char, int, float))handler
    __attribute__((swift_async(swift_private, 1)))
    __attribute__((swift_async_error(zero_argument, 2)));
@end
```

Here, the `swift_async` attribute specifies that `handler` is the completion
handler for this method, and the `swift_async_error` attribute specifies that
the `int` parameter is the one that represents the error.
}];
}

def SuppressDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
The `suppress` attribute suppresses unwanted warnings coming from static
analysis tools such as the Clang Static Analyzer. The tool will not report
any issues in source code annotated with the attribute.

The attribute cannot be used to suppress traditional Clang warnings, because
many such warnings are emitted before the attribute is fully parsed.
Consider using `#pragma clang diagnostic` to control such diagnostics,
as described in
{ref}`Controlling Diagnostics via Pragmas <pragma-gcc-diagnostic>`.

The `suppress` attribute can be placed on an individual statement in order to
suppress warnings about undesirable behavior occurring at that statement:

```c++
int foo() {
  int *x = nullptr;
  ...
  [[clang::suppress]]
  return *x;  // null pointer dereference warning suppressed here
}
```

Putting the attribute on a compound statement suppresses all warnings in scope:

```c++
int foo() {
  [[clang::suppress]] {
    int *x = nullptr;
    ...
    return *x;  // warnings suppressed in the entire scope
  }
}
```

The attribute can also be placed on entire declarations of functions, classes,
variables, member variables, and so on, to suppress warnings related
to the declarations themselves. When used this way, the attribute additionally
suppresses all warnings in the lexical scope of the declaration:

```c++
class [[clang::suppress]] C {
  int foo() {
    int *x = nullptr;
    ...
    return *x;  // warnings suppressed in the entire class scope
  }

  int bar();
};

int C::bar() {
  int *x = nullptr;
  ...
  return *x;  // warning NOT suppressed! - not lexically nested in 'class C{}'
}
```

Some static analysis warnings are accompanied by one or more notes, and the
line of code against which the warning is emitted isn't necessarily the best
for suppression purposes. In such cases the tools are allowed to implement
additional ways to suppress specific warnings based on the attribute attached
to a note location.

For example, the Clang Static Analyzer suppresses memory leak warnings when
the suppression attribute is placed at the allocation site (highlited by
a "note: memory is allocated"), which may be different from the line of code
at which the program "loses track" of the pointer (where the warning
is ultimately emitted):

```c
int bar1(bool coin_flip) {
  __attribute__((suppress))
  int *result = (int *)malloc(sizeof(int));
  if (coin_flip)
    return 1;  // warning about this leak path is suppressed

  return *result;  // warning about this leak path is also suppressed
}

int bar2(bool coin_flip) {
  int *result = (int *)malloc(sizeof(int));
  if (coin_flip)
    return 1;  // leak warning on this path NOT suppressed

  __attribute__((suppress))
  return *result;  // leak warning is suppressed only on this path
}
```

When written as `[[gsl::suppress]]`, this attribute suppresses specific
clang-tidy diagnostics for rules of the [C++ Core Guidelines][c++ core guidelines] in a portable
way. The attribute can be attached to declarations, statements, and at
namespace scope.

```c++
[[gsl::suppress("Rh-public")]]
void f_() {
  int *p;
  [[gsl::suppress("type")]] {
    p = reinterpret_cast<int*>(7);
  }
}
namespace N {
  [[clang::suppress("type", "bounds")]];
  ...
}
```

[c++ core guidelines]: https://github.com/isocpp/CppCoreGuidelines/blob/master/CppCoreGuidelines.md#inforce-enforcement
  }];
}

def AbiTagsDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `abi_tag` attribute can be applied to a function, variable, class or
inline namespace declaration to modify the mangled name of the entity. It gives
the ability to distinguish between different versions of the same entity but
with different ABI versions supported. For example, a newer version of a class
could have a different set of data members and thus have a different size. Using
the `abi_tag` attribute, it is possible to have different mangled names for
a global variable of the class type. Therefore, the old code could keep using
the old mangled name and the new code will use the new mangled name with tags.
  }];
}

def BuiltinAliasDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "clang::builtin_alias, clang_builtin_alias";
  let Content = [{
This attribute is used in the implementation of the C intrinsics.
It allows the C intrinsic functions to be declared using the names defined
in target builtins, and still be recognized as clang builtins equivalent to the
underlying name. For example, `riscv_vector.h` declares the function `vadd`
with `__attribute__((clang_builtin_alias(__builtin_rvv_vadd_vv_i8m1)))`.
This ensures that both functions are recognized as that clang builtin,
and in the latter case, the choice of which builtin to identify the
function as can be deferred until after overload resolution.

This attribute can only be used to set up the aliases for certain ARM/RISC-V
C intrinsic functions; it is intended for use only inside `arm_*.h` and
`riscv_*.h` and is not a general mechanism for declaring arbitrary aliases
for clang builtin functions.
  }];
}

def RISCVVectorCCDocs : Documentation {
 let Category = DocCatCallingConvs;
 let Heading = "riscv::vector_cc, riscv_vector_cc, clang::riscv_vector_cc";
 let Content = [{
The `riscv_vector_cc` attribute can be applied to a function. It preserves 15
registers namely, v1-v7 and v24-v31 as callee-saved. Callers thus don't need
to save these registers before function calls, and callees only need to save
them if they use them.
 }];
}

def RISCVVLSCCDocs : Documentation {
 let Category = DocCatCallingConvs;
 let Heading = "riscv::vls_cc, riscv_vls_cc, clang::riscv_vls_cc";
 let Content = [{
The `riscv_vls_cc` attribute can be applied to a function. Functions
declared with this attribute will utilize the standard fixed-length vector
calling convention variant instead of the default calling convention defined by
the ABI. This variant aims to pass fixed-length vectors via vector registers,
if possible, rather than through general-purpose registers.}];
}

def PreferredNameDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `preferred_name` attribute can be applied to a class template, and
specifies a preferred way of naming a specialization of the template. The
preferred name will be used whenever the corresponding template specialization
would otherwise be printed in a diagnostic or similar context.

The preferred name must be a typedef or type alias declaration that refers to a
specialization of the class template (not including any type qualifiers). In
general this requires the template to be declared at least twice. For example:

```c++
template<typename T> struct basic_string;
using string = basic_string<char>;
using wstring = basic_string<wchar_t>;
template<typename T> struct [[clang::preferred_name(string),
                              clang::preferred_name(wstring)]] basic_string {
  // ...
};
```

Note that the `preferred_name` attribute will be ignored when the compiler
writes a C++20 Module interface now. This is due to a compiler issue
(<https://github.com/llvm/llvm-project/issues/56490>) that blocks users to modularize
declarations with `preferred_name`. This is intended to be fixed in the future.
  }];
}

def PreserveMostDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On X86-64 and AArch64 targets, this attribute changes the calling convention of
a function. The `preserve_most` calling convention attempts to make the code
in the caller as unintrusive as possible. This convention behaves identically
to the `C` calling convention on how arguments and return values are passed,
but it uses a different set of caller/callee-saved registers. This alleviates
the burden of saving and recovering a large register set before and after the
call in the caller. If the arguments are passed in callee-saved registers,
then they will be preserved by the callee across the call. This doesn't
apply for values returned in callee-saved registers.

- On X86-64 the callee preserves all general purpose registers, except for
  R11. R11 can be used as a scratch register. Floating-point registers
  (XMMs/YMMs) are not preserved and need to be saved by the caller.
- On AArch64 the callee preserve all general purpose registers, except X0-X8 and
  X16-X18.

The idea behind this convention is to support calls to runtime functions
that have a hot path and a cold path. The hot path is usually a small piece
of code that doesn't use many registers. The cold path might need to call out to
another function and therefore only needs to preserve the caller-saved
registers, which haven't already been saved by the caller. The
`preserve_most` calling convention is very similar to the `cold` calling
convention in terms of caller/callee-saved registers, but they are used for
different types of function calls. `coldcc` is for function calls that are
rarely executed, whereas `preserve_most` function calls are intended to be
on the hot path and definitely executed a lot. Furthermore `preserve_most`
doesn't prevent the inliner from inlining the function call.

This calling convention will be used by a future version of the Objective-C
runtime and should therefore still be considered experimental at this time.
Although this convention was created to optimize certain runtime calls to
the Objective-C runtime, it is not limited to this runtime and might be used
by other runtimes in the future too. The current implementation only
supports X86-64 and AArch64, but the intention is to support more architectures
in the future.
  }];
}

def PreserveAllDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On X86-64 and AArch64 targets, this attribute changes the calling convention of
a function. The `preserve_all` calling convention attempts to make the code
in the caller even less intrusive than the `preserve_most` calling convention.
This calling convention also behaves identical to the `C` calling convention
on how arguments and return values are passed, but it uses a different set of
caller/callee-saved registers. This removes the burden of saving and
recovering a large register set before and after the call in the caller. If
the arguments are passed in callee-saved registers, then they will be
preserved by the callee across the call. This doesn't apply for values
returned in callee-saved registers.

- On X86-64 the callee preserves all general purpose registers, except for
  R11. R11 can be used as a scratch register. Furthermore it also preserves
  all floating-point registers (XMMs/YMMs).
- On AArch64 the callee preserve all general purpose registers, except X0-X8 and
  X16-X18. Furthermore it also preserves lower 128 bits of V8-V31 SIMD - floating
  point registers.

The idea behind this convention is to support calls to runtime functions
that don't need to call out to any other functions.

This calling convention, like the `preserve_most` calling convention, will be
used by a future version of the Objective-C runtime and should be considered
experimental at this time.
  }];
}

def PreserveNoneDocs : Documentation {
  let Category = DocCatCallingConvs;
  let Content = [{
On X86-64 and AArch64 targets, this attribute changes the calling convention of a function.
The `preserve_none` calling convention tries to preserve as few general
registers as possible. So all general registers are caller saved registers. It
also uses more general registers to pass arguments. This attribute doesn't
impact floating-point registers. `preserve_none`'s ABI is still unstable, and
may be changed in the future.

- On X86-64, only RSP and RBP are preserved by the callee.
  Registers R12, R13, R14, R15, RDI, RSI, RDX, RCX, R8, R9, R11, and RAX now can
  be used to pass function arguments. Floating-point registers (XMMs/YMMs) still
  follow the C calling convention.
- On AArch64, only LR and FP are preserved by the callee.
  Registers X20-X28, X0-X7, and X9-X14 are used to pass function arguments.
  X8, X16-X19, SIMD and floating-point registers follow the AAPCS calling
  convention. X15 is not available for argument passing on Windows, but is
  used to pass arguments on other platforms.
  }];
}

def DeprecatedDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `deprecated` attribute can be applied to a function, a variable, or a
type. This is useful when identifying functions, variables, or types that are
expected to be removed in a future version of a program.

Consider the function declaration for a hypothetical function `f`:

```c++
void f(void) __attribute__((deprecated("message", "replacement")));
```

When spelled as `__attribute__((deprecated))`, the deprecated attribute can have
two optional string arguments. The first one is the message to display when
emitting the warning; the second one enables the compiler to provide a Fix-It
to replace the deprecated name with a new name. Otherwise, when spelled as
`[[gnu::deprecated]]` or `[[deprecated]]`, the attribute can have one optional
string argument which is the message to display when emitting the warning.
  }];
}

def IFuncDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
`__attribute__((ifunc("resolver")))` is used to mark that the address of a
declaration should be resolved at runtime by calling a resolver function.

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 return a pointer.

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.

Not all targets support this attribute:

- ELF target support depends on both the linker and runtime linker, and is
  available in at least lld 4.0 and later, binutils 2.20.1 and later, glibc
  v2.11.1 and later, and FreeBSD 9.1 and later.
- Mach-O targets support it, but with slightly different semantics: the resolver
  is run at first call, instead of at load time by the runtime linker.
- Windows target supports it on AArch64, but with different semantics: the
  `ifunc` is replaced with a global function pointer, and the call is replaced
  with an indirect call. The function pointer is initialized by a constructor
  that calls the resolver.
- Baremetal target supports it on AVR.
- AIX/XCOFF supports it via a compiler-only solution. An ifunc appears as a
  regular function (has an entry point `.foo[PR]` and a function descriptor
  `foo[DS]`). The entry point is a stub that branches to the function address
  in the descriptor, and the descriptor is initialized via a constructor
  function (`__init_ifuncs`) that is linked into every shared object and
  executable. `__init_ifuncs` calls the resolver of each ifunc and stores the
  result in the corresponding descriptor.
- Other targets currently do not support this attribute.
  }];
}

def LTOVisibilityDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
See {doc}`LTOVisibility`.
  }];
}

def TypeVisibilityDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `type_visibility` attribute allows the visibility of a type and its vague
linkage objects (vtable, typeinfo, typeinfo name) to be controlled separately from
the visibility of functions and data members of the type.

For example, this can be used to give default visibility to the typeinfo and the vtable
of a type while still keeping hidden visibility on its member functions and static data
members.

This attribute can only be applied to types and namespaces.

If both `visibility` and `type_visibility` are applied to a type or a namespace, the
visibility specified with the `type_visibility` attribute overrides the visibility
provided with the regular `visibility` attribute.
  }];
}

def XRayDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "xray_always_instrument, xray_never_instrument, xray_log_args";
  let Content = [{
`__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.

Conversely, `__attribute__((xray_never_instrument))` or
`[[clang::xray_never_instrument]]` will inhibit the insertion of these
instrumentation points.

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.

`__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.
  }];
}

def PatchableFunctionEntryDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
`__attribute__((patchable_function_entry(N,M,Section)))` is used to generate M
NOPs before the function entry and N-M NOPs after the function entry, with a record of
the entry stored in section `Section`. This attribute takes precedence over the
command line option `-fpatchable-function-entry=N,M,Section`. `M` defaults to 0
if omitted. `Section` defaults to the `-fpatchable-function-entry` section name if
set, or to `__patchable_function_entries` otherwise.

This attribute is only supported on
aarch64/aarch64-be/loongarch32/loongarch64/riscv32/riscv64/i386/x86-64/ppc/ppc64/ppc64le/s390x targets.
For ppc/ppc64 targets, AIX is still not supported.
}];
}

def HotFunctionEntryDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
`__attribute__((hot))` marks a function as hot, as a manual alternative to PGO hotness data.
If PGO data is available, the annotation `__attribute__((hot))` overrides the profile count based hotness (unlike `__attribute__((cold))`).
}];
}

def ColdFunctionEntryDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
`__attribute__((cold))` marks a function as cold, as a manual alternative to PGO hotness data.
If PGO data is available, the profile count based hotness overrides the `__attribute__((cold))` annotation (unlike `__attribute__((hot))`).
}];
}
def TransparentUnionDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute can be applied to a union to change the behavior of calls to
functions that have an argument with a transparent union type. The compiler
behavior is changed in the following manner:

- A value whose type is any member of the transparent union can be passed as an
  argument without the need to cast that value.
- The argument is passed to the function using the calling convention of the
  first member of the transparent union. Consequently, all the members of the
  transparent union should have the same calling convention as its first member.

Transparent unions are not supported in C++.
  }];
}

def ObjCSubclassingRestrictedDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute can be added to an Objective-C `@interface` declaration to
ensure that this class cannot be subclassed.
  }];
}

def ObjCNonLazyClassDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute can be added to an Objective-C `@interface` or
`@implementation` declaration to add the class to the list of non-lazily
initialized classes. A non-lazy class will be initialized eagerly when the
Objective-C runtime is loaded. This is required for certain system classes which
have instances allocated in non-standard ways, such as the classes for blocks
and constant strings. Adding this attribute is essentially equivalent to
providing a trivial `+load` method but avoids the (fairly small) load-time
overheads associated with defining and calling such a method.
  }];
}

def ObjCDirectDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `objc_direct` attribute can be used to mark an Objective-C method as
being *direct*. A direct method is treated statically like an ordinary method,
but dynamically it behaves more like a C function. This lowers some of the costs
associated with the method but also sacrifices some of the ordinary capabilities
of Objective-C methods.

A message send of a direct method calls the implementation directly, as if it
were a C function, rather than using ordinary Objective-C method dispatch. This
is substantially faster and potentially allows the implementation to be inlined,
but it also means the method cannot be overridden in subclasses or replaced
dynamically, as ordinary Objective-C methods can.

Furthermore, a direct method is not listed in the class's method lists. This
substantially reduces the code-size overhead of the method but also means it
cannot be called dynamically using ordinary Objective-C method dispatch at all;
in particular, this means that it cannot override a superclass method or satisfy
a protocol requirement.

Because a direct method cannot be overridden, it is an error to perform
a `super` message send of one.

Although a message send of a direct method causes the method to be called
directly as if it were a C function, it still obeys Objective-C semantics in other
ways:

- If the receiver is `nil`, the message send does nothing and returns the zero value
  for the return type.
- A message send of a direct class method will cause the class to be initialized,
  including calling the `+initialize` method if present.
- The implicit `_cmd` parameter containing the method's selector is still defined.
  In order to minimize code-size costs, the implementation will not emit a reference
  to the selector if the parameter is unused within the method.

Symbols for direct method implementations are implicitly given hidden
visibility, meaning that they can only be called within the same linkage unit.

It is an error to do any of the following:

- declare a direct method in a protocol,
- declare an override of a direct method with a method in a subclass,
- declare an override of a non-direct method with a direct method in a subclass,
- declare a method with different directness in different class interfaces, or
- implement a non-direct method (as declared in any class interface) with a direct method.

If any of these rules would be violated if every method defined in an
`@implementation` within a single linkage unit were declared in an
appropriate class interface, the program is ill-formed with no diagnostic
required. If a violation of this rule is not diagnosed, behavior remains
well-defined; this paragraph is simply reserving the right to diagnose such
conflicts in the future, not to treat them as undefined behavior.

Additionally, Clang will warn about any `@selector` expression that
names a selector that is only known to be used for direct methods.

For the purpose of these rules, a "class interface" includes a class's primary
`@interface` block, its class extensions, its categories, its declared protocols,
and all the class interfaces of its superclasses.

An Objective-C property can be declared with the `direct` property
attribute. If a direct property declaration causes an implicit declaration of
a getter or setter method (that is, if the given method is not explicitly
declared elsewhere), the method is declared to be direct.

Some programmers may wish to make many methods direct at once. In order
to simplify this, the `objc_direct_members` attribute is provided; see its
documentation for more information.
  }];
}

def ObjCDirectMembersDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `objc_direct_members` attribute can be placed on an Objective-C
`@interface` or `@implementation` to mark that methods declared
therein should be considered direct by default. See the documentation
for `objc_direct` for more information about direct methods.

When `objc_direct_members` is placed on an `@interface` block, every
method in the block is considered to be declared as direct. This includes any
implicit method declarations introduced by property declarations. If the method
redeclares a non-direct method, the declaration is ill-formed, exactly as if the
method was annotated with the `objc_direct` attribute.

When `objc_direct_members` is placed on an `@implementation` block,
methods defined in the block are considered to be declared as direct unless
they have been previously declared as non-direct in any interface of the class.
This includes the implicit method definitions introduced by synthesized
properties, including auto-synthesized properties.
  }];
}

def ObjCNonRuntimeProtocolDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `objc_non_runtime_protocol` attribute can be used to mark that an
Objective-C protocol is only used during static type-checking and doesn't need
to be represented dynamically. This avoids several small code-size and run-time
overheads associated with handling the protocol's metadata. A non-runtime
protocol cannot be used as the operand of a `@protocol` expression, and
dynamic attempts to find it with `objc_getProtocol` will fail.

If a non-runtime protocol inherits from any ordinary protocols, classes and
derived protocols that declare conformance to the non-runtime protocol will
dynamically list their conformance to those bare protocols.
  }];
}

def SelectAnyDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
This attribute appertains to a global symbol, causing it to have a weak
definition (
[linkonce](https://llvm.org/docs/LangRef.html#linkage-types)
), allowing the linker to select any definition.

For more information see
[gcc documentation](https://gcc.gnu.org/onlinedocs/gcc-7.2.0/gcc/Microsoft-Windows-Variable-Attributes.html)
or [msvc documentation](https://docs.microsoft.com/pl-pl/cpp/cpp/selectany).
}]; }

def HybridPatchableDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `hybrid_patchable` attribute declares an ARM64EC function with an additional
x86-64 thunk, which may be patched at runtime.

For more information see
[ARM64EC ABI documentation](https://learn.microsoft.com/en-us/windows/arm/arm64ec-abi).
}]; }

def WebAssemblyExportNameDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((export_name(<name>)))`
attribute for the WebAssembly target. This attribute may be attached to a
function declaration, where it modifies how the symbol is to be exported
from the linked WebAssembly.

WebAssembly functions are exported via string name. By default when a symbol
is exported, the export name for C/C++ symbols are the same as their C/C++
symbol names. This attribute can be used to override the default behavior, and
request a specific string name be used instead.
  }];
}

def WebAssemblyImportModuleDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((import_module(<module_name>)))`
attribute for the WebAssembly target. This attribute may be attached to a
function declaration, where it modifies how the symbol is to be imported
within the WebAssembly linking environment.

WebAssembly imports use a two-level namespace scheme, consisting of a module
name, which typically identifies a module from which to import, and a field
name, which typically identifies a field from that module to import. By
default, module names for C/C++ symbols are assigned automatically by the
linker. This attribute can be used to override the default behavior, and
request a specific module name be used instead.
  }];
}

def WebAssemblyImportNameDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Clang supports the `__attribute__((import_name(<name>)))`
attribute for the WebAssembly target. This attribute may be attached to a
function declaration, where it modifies how the symbol is to be imported
within the WebAssembly linking environment.

WebAssembly imports use a two-level namespace scheme, consisting of a module
name, which typically identifies a module from which to import, and a field
name, which typically identifies a field from that module to import. By
default, field names for C/C++ symbols are the same as their C/C++ symbol
names. This attribute can be used to override the default behavior, and
request a specific field name be used instead.
  }];
}

def ArtificialDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `artificial` attribute can be applied to an inline function. If such a
function is inlined, the attribute indicates that debuggers should associate
the resulting instructions with the call site, rather than with the
corresponding line within the inlined callee.
  }];
}

def NoDerefDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `noderef` attribute causes clang to diagnose dereferences of annotated pointer types.
This is ideally used with pointers that point to special memory which cannot be read
from or written to, but allowing for the pointer to be used in pointer arithmetic.
The following are examples of valid expressions where dereferences are diagnosed:

```c
int __attribute__((noderef)) *p;
int x = *p;  // warning

int __attribute__((noderef)) **p2;
x = **p2;  // warning

int * __attribute__((noderef)) *p3;
p = *p3;  // warning

struct S {
  int a;
};
struct S __attribute__((noderef)) *s;
x = s->a;    // warning
x = (*s).a;  // warning
```

Not all dereferences may diagnose a warning if the value directed by the pointer may not be
accessed. The following are examples of valid expressions where may not be diagnosed:

```c
int *q;
int __attribute__((noderef)) *p;
q = &*p;
q = *&p;

struct S {
  int a;
};
struct S __attribute__((noderef)) *s;
p = &s->a;
p = &(*s).a;
```

`noderef` is currently only supported for pointers and arrays and not usable
for references or Objective-C object pointers.

```c++
int x = 2;
int __attribute__((noderef)) &y = x;  // warning: 'noderef' can only be used on an array or pointer type
```

```objc
id __attribute__((noderef)) obj = [NSObject new]; // warning: 'noderef' can only be used on an array or pointer type
```
}];
}

def CFIUncheckedCalleeDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
`cfi_unchecked_callee` is a function type attribute which prevents the
compiler from instrumenting
{doc}`Control Flow Integrity <ControlFlowIntegrity>` checks on indirect
function calls. This also includes control flow checks added by
`-fsanitize=function`; see {ref}`Available checks <ubsan-checks>`.
Specifically, the attribute has the following semantics:

1. Indirect calls to a function type with this attribute will not be instrumented with CFI. That is,
   the indirect call will not be checked. Note that this only changes the behavior for indirect calls
   on pointers to function types having this attribute. It does not prevent all indirect function calls
   for a given type from being checked.
2. All direct references to a function whose type has this attribute will always reference the
   function definition rather than an entry in the CFI jump table.
3. When a pointer to a function with this attribute is implicitly cast to a pointer to a function
   without this attribute, the compiler will give a warning saying this attribute is discarded. This
   warning can be silenced with an explicit cast. Note an explicit cast just disables the warning, so
   direct references to a function with a `cfi_unchecked_callee` attribute will still reference the
   function definition rather than the CFI jump table.

```c
#define CFI_UNCHECKED_CALLEE __attribute__((cfi_unchecked_callee))

void no_cfi() CFI_UNCHECKED_CALLEE {}

void (*with_cfi)() = no_cfi;  // warning: implicit conversion discards `cfi_unchecked_callee` attribute.
                              // `with_cfi` also points to the actual definition of `no_cfi` rather than
                              // its jump table entry.

void invoke(void (CFI_UNCHECKED_CALLEE *func)()) {
  func();  // CFI will not instrument this indirect call.

  void (*func2)() = func;  // warning: implicit conversion discards `cfi_unchecked_callee` attribute.

  func2();  // CFI will instrument this indirect call. Users should be careful however because if this
            // references a function with type `cfi_unchecked_callee`, then the CFI check may incorrectly
            // fail because the reference will be to the function definition rather than the CFI jump
            // table entry.
}
```

This attribute can only be applied on functions or member functions. This attribute can be a good
alternative to `no_sanitize("cfi")` if you only want to disable innstrumentation for specific indirect
calls rather than applying `no_sanitize("cfi")` on the whole function containing indirect call. Note
that `cfi_unchecked_attribute` is a type attribute doesn't disable CFI instrumentation on a function
body.
}];
}

def ReinitializesDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `reinitializes` attribute can be applied to a non-static, non-const C++
member function to indicate that this member function reinitializes the entire
object to a known state, independent of the previous state of the object.

This attribute can be interpreted by static analyzers that warn about uses of an
object that has been left in an indeterminate state by a move operation. If a
member function marked with the `reinitializes` attribute is called on a
moved-from object, the analyzer can conclude that the object is no longer in an
indeterminate state.

A typical example where this attribute would be used is on functions that clear
a container class:

```c++
template <class T>
class Container {
public:
  ...
  [[clang::reinitializes]] void Clear();
  ...
};
```
  }];
}

def AlwaysDestroyDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `always_destroy` attribute specifies that a variable with static or thread
storage duration should have its exit-time destructor run. This attribute is the
default unless clang was invoked with -fno-c++-static-destructors.

If a variable is explicitly declared with this attribute, Clang will silence
otherwise applicable `-Wexit-time-destructors` warnings.
  }];
}

def NoDestroyDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `no_destroy` attribute specifies that a variable with static or thread
storage duration shouldn't have its exit-time destructor run. Annotating every
static and thread duration variable with this attribute is equivalent to
invoking clang with -fno-c++-static-destructors.

If a variable is declared with this attribute, clang doesn't access check or
generate the type's destructor. If you have a type that you only want to be
annotated with `no_destroy`, you can therefore declare the destructor private:

```c++
struct only_no_destroy {
  only_no_destroy();
private:
  ~only_no_destroy();
};

[[clang::no_destroy]] only_no_destroy global; // fine!
```

Note that destructors are still required for subobjects of aggregates annotated
with this attribute. This is because previously constructed subobjects need to
be destroyed if an exception gets thrown before the initialization of the
complete object is complete. For instance:

```c++
void f() {
  try {
    [[clang::no_destroy]]
    static only_no_destroy array[10]; // error, only_no_destroy has a private destructor.
  } catch (...) {
    // Handle the error
  }
}
```

Here, if the construction of `array[9]` fails with an exception, `array[0..8]`
will be destroyed, so the element's destructor needs to be accessible.
  }];
}

def UninitializedDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The command-line parameter `-ftrivial-auto-var-init=*` can be used to
initialize trivial automatic stack variables. By default, trivial automatic
stack variables are uninitialized. This attribute is used to override the
command-line parameter, forcing variables to remain uninitialized. It has no
semantic meaning in that using uninitialized values is undefined behavior,
it rather documents the programmer's intent.
  }];
}

def LoaderUninitializedDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `loader_uninitialized` attribute can be placed on global variables to
indicate that the variable does not need to be zero initialized by the loader.
On most targets, zero-initialization does not incur any additional cost.
For example, most general purpose operating systems deliberately ensure
that all memory is properly initialized in order to avoid leaking privileged
information from the kernel or other programs. However, some targets
do not make this guarantee, and on these targets, avoiding an unnecessary
zero-initialization can have a significant impact on load times and/or code
size.

A declaration with this attribute is a non-tentative definition just as if it
provided an initializer. Variables with this attribute are considered to be
uninitialized in the same sense as a local variable, and the programs must
write to them before reading from them. If the variable's type is a C++ class
type with a non-trivial default constructor, or an array thereof, this attribute
only suppresses the static zero-initialization of the variable, not the dynamic
initialization provided by executing the default constructor.
  }];
}

def CallbackDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `callback` attribute specifies that the annotated function may invoke the
specified callback zero or more times. The callback, as well as the passed
arguments, are identified by their parameter name or position (starting with
1!) in the annotated function. The first position in the attribute identifies
the callback callee, the following positions declare describe its arguments.
The callback callee is required to be callable with the number, and order, of
the specified arguments. The index `0`, or the identifier `this`, is used to
represent an implicit "this" pointer in class methods. If there is no implicit
"this" pointer it shall not be referenced. The index '-1', or the name "\_\_",
represents an unknown callback callee argument. This can be a value which is
not present in the declared parameter list, or one that is, but is potentially
inspected, captured, or modified. Parameter names and indices can be mixed in
the callback attribute.

The `callback` attribute, which is directly translated to `callback`
metadata \<<http://llvm.org/docs/LangRef.html#callback-metadata>>, make the
connection between the call to the annotated function and the callback callee.
This can enable interprocedural optimizations which were otherwise impossible.
If a function parameter is mentioned in the `callback` attribute, through its
position, it is undefined if that parameter is used for anything other than the
actual callback. Inspected, captured, or modified parameters shall not be
listed in the `callback` metadata.

Example encodings for the callback performed by `pthread_create` are shown
below. The explicit attribute annotation indicates that the third parameter
(`start_routine`) is called zero or more times by the `pthread_create` function,
and that the fourth parameter (`arg`) is passed along. Note that the callback
behavior of `pthread_create` is automatically recognized by Clang. In addition,
the declarations of `__kmpc_fork_teams` and `__kmpc_fork_call`, generated for
`#pragma omp target teams` and `#pragma omp parallel`, respectively, are also
automatically recognized as broker functions. Further functions might be added
in the future.

```c
__attribute__((callback (start_routine, arg)))
int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
                   void *(*start_routine) (void *), void *arg);

__attribute__((callback (3, 4)))
int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
                   void *(*start_routine) (void *), void *arg);
```

  }];
}

def CalledOnceDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `called_once` attribute specifies that the annotated function or method
parameter is invoked exactly once on all execution paths. It only applies
to parameters with function-like types, i.e. function pointers or blocks. This
concept is particularly useful for asynchronous programs.

Clang implements a check for `called_once` parameters,
`-Wcalled-once-parameter`. It is on by default and finds the following
violations:

- Parameter is not called at all.
- Parameter is called more than once.
- Parameter is not called on one of the execution paths.

In the latter case, Clang pinpoints the path where parameter is not invoked
by showing the control-flow statement where the path diverges.

```objc
void fooWithCallback(void (^callback)(void) __attribute__((called_once))) {
  if (somePredicate()) {
    ...
    callback();
  } else {
    callback(); // OK: callback is called on every path
  }
}

void barWithCallback(void (^callback)(void) __attribute__((called_once))) {
  if (somePredicate()) {
    ...
    callback(); // note: previous call is here
  }
  callback(); // warning: callback is called twice
}

void foobarWithCallback(void (^callback)(void) __attribute__((called_once))) {
  if (somePredicate()) {  // warning: callback is not called when condition is false
    ...
    callback();
  }
}
```

This attribute is useful for API developers who want to double-check if they
implemented their method correctly.

  }];
}

def GnuInlineDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `gnu_inline` changes the meaning of `extern inline` to use GNU inline
semantics, meaning:

- If any declaration that is declared `inline` is not declared `extern`,
  then the `inline` keyword is just a hint. In particular, an out-of-line
  definition is still emitted for a function with external linkage, even if all
  call sites are inlined, unlike in C99 and C++ inline semantics.
- If all declarations that are declared `inline` are also declared
  `extern`, then the function body is present only for inlining and no
  out-of-line version is emitted.

Some important consequences: `static inline` emits an out-of-line
version if needed, a plain `inline` definition emits an out-of-line version
always, and an `extern inline` definition (in a header) followed by a
(non-`extern`) `inline` declaration in a source file emits an out-of-line
version of the function in that source file but provides the function body for
inlining to all includers of the header.

Either `__GNUC_GNU_INLINE__` (GNU inline semantics) or
`__GNUC_STDC_INLINE__` (C99 semantics) will be defined (they are mutually
exclusive). If `__GNUC_STDC_INLINE__` is defined, then the `gnu_inline`
function attribute can be used to get GNU inline semantics on a per function
basis. If `__GNUC_GNU_INLINE__` is defined, then the translation unit is
already being compiled with GNU inline semantics as the implied default. It is
unspecified which macro is defined in a C++ compilation.

GNU inline semantics are the default behavior with `-std=gnu89`,
`-std=c89`, `-fgnu89-inline`, or `-std=iso9899:199409`.
  }];
}

def SpeculativeLoadHardeningDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute can be applied to a function declaration in order to indicate
that [Speculative Load Hardening][slh]
should be enabled for the function body. This can also be applied to a method
in Objective C. This attribute will take precedence over the command line flag
in the case where {option}`-mno-speculative-load-hardening` is specified.

[slh]: https://llvm.org/docs/SpeculativeLoadHardening.html

Speculative Load Hardening is a best-effort mitigation against
information leak attacks that make use of control flow
miss-speculation - specifically miss-speculation of whether a branch
is taken or not. Typically vulnerabilities enabling such attacks are
classified as "Spectre variant #1". Notably, this does not attempt to
mitigate against miss-speculation of branch target, classified as
"Spectre variant #2" vulnerabilities.

When inlining, the attribute is sticky. Inlining a function that
carries this attribute will cause the caller to gain the
attribute. This is intended to provide a maximally conservative model
where the code in a function annotated with this attribute will always
(even after inlining) end up hardened.
  }];
}

def NoSpeculativeLoadHardeningDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute can be applied to a function declaration in order to indicate
that [Speculative Load Hardening][slh] is *not* needed for the function body.
This can also be applied to a method in Objective C. This attribute will take
precedence over the command line flag in the case where
{option}`-mspeculative-load-hardening` is specified.

Warning: This attribute may not prevent Speculative Load Hardening from being
enabled for a function which inlines a function that has the
'speculative_load_hardening' attribute. This is intended to provide a
maximally conservative model where the code that is marked with the
'speculative_load_hardening' attribute will always (even when inlined)
be hardened. A user of this attribute may want to mark functions called by
a function they do not want to be hardened with the 'noinline' attribute.

For example:

```c
__attribute__((speculative_load_hardening))
int foo(int i) {
  return i;
}

// Note: bar() may still have speculative load hardening enabled if
// foo() is inlined into bar(). Mark foo() with __attribute__((noinline))
// to avoid this situation.
__attribute__((no_speculative_load_hardening))
int bar(int i) {
  return foo(i);
}
```
  }];
}

def ObjCExternallyRetainedDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `objc_externally_retained` attribute can be applied to strong local
variables, functions, methods, or blocks to opt into
{ref}`externally-retained semantics <arc.misc.externally_retained>`.

When applied to the definition of a function, method, or block, every parameter
of the function with implicit strong retainable object pointer type is
considered externally-retained, and becomes `const`. By explicitly annotating
a parameter with `__strong`, you can opt back into the default
non-externally-retained behavior for that parameter. For instance,
`first_param` is externally-retained below, but not `second_param`:

```objc
__attribute__((objc_externally_retained))
void f(NSArray *first_param, __strong NSArray *second_param) {
  // ...
}
```

Likewise, when applied to a strong local variable, that variable becomes
`const` and is considered externally-retained.

When compiled without `-fobjc-arc`, this attribute is ignored.
}]; }

def MIGConventionDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The Mach Interface Generator release-on-success convention dictates

functions that follow it to only release arguments passed to them when they
return "success" (a `kern_return_t` error code that indicates that
no errors have occurred). Otherwise the release is performed by the MIG client
that called the function. The annotation `__attribute__((mig_server_routine))`
is applied in order to specify which functions are expected to follow the
convention. This allows the Static Analyzer to find bugs caused by violations of
that convention. The attribute would normally appear on the forward declaration
of the actual server routine in the MIG server header, but it may also be
added to arbitrary functions that need to follow the same convention - for
example, a user can add them to auxiliary functions called by the server routine
that have their return value of type `kern_return_t` unconditionally returned
from the routine. The attribute can be applied to C++ methods, and in this case
it will be automatically applied to overrides if the method is virtual. The
attribute can also be written using C++11 syntax: `[[mig::server_routine]]`.
}];
}

def MinSizeDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This function attribute indicates that optimization passes and code generator passes
make choices that keep the function code size as small as possible. Optimizations may
also sacrifice runtime performance in order to minimize the size of the generated code.
  }];
}

def MSAllocatorDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `__declspec(allocator)` attribute is applied to functions that allocate
memory, such as operator new in C++. When CodeView debug information is emitted
(enabled by `clang -gcodeview` or `clang-cl /Z7`), Clang will attempt to
record the code offset of heap allocation call sites in the debug info. It will
also record the type being allocated using some local heuristics. The Visual
Studio debugger uses this information to [profile memory usage][profile memory usage].

This attribute does not affect optimizations in any way, unlike GCC's
`__attribute__((malloc))`.

[profile memory usage]: https://docs.microsoft.com/en-us/visualstudio/profiling/memory-usage
}];
}

def CFGuardDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Code can indicate CFG checks are not wanted with the `__declspec(guard(nocf))`
attribute. This directs the compiler to not insert any CFG checks for the entire
function. This approach is typically used only sparingly in specific situations
where the programmer has manually inserted "CFG-equivalent" protection. The
programmer knows that they are calling through some read-only function table
whose address is obtained through read-only memory references and for which the
index is masked to the function table limit. This approach may also be applied
to small wrapper functions that are not inlined and that do nothing more than
make a call through a function pointer. Since incorrect usage of this directive
can compromise the security of CFG, the programmer must be very careful using
the directive. Typically, this usage is limited to very small functions that
only call one function.

`Control Flow Guard documentation <https://docs.microsoft.com/en-us/windows/win32/secbp/pe-metadata>`
}];
}

def CUDADeviceBuiltinSurfaceTypeDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `device_builtin_surface_type` attribute can be applied to a class
template when declaring the surface reference. A surface reference variable
could be accessed on the host side and, on the device side, might be translated
into an internal surface object, which is established through surface bind and
unbind runtime APIs.
  }];
}

def CUDADeviceBuiltinTextureTypeDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `device_builtin_texture_type` attribute can be applied to a class
template when declaring the texture reference. A texture reference variable
could be accessed on the host side and, on the device side, might be translated
into an internal texture object, which is established through texture bind and
unbind runtime APIs.
  }];
}

def CUDAGridConstantAttrDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `__grid_constant__` attribute can be applied to a `const`-qualified kernel
function argument and allows compiler to take the address of that argument without
making a copy. The argument applies to sm_70 or newer GPUs, during compilation
with CUDA-11.7(PTX 7.7) or newer, and is ignored otherwise.
  }];
}

def HIPManagedAttrDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `__managed__` attribute can be applied to a global variable declaration in HIP.
A managed variable is emitted as an undefined global symbol in the device binary and is
registered by `__hipRegisterManagedVar` in init functions. The HIP runtime allocates
managed memory and uses it to define the symbol when loading the device binary.
A managed variable can be accessed in both device and host code.
  }];
}

def CUDAClusterDimsAttrDoc : Documentation {
  let Category = DocCatDecl;
  let Content = [{
In CUDA/HIP programming, the `cluster_dims` attribute, conventionally exposed as the
`__cluster_dims__` macro, can be applied to a kernel function to set the dimensions of a
thread block cluster, which is an optional level of hierarchy and made up of thread blocks.
`__cluster_dims__` defines the cluster size as `(X, Y, Z)`, where each value is the number
of thread blocks in that dimension. The `cluster_dims` and `no_cluster` attributes are
mutually exclusive.

```
__global__ __cluster_dims__(2, 1, 1) void kernel(...) {
  ...
}
```

  }];
}

def CUDANoClusterAttrDoc : Documentation {
  let Category = DocCatDecl;
  let Content = [{
In CUDA/HIP programming, a kernel function can still be launched with the cluster feature enabled
at runtime, even without being annotated with `__cluster_dims__`. The LLVM/Clang-exclusive
`no_cluster` attribute, conventionally exposed as the `__no_cluster__` macro, can be applied to
a kernel function to explicitly indicate that the cluster feature will not be enabled either at
compile time or at kernel launch time. This allows the compiler to apply certain optimizations
without assuming that clustering could be enabled at runtime. It is undefined behavior to launch a
kernel annotated with `__no_cluster__` if the cluster feature is enabled at runtime.
The `cluster_dims` and `no_cluster` attributes are mutually exclusive.

```
__global__ __no_cluster__ void kernel(...) {
  ...
}
```
  }];
}

def LifetimeOwnerDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
:::{Note}
This attribute is experimental and its effect on analysis is subject to change in
a future version of clang.
:::

The attribute `[[gsl::Owner(T)]]` applies to structs and classes that own an
object of type `T`:

```
class [[gsl::Owner(int)]] IntOwner {
private:
  int value;
public:
  int *getInt() { return &value; }
};
```

The argument `T` is optional and is ignored.
This attribute may be used by analysis tools and has no effect on code
generation. A `void` argument means that the class can own any type.

See [Pointer] for an example.
}];
}

def LifetimePointerDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
:::{Note}
This attribute is experimental and its effect on analysis is subject to change in
a future version of clang.
:::

The attribute `[[gsl::Pointer(T)]]` applies to structs and classes that behave
like pointers to an object of type `T`:

```
class [[gsl::Pointer(int)]] IntPointer {
private:
  int *valuePointer;
public:
  IntPointer(const IntOwner&);
  int *getInt() { return valuePointer; }
};
```

The argument `T` is optional and is ignored.
This attribute may be used by analysis tools and has no effect on code
generation. A `void` argument means that the pointer can point to any type.

Example:
When constructing an instance of a class annotated like this (a Pointer) from
an instance of a class annotated with `[[gsl::Owner]]` (an Owner),
then the analysis will consider the Pointer to point inside the Owner.
When the Owner's lifetime ends, it will consider the Pointer to be dangling.

```c++
int f() {
  IntPointer P(IntOwner{}); // P "points into" a temporary IntOwner object
  P.getInt(); // P is dangling
}
```

**Transparent Member Functions**

The analysis automatically tracks certain member functions of `[[gsl::Pointer]]` types
that provide transparent access to the pointed-to object. These include:

- Dereference operators: `operator*`, `operator->`
- Data access methods: `data()`, `c_str()`, `get()`
- Iterator operations: `begin()`, `end()`, `rbegin()`, `rend()`, `cbegin()`, `cend()`, `crbegin()`, `crend()`, `operator+`, `operator-`, `operator++`, `operator--`

When these methods return pointers, view types, or references, the analysis treats them as
transparently borrowing from the same object that the pointer itself borrows from,
enabling detection of use-after-free through these access patterns:

```c++
// For example, .data() here returns a borrow to 's' instead of 'v'.
const char* f() {
  std::string s = "hello";
  std::string_view v = s; // warning: address of stack memory returned
  return v.data();        // note: returned here
}

const MyObj& g(MyObj obj) {
  View v = obj; // warning: address of stack memory returned
  return *v;    // note: returned here
}
```

This tracking also applies to range-based for loops, where the `begin()` and `end()`
iterators are used to access elements:

```c++
std::string_view f(std::vector<std::string> vec) {
  for (const std::string& s : vec) {  // warning: address of stack memory returned
    return s; // note: returned here
  }
}
```

**Container Template Specialization**

If a template class is annotated with `[[gsl::Owner]]`, and the first
instantiated template argument is a pointer type (raw pointer, or `[[gsl::Pointer]]`),
the analysis will consider the instantiated class as a container of the pointer.
When constructing such an object from a GSL owner object, the analysis will
assume that the container holds a pointer to the owner object. Consequently,
when the owner object is destroyed, the pointer will be considered dangling.

```c++
int f() {
  std::vector<std::string_view> v = {std::string()}; // v holds a dangling pointer.
  std::optional<std::string_view> o = std::string(); // o holds a dangling pointer.
}
```

}];
}

def ArmBuiltinAliasDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute is used in the implementation of the ACLE intrinsics.
It allows the intrinsic functions to
be declared using the names defined in ACLE, and still be recognized
as clang builtins equivalent to the underlying name. For example,
`arm_mve.h` declares the function `vaddq_u32` with
`__attribute__((__clang_arm_mve_alias(__builtin_arm_mve_vaddq_u32)))`,
and similarly, one of the type-overloaded declarations of `vaddq`
will have the same attribute. This ensures that both functions are
recognized as that clang builtin, and in the latter case, the choice
of which builtin to identify the function as can be deferred until
after overload resolution.

This attribute can only be used to set up the aliases for certain Arm
intrinsic functions; it is intended for use only inside `arm_*.h`
and is not a general mechanism for declaring arbitrary aliases for
clang builtin functions.

In order to avoid duplicating the attribute definitions for similar
purpose for other architecture, there is a general form for the
attribute `clang_builtin_alias`.
  }];
}

def NoBuiltinDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `__attribute__((no_builtin))` is similar to the `-fno-builtin` flag
except it is specific to the body of a function. The attribute may also be
applied to a virtual function but has no effect on the behavior of overriding
functions in a derived class.

It accepts one or more strings corresponding to the specific names of the
builtins to disable (e.g. "memcpy", "memset").
If the attribute is used without parameters it will disable all buitins at
once.

```c++
// The compiler is not allowed to add any builtin to foo's body.
void foo(char* data, size_t count) __attribute__((no_builtin)) {
  // The compiler is not allowed to convert the loop into
  // `__builtin_memset(data, 0xFE, count);`.
  for (size_t i = 0; i < count; ++i)
    data[i] = 0xFE;
}

// The compiler is not allowed to add the `memcpy` builtin to bar's body.
void bar(char* data, size_t count) __attribute__((no_builtin("memcpy"))) {
  // The compiler is allowed to convert the loop into
  // `__builtin_memset(data, 0xFE, count);` but cannot generate any
  // `__builtin_memcpy`
  for (size_t i = 0; i < count; ++i)
    data[i] = 0xFE;
}
```
  }];
}

def UsingIfExistsDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `using_if_exists` attribute applies to a using-declaration. It allows
programmers to import a declaration that potentially does not exist, instead
deferring any errors to the point of use. For instance:

```c++
namespace empty_namespace {};
__attribute__((using_if_exists))
using empty_namespace::does_not_exist; // no error!

does_not_exist x; // error: use of unresolved 'using_if_exists'
```

The C++ spelling of the attribute (`[[clang::using_if_exists]]`) is also
supported as a clang extension, since ISO C++ doesn't support attributes in this
position. If the entity referred to by the using-declaration is found by name
lookup, the attribute has no effect. This attribute is useful for libraries
(primarily, libc++) that wish to redeclare a set of declarations in another
namespace, when the availability of those declarations is difficult or
impossible to detect at compile time with the preprocessor.
  }];
}

def HandleDocs : DocumentationCategory<"Handle Attributes"> {
  let Content = [{
Handles are a way to identify resources like files, sockets, and processes.
They are more opaque than pointers and widely used in system programming. They
have similar risks such as never releasing a resource associated with a handle,
attempting to use a handle that was already released, or trying to release a
handle twice. Using the annotations below it is possible to make the ownership
of the handles clear: whose responsibility is to release them. They can also
aid static analysis tools to find bugs.
  }];
}

def AcquireHandleDocs : Documentation {
  let Category = HandleDocs;
  let Content = [{
If this annotation is on a function or a function type it is assumed to return
a new handle. In case this annotation is on an output parameter,
the function is assumed to fill the corresponding argument with a new
handle. The attribute requires a string literal argument which used to
identify the handle with later uses of `use_handle` or
`release_handle`.

```c++
// Output arguments from Zircon.
zx_status_t zx_socket_create(uint32_t options,
                             zx_handle_t __attribute__((acquire_handle("zircon"))) * out0,
                             zx_handle_t* out1 [[clang::acquire_handle("zircon")]]);


// Returned handle.
[[clang::acquire_handle("tag")]] int open(const char *path, int oflag, ... );
int open(const char *path, int oflag, ... ) __attribute__((acquire_handle("tag")));
```
  }];
}

def UseHandleDocs : Documentation {
  let Category = HandleDocs;
  let Content = [{
A function taking a handle by value might close the handle. If a function
parameter is annotated with `use_handle(tag)` it is assumed to not to change
the state of the handle. It is also assumed to require an open handle to work with.
The attribute requires a string literal argument to identify the handle being used.

```c++
zx_status_t zx_port_wait(zx_handle_t handle [[clang::use_handle("zircon")]],
                         zx_time_t deadline,
                         zx_port_packet_t* packet);
```
  }];
}

def ReleaseHandleDocs : Documentation {
  let Category = HandleDocs;
  let Content = [{
If a function parameter is annotated with `release_handle(tag)` it is assumed to
close the handle. It is also assumed to require an open handle to work with. The
attribute requires a string literal argument to identify the handle being released.

```c++
zx_status_t zx_handle_close(zx_handle_t handle [[clang::release_handle("tag")]]);
```
  }];
}

def UnsafeBufferUsageDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The attribute `[[clang::unsafe_buffer_usage]]` should be placed on functions
that need to be avoided as they are prone to buffer overflows or unsafe buffer
struct fields. It is designed to work together with the off-by-default compiler
warning `-Wunsafe-buffer-usage` to help codebases transition away from raw pointer
based buffer management, in favor of safer abstractions such as C++20 `std::span`.
The attribute causes `-Wunsafe-buffer-usage` to warn on every use of the function or
the field it is attached to, and it may also lead to emission of automatic fix-it
hints which would help the user replace the use of unsafe functions(/fields) with safe
alternatives, though the attribute can be used even when the fix can't be automated.

- Attribute attached to functions: The attribute suppresses all
  `-Wunsafe-buffer-usage` warnings within the function it is attached to, as the
  function is now classified as unsafe. The attribute should be used carefully, as it
  will silence all unsafe operation warnings inside the function; including any new
  unsafe operations introduced in the future.

  The attribute is warranted even if the only way a function can overflow
  the buffer is by violating the function's preconditions. For example, it
  would make sense to put the attribute on function `foo()` below because
  passing an incorrect size parameter would cause a buffer overflow:

  ```c++
  [[clang::unsafe_buffer_usage]]
  void foo(int *buf, size_t size) {
    for (size_t i = 0; i < size; ++i) {
      buf[i] = i;
    }
  }
  ```

  The attribute is NOT warranted when the function uses safe abstractions,
  assuming that these abstractions weren't misused outside the function.
  For example, function `bar()` below doesn't need the attribute,
  because assuming that the container `buf` is well-formed (has size that
  fits the original buffer it refers to), overflow cannot occur:

  ```c++
  void bar(std::span<int> buf) {
    for (size_t i = 0; i < buf.size(); ++i) {
      buf[i] = i;
    }
  }
  ```

  In this case function `bar()` enables the user to keep the buffer
  "containerized" in a span for as long as possible. On the other hand,
  Function `foo()` in the previous example may have internal
  consistency, but by accepting a raw buffer it requires the user to unwrap
  their span, which is undesirable according to the programming model
  behind `-Wunsafe-buffer-usage`.

  The attribute is warranted when a function accepts a raw buffer only to
  immediately put it into a span:

  ```c++
  [[clang::unsafe_buffer_usage]]
  void baz(int *buf, size_t size) {
    std::span<int> sp{ buf, size };
    for (size_t i = 0; i < sp.size(); ++i) {
      sp[i] = i;
    }
  }
  ```

  In this case `baz()` does not contain any unsafe operations, but the awkward
  parameter type causes the caller to unwrap the span unnecessarily.
  Note that regardless of the attribute, code inside `baz()` isn't flagged
  by `-Wunsafe-buffer-usage` as unsafe. It is definitely undesirable,
  but if `baz()` is on an API surface, there is no way to improve it
  to make it as safe as `bar()` without breaking the source and binary
  compatibility with existing users of the function. In such cases
  the proper solution would be to create a different function (possibly
  an overload of `baz()`) that accepts a safe container like `bar()`,
  and then use the attribute on the original `baz()` to help the users
  update their code to use the new function.

- Attribute attached to fields: The attribute should only be attached to
  struct fields, if the fields can not be updated to a safe type with bounds
  check, such as std::span. In other words, the buffers prone to unsafe accesses
  should always be updated to use safe containers/views and attaching the attribute
  must be last resort when such an update is infeasible.

  The attribute can be placed on individual fields or a set of them as shown below.

  ```c++
  struct A {
    [[clang::unsafe_buffer_usage]]
    int *ptr1;

    [[clang::unsafe_buffer_usage]]
    int *ptr2, buf[10];

    [[clang::unsafe_buffer_usage]]
    size_t sz;
  };
  ```

  Here, every read/write to the fields ptr1, ptr2, buf and sz will trigger a warning
  that the field has been explcitly marked as unsafe due to unsafe-buffer operations.

  }];
}

def DiagnoseAsBuiltinDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `diagnose_as_builtin` attribute indicates that Fortify diagnostics are to
be applied to the declared function as if it were the function specified by the
attribute. The builtin function whose diagnostics are to be mimicked should be
given. In addition, the order in which arguments should be applied must also
be given.

For example, the attribute can be used as follows.

```c
__attribute__((diagnose_as_builtin(__builtin_memset, 3, 2, 1)))
void *mymemset(int n, int c, void *s) {
  // ...
}
```

This indicates that calls to `mymemset` should be diagnosed as if they were
calls to `__builtin_memset`. The arguments `3, 2, 1` indicate by index the
order in which arguments of `mymemset` should be applied to
`__builtin_memset`. The third argument should be applied first, then the
second, and then the first. Thus (when Fortify warnings are enabled) the call
`mymemset(n, c, s)` will diagnose overflows as if it were the call
`__builtin_memset(s, c, n)`.

For variadic functions, the variadic arguments must come in the same order as
they would to the builtin function, after all normal arguments. For instance,
to diagnose a new function as if it were `sscanf`, we can use the attribute as
follows.

```c
__attribute__((diagnose_as_builtin(sscanf, 1, 2)))
int mysscanf(const char *str, const char *format, ...)  {
  // ...
}
```

Then the call `mysscanf("abc def", "%4s %4s", buf1, buf2)` will be diagnosed as
if it were the call `sscanf("abc def", "%4s %4s", buf1, buf2)`.

This attribute cannot be applied to non-static member functions.
}];
}

def ArmSveVectorBitsDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `arm_sve_vector_bits(N)` attribute is defined by the Arm C Language
Extensions (ACLE) for SVE. It is used to define fixed-length (VLST) variants of
sizeless types (VLAT).

For example:

```c
#include <arm_sve.h>

#if __ARM_FEATURE_SVE_BITS==512
typedef svint32_t fixed_svint32_t __attribute__((arm_sve_vector_bits(512)));
#endif
```

Creates a type `fixed_svint32_t` that is a fixed-length variant of
`svint32_t` that contains exactly 512-bits. Unlike `svint32_t`, this type
can be used in globals, structs, unions, and arrays, all of which are
unsupported for sizeless types.

The attribute can be attached to a single SVE vector (such as `svint32_t`) or
to the SVE predicate type `svbool_t`, this excludes tuple types such as
`svint32x4_t`. The behavior of the attribute is undefined unless
`N==__ARM_FEATURE_SVE_BITS`, the implementation defined feature macro that is
enabled under the `-msve-vector-bits` flag.

For more information See [Arm C Language Extensions for SVE](https://developer.arm.com/documentation/100987/latest) for more information.
}];
}

def ArmMveStrictPolymorphismDocs : Documentation {
    let Category = DocCatType;
    let Content = [{
This attribute is used in the implementation of the ACLE intrinsics for the Arm
MVE instruction set. It is used to define the vector types used by the MVE
intrinsics.

Its effect is to modify the behavior of a vector type with respect to function
overloading. If a candidate function for overload resolution has a parameter
type with this attribute, then the selection of that candidate function will be
disallowed if the actual argument can only be converted via a lax vector
conversion. The aim is to prevent spurious ambiguity in ARM MVE polymorphic
intrinsics.

```c++
void overloaded(uint16x8_t vector, uint16_t scalar);
void overloaded(int32x4_t vector, int32_t scalar);
uint16x8_t myVector;
uint16_t myScalar;

// myScalar is promoted to int32_t as a side effect of the addition,
// so if lax vector conversions are considered for myVector, then
// the two overloads are equally good (one argument conversion
// each). But if the vector has the __clang_arm_mve_strict_polymorphism
// attribute, only the uint16x8_t,uint16_t overload will match.
overloaded(myVector, myScalar + 1);
```

However, this attribute does not prohibit lax vector conversions in contexts
other than overloading.

```c++
uint16x8_t function();

// This is still permitted with lax vector conversion enabled, even
// if the vector types have __clang_arm_mve_strict_polymorphism
int32x4_t result = function();
```

    }];
}

def ArmCmseNSCallDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
This attribute declares a non-secure function type. When compiling for secure
state, a call to such a function would switch from secure to non-secure state.
All non-secure function calls must happen only through a function pointer, and
a non-secure function type should only be used as a base type of a pointer.
See [ARMv8-M Security Extensions: Requirements on Development
Tools - Engineering Specification Documentation](https://developer.arm.com/docs/ecm0359818/latest/) for more information.
  }];
}

def ArmCmseNSEntryDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute declares a function that can be called from non-secure state, or
from secure state. Entering from and returning to non-secure state would switch
to and from secure state, respectively, and prevent flow of information
to non-secure state, except via return values. See [ARMv8-M Security Extensions:
Requirements on Development Tools - Engineering Specification Documentation](https://developer.arm.com/docs/ecm0359818/latest/) for more information.
  }];
}

def DocCatArmSmeAttributes : DocumentationCategory<"AArch64 SME Attributes"> {
  let Content = [{
Clang supports a number of AArch64-specific attributes to manage state
added by the Scalable Matrix Extension (SME). This state includes the
runtime mode that the processor is in (e.g. non-streaming or streaming)
as well as the state of the `ZA` Matrix Storage.

The attributes come in the form of type- and declaration attributes:

- The SME declaration attributes can appear anywhere that a standard
  `[[...]]` declaration attribute can appear.
- The SME type attributes apply only to prototyped functions and can appear
  anywhere that a standard `[[...]]` type attribute can appear. The SME
  type attributes do not apply to functions having a K&R-style
  unprototyped function type.

See [Arm C Language Extensions](https://github.com/ARM-software/acle)
for more details about the features related to the SME extension.

See [Procedure Call Standard for the Arm® 64-bit Architecture (AArch64)](https://github.com/ARM-software/abi-aa) for more details about
streaming-interface functions and shared/private-ZA interface functions.
  }];
}

def ArmSmeStreamingDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_streaming` keyword applies to prototyped function types and specifies
that the function has a "streaming interface". This means that:

- the function requires that the processor implements the Scalable Matrix
  Extension (SME).
- the function must be entered in streaming mode (that is, with PSTATE.SM
  set to 1)
- the function must return in streaming mode

Clang manages PSTATE.SM automatically; it is not the source code's
responsibility to do this. For example, if a non-streaming
function calls an `__arm_streaming` function, Clang generates code
that switches into streaming mode before calling the function and
switches back to non-streaming mode on return.
  }];
}

def ArmSmeStreamingCompatibleDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_streaming_compatible` keyword applies to prototyped function types and
specifies that the function has a "streaming compatible interface". This
means that:

- the function may be entered in either non-streaming mode (PSTATE.SM=0) or
  in streaming mode (PSTATE.SM=1).
- the function must return in the same mode as it was entered.
- the code executed in the function is compatible with either mode.

Clang manages PSTATE.SM automatically; it is not the source code's
responsibility to do this. Clang will ensure that the generated code in
streaming-compatible functions is valid in either mode (PSTATE.SM=0 or
PSTATE.SM=1). For example, if an `__arm_streaming_compatible` function calls a
non-streaming function, Clang generates code to temporarily switch out of streaming
mode before calling the function and switch back to streaming-mode on return if
`PSTATE.SM` is `1` on entry of the caller. If `PSTATE.SM` is `0` on
entry to the `__arm_streaming_compatible` function, the call will be executed
without changing modes.
  }];
}

def ArmInDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_in` keyword applies to prototyped function types and specifies
that the function shares a given state S with its caller. For `__arm_in`, the
function takes the state S as input and returns with the state S unchanged.

The attribute takes string arguments to instruct the compiler which state
is shared. The supported states for S are:

- `"za"` for Matrix Storage (requires SME)

The attributes `__arm_in(S)`, `__arm_out(S)`, `__arm_inout(S)` and
`__arm_preserves(S)` are all mutually exclusive for the same state S.
  }];
}

def ArmOutDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_out` keyword applies to prototyped function types and specifies
that the function shares a given state S with its caller. For `__arm_out`,
the function ignores the incoming state for S and returns new state for S.

The attribute takes string arguments to instruct the compiler which state
is shared. The supported states for S are:

- `"za"` for Matrix Storage (requires SME)

The attributes `__arm_in(S)`, `__arm_out(S)`, `__arm_inout(S)` and
`__arm_preserves(S)` are all mutually exclusive for the same state S.
  }];
}

def ArmInOutDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_inout` keyword applies to prototyped function types and specifies
that the function shares a given state S with its caller. For `__arm_inout`,
the function takes the state S as input and returns new state for S.

The attribute takes string arguments to instruct the compiler which state
is shared. The supported states for S are:

- `"za"` for Matrix Storage (requires SME)

The attributes `__arm_in(S)`, `__arm_out(S)`, `__arm_inout(S)` and
`__arm_preserves(S)` are all mutually exclusive for the same state S.
  }];
}

def ArmPreservesDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_preserves` keyword applies to prototyped function types and
specifies that the function does not read a given state S and returns
with state S unchanged.

The attribute takes string arguments to instruct the compiler which state
is shared. The supported states for S are:

- `"za"` for Matrix Storage (requires SME)

The attributes `__arm_in(S)`, `__arm_out(S)`, `__arm_inout(S)` and
`__arm_preserves(S)` are all mutually exclusive for the same state S.
  }];
}

def ArmAgnosticDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_agnostic` keyword applies to prototyped function types and
affects the function's calling convention for a given state S. This
attribute allows the user to describe a function that preserves S, without
requiring the function to share S with its callers and without making
the assumption that S exists.

If a function has the `__arm_agnostic(S)` attribute and calls a function
without this attribute, then the function's object code will contain code
to preserve state S. Otherwise, the function's object code will be the same
as if it did not have the attribute.

The attribute takes string arguments to describe state S. The supported
states are:

- `"sme_za_state"` for state enabled by PSTATE.ZA, such as ZA and ZT0.

The attribute `__arm_agnostic("sme_za_state")` cannot be used in conjunction
with `__arm_in(S)`, `__arm_out(S)`, `__arm_inout(S)` or
`__arm_preserves(S)` where state S describes state enabled by PSTATE.ZA,
such as "za" or "zt0".
  }];
}

def ArmSmeLocallyStreamingDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_locally_streaming` keyword applies to function declarations
and specifies that all the statements in the function are executed in
streaming mode. This means that:

- the function requires that the target processor implements the Scalable Matrix
  Extension (SME).
- the program automatically puts the machine into streaming mode before
  executing the statements and automatically restores the previous mode
  afterwards.

Clang manages PSTATE.SM automatically; it is not the source code's
responsibility to do this. For example, Clang will emit code to enable
streaming mode at the start of the function, and disable streaming mode
at the end of the function.
  }];
}

def ArmNewDocs : Documentation {
  let Category = DocCatArmSmeAttributes;
  let Content = [{
The `__arm_new` keyword applies to function declarations and specifies
that the function will create a new scope for state S.

The attribute takes string arguments to instruct the compiler for which state
to create new scope. The supported states for S are:

- `"za"` for Matrix Storage (requires SME)

For state `"za"`, this means that:

- the function requires that the target processor implements the Scalable Matrix
  Extension (SME).
- the function will commit any lazily saved ZA data.
- the function will create a new ZA context and enable PSTATE.ZA.
- the function will disable PSTATE.ZA (by setting it to 0) before returning.

For `__arm_new("za")` functions Clang will set up the ZA context automatically
on entry to the function and disable it before returning. For example, if ZA is
in a dormant state Clang will generate the code to commit a lazy-save and set up
a new ZA state before executing user code.
  }];
}

def AlwaysInlineDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Inlining heuristics are disabled and inlining is always attempted regardless of
optimization level.

`[[clang::always_inline]]` spelling can be used as a statement attribute; other
spellings of the attribute are not supported on statements. If a statement is
marked `[[clang::always_inline]]` and contains calls, the compiler attempts
to inline those calls.

```c
int example(void) {
  int i;
  [[clang::always_inline]] foo(); // attempts to inline foo
  [[clang::always_inline]] i = bar(); // attempts to inline bar
  [[clang::always_inline]] return f(42, baz(bar())); // attempts to inline everything
}
```

A declaration statement, which is a statement, is not a statement that can have an
attribute associated with it (the attribute applies to the declaration, not the
statement in that case). So this use case will not work:

```c
int example(void) {
  [[clang::always_inline]] int i = bar();
  return i;
}
```

This attribute does not guarantee that inline substitution actually occurs.

\<ins>Note: applying this attribute to a coroutine at the `-O0` optimization level
has no effect; other optimization levels may only partially inline and result in a
diagnostic.\</ins>

See also [the Microsoft Docs on Inline Functions][the microsoft docs on inline functions], [the GCC Common Function
Attribute docs][the gcc common function attribute docs], and [the GCC Inline docs][the gcc inline docs].

[the gcc common function attribute docs]: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html
[the gcc inline docs]: https://gcc.gnu.org/onlinedocs/gcc/Inline.html
[the microsoft docs on inline functions]: https://docs.microsoft.com/en-us/cpp/cpp/inline-functions-cpp

}];
  let Heading = "always_inline, __force_inline";
}

def EnforceTCBDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `enforce_tcb` attribute can be placed on functions to enforce that a
trusted compute base (TCB) does not call out of the TCB. This generates a
warning every time a function not marked with an `enforce_tcb` attribute is
called from a function with the `enforce_tcb` attribute. A function may be a
part of multiple TCBs. Invocations through function pointers are currently
not checked. Builtins are considered to a part of every TCB.

- `enforce_tcb(Name)` indicates that this function is a part of the TCB named `Name`
  }];
}

def EnforceTCBLeafDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `enforce_tcb_leaf` attribute satisfies the requirement enforced by
`enforce_tcb` for the marked function to be in the named TCB but does not
continue to check the functions called from within the leaf function.

- `enforce_tcb_leaf(Name)` indicates that this function is a part of the TCB named `Name`
  }];
}

def ErrorAttrDocs : Documentation {
  let Category = DocCatFunction;
  let Heading = "error, warning";
  let Content = [{
The `error` and `warning` function attributes can be used to specify a
custom diagnostic to be emitted when a call to such a function is not
eliminated via optimizations. This can be used to create compile time
assertions that depend on optimizations, while providing diagnostics
pointing to precise locations of the call site in the source.

```c++
__attribute__((warning("oh no"))) void dontcall();
void foo() {
  if (someCompileTimeAssertionThatsTrue)
    dontcall(); // Warning

  dontcall(); // Warning

  if (someCompileTimeAssertionThatsFalse)
    dontcall(); // No Warning
  sizeof(dontcall()); // No Warning
}
```

When the call occurs through inlined functions, the
`-fdiagnostics-show-inlining-chain` option can be used to show the
inlining chain that led to the call. This helps identify which call site
triggered the diagnostic when the attributed function is called from
multiple locations through inline functions.

When enabled, this option automatically uses debug info for accurate source
locations if available (`-gline-directives-only` (implicitly enabled at
`-g1`) or higher), or falls back to a heuristic based on metadata tracking.
When falling back, a note is emitted suggesting `-gline-directives-only` for
more accurate locations.
  }];
}

def ZeroCallUsedRegsDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
This attribute, when attached to a function, causes the compiler to zero a
subset of all call-used registers before the function returns. It's used to
increase program security by either mitigating [Return-Oriented Programming][return-oriented programming]
(ROP) attacks or preventing information leakage through registers.

The term "call-used" means registers which are not guaranteed to be preserved
unchanged for the caller by the current calling convention. This could also be
described as "caller-saved" or "not callee-saved".

The `choice` parameters gives the programmer flexibility to choose the subset
of the call-used registers to be zeroed:

- `skip` doesn't zero any call-used registers. This choice overrides any
  command-line arguments.
- `used` only zeros call-used registers used in the function. By `used`, we
  mean a register whose contents have been set or referenced in the function.
- `used-gpr` only zeros call-used GPR registers used in the function.
- `used-arg` only zeros call-used registers used to pass arguments to the
  function.
- `used-gpr-arg` only zeros call-used GPR registers used to pass arguments to
  the function.
- `all` zeros all call-used registers.
- `all-gpr` zeros all call-used GPR registers.
- `all-arg` zeros all call-used registers used to pass arguments to the
  function.
- `all-gpr-arg` zeros all call-used GPR registers used to pass arguments to
  the function.

The default for the attribute is controlled by the `-fzero-call-used-regs`
flag.

[return-oriented programming]: https://en.wikipedia.org/wiki/Return-oriented_programming
  }];
}

def DocHLSLSemantics : DocumentationCategory<"HLSL Semantics"> {
  let Content = [{
A semantic is a string attached to a shader input or output that conveys
information about the intended use of a parameter. Semantics are required on
all variables passed between shader stages. The syntax for adding a semantic
to a shader variable is shown here (Variable Syntax (DirectX HLSL)).

In general, data passed between pipeline stages is completely generic and is
not uniquely interpreted by the system; arbitrary semantics are allowed which
have no special meaning. Parameters (in Direct3D 10 and later) which contain
these special semantics are referred to as System-Value Semantics.

More information is available here:
<https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-semantics>
  }];
}

def WaveSizeDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `WaveSize` attribute specify a wave size on a shader entry point in order
to indicate either that a shader depends on or strongly prefers a specific wave
size.
There're 2 versions of the attribute: `WaveSize` and `RangedWaveSize`.
The syntax for `WaveSize` is:

```text
``[WaveSize(<numLanes>)]``
```

The allowed wave sizes that an HLSL shader may specify are the powers of 2
between 4 and 128, inclusive.
In other words, the set: [4, 8, 16, 32, 64, 128].

The syntax for `RangedWaveSize` is:

```text
``[WaveSize(<minWaveSize>, <maxWaveSize>, [prefWaveSize])]``
```

Where minWaveSize is the minimum wave size supported by the shader representing
the beginning of the allowed range, maxWaveSize is the maximum wave size
supported by the shader representing the end of the allowed range, and
prefWaveSize is the optional preferred wave size representing the size expected
to be the most optimal for this shader.

`WaveSize` is available for HLSL shader model 6.6 and later.
`RangedWaveSize` available for HLSL shader model 6.8 and later.

The full documentation is available here: <https://microsoft.github.io/DirectX-Specs/d3d/HLSL_SM_6_6_WaveSize.html>
and <https://microsoft.github.io/hlsl-specs/proposals/0013-wave-size-range.html>
  }];
}

def VkConstantIdDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `vk::constant_id` attribute specifies the id for a SPIR-V specialization
constant. The attribute applies to const global scalar variables. The variable must be initialized with a C++11 constexpr.
In SPIR-V, the
variable will be replaced with an `OpSpecConstant` with the given id.
The syntax is:

```text
``[[vk::constant_id(<Id>)]] const T Name = <Init>``
```
}];
}

def RootSignatureDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `RootSignature` attribute applies to HLSL entry functions to define what
types of resources are bound to the graphics pipeline.

For details about the use and specification of Root Signatures please see here:
<https://learn.microsoft.com/en-us/windows/win32/direct3d12/root-signatures>
  }];
}

def NumThreadsDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `numthreads` attribute applies to HLSL shaders where explcit thread counts
are required. The `X`, `Y`, and `Z` values provided to the attribute
dictate the thread id. Total number of threads executed is `X * Y * Z`.

The full documentation is available here: <https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/sm5-attributes-numthreads>
  }];
}

def HLSLSV_ShaderTypeAttrDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `shader` type attribute applies to HLSL shader entry functions to
identify the shader type for the entry function.
The syntax is:

```text
``[shader(string-literal)]``
```

where the string literal is one of: "pixel", "vertex", "geometry", "hull",
"domain", "compute", "raygeneration", "intersection", "anyhit", "closesthit",
"miss", "callable", "mesh", "amplification". Normally the shader type is set
by shader target with the `-T` option like `-Tps_6_1`. When compiling to a
library target like `lib_6_3`, the shader type attribute can help the
compiler to identify the shader type. It is mostly used by Raytracing shaders
where shaders must be compiled into a library and linked at runtime.
  }];
}

def HLSLLoopHintDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "[loop]";
  let Content = [{
The `[loop]` directive allows loop optimization hints to be
specified for the subsequent loop. The directive allows unrolling to
be disabled and is not compatible with [unroll(x)].

Specifying the parameter, `[loop]`, directs the
unroller to not unroll the loop.

```hlsl
[loop]
for (...) {
  ...
}
```

```hlsl
[loop]
while (...) {
  ...
}
```

```hlsl
[loop]
do {
  ...
} while (...)
```

See [hlsl loop extensions](https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-for)
for details.
  }];
}

def HLSLUnrollHintDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "[unroll(x)], [unroll]";
  let Content = [{
Loop unrolling optimization hints can be specified with `[unroll(x)]`
. The attribute is placed immediately before a for, while,
or do-while.
Specifying the parameter, `[unroll(_value_)]`, directs the
unroller to unroll the loop `_value_` times. Note: [unroll(x)] is not compatible with [loop].

```hlsl
[unroll(4)]
for (...) {
  ...
}
```

```hlsl
[unroll]
for (...) {
  ...
}
```

```hlsl
[unroll(4)]
while (...) {
  ...
}
```

```hlsl
[unroll]
while (...) {
  ...
}
```

```hlsl
[unroll(4)]
do {
  ...
} while (...)
```

```hlsl
[unroll]
do {
  ...
} while (...)
```

See [hlsl loop extensions](https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-for)
for details.
  }];
}

def HLSLControlFlowHintDocs : Documentation {
  let Category = DocCatStmt;
  let Heading = "branch, flatten";
  let Content = [{

The ``branch`` and ``flatten`` attributes can be applied to *if* and *switch*
statements in the HLSL language mode to provide hints for how the backend
should execute them.

- ``branch`` means that control flow is preferred. The condition should be
  evaluated first and we should only execute the block guarded by it.

- ``flatten`` means that control flow should be avoided. All blocks should be
  executed and variables that are modified should be conditionally assigned.

These control flow hints are preserved through the compilation and emitted in a
backend-specific way.

For details, see the Direct3D documentation for `if Statement`_ and `switch Statement`_.

.. _`if Statement`: https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-if
.. _`switch Statement`: https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-switchhttps://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-switch
  }];
}

def AtomicDocs : Documentation {
  let Category = DocCatStmt;
  let Content = [{
The `atomic` attribute can be applied to *compound statements* to override or
further specify the default atomic code-generation behavior, especially on
targets such as AMDGPU. You can annotate compound statements with options
to modify how atomic instructions inside that statement are emitted at the IR
level.

For details, see the documentation for
{ref}`@atomic <langext-atomic-code-generation>`

  }];
}

def ClangRandomizeLayoutDocs : Documentation {
  let Category = DocCatDecl;
  let Heading = "randomize_layout, no_randomize_layout";
  let Content = [{
The attribute `randomize_layout`, when attached to a C structure, selects it
for structure layout field randomization; a compile-time hardening technique. A
"seed" value, is specified via the `-frandomize-layout-seed=` command line flag.
For example:

```bash
SEED=`od -A n -t x8 -N 32 /dev/urandom | tr -d ' \n'`
make ... CFLAGS="-frandomize-layout-seed=$SEED" ...
```

You can also supply the seed in a file with `-frandomize-layout-seed-file=`.
For example:

```bash
od -A n -t x8 -N 32 /dev/urandom | tr -d ' \n' > /tmp/seed_file.txt
make ... CFLAGS="-frandomize-layout-seed-file=/tmp/seed_file.txt" ...
```

The randomization is deterministic based for a given seed, so the entire
program should be compiled with the same seed, but keep the seed safe
otherwise.

The attribute `no_randomize_layout`, when attached to a C structure,
instructs the compiler that this structure should not have its field layout
randomized.
  }];
}

def HLSLResourceBindingDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The resource binding attribute sets the virtual register and logical register space for a resource.
Attribute spelling in HLSL is: `register(slot [, space])`.
`slot` takes the format `[type][number]`,
where `type` is a single character specifying the resource type and `number` is the virtual register number.

Register types are:
t for shader resource views (SRV),
s for samplers,
u for unordered access views (UAV),
b for constant buffer views (CBV).

Register space is specified in the format `space[number]` and defaults to `space0` if omitted.
Here're resource binding examples with and without space:

```hlsl
RWBuffer<float> Uav : register(u3, space1);
Buffer<float> Buf : register(t1);
```

The full documentation is available here: <https://docs.microsoft.com/en-us/windows/win32/direct3d12/resource-binding-in-hlsl>
  }];
}

def HLSLPackOffsetDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The packoffset attribute is used to change the layout of a cbuffer.
Attribute spelling in HLSL is: `packoffset( c[Subcomponent][.component] )`.
A subcomponent is a register number, which is an integer. A component is in the form of [.xyzw].

Examples:

```hlsl
cbuffer A {
  float3 a : packoffset(c0.y);
  float4 b : packoffset(c4);
}
```

The full documentation is available here: <https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-variable-packoffset>
  }];
}

def HLSLResourceClassDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::resource_class`` attribute specifies the resource class of the HLSL
resource represented by a member variable of type ``__hlsl_resource_t``,
declaring it to be an SRV, UAV, CBuffer, or Sampler resource.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLResourceDimensionDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::dimension`` attribute specifies the dimensions of the HLSL resource
represented by a member variable of type ``__hlsl_resource_t``, declaring the
resource to have Unknown, 1D, 2D, 3D, or Cube dimension.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLContainedTypeDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::contained_type`` attribute specifies the type of the HLSL resource
represented by a member variable of type ``__hlsl_resource_t``.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLRawBufferDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::raw_buffer`` attribute specifies that the HLSL resource represented
by a member variable of type ``__hlsl_resource_t`` has raw buffer semantics.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLIsArrayDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::is_array`` attribute specifies that the HLSL resource represented
by a member variable of type ``__hlsl_resource_t`` has array dimensions.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLIsMultiSampledDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::is_array`` attribute specifies that the HLSL resource represented
by a member variable of type ``__hlsl_resource_t`` is multisampled.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLIsROVDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::is_rov`` attribute specifies that the HLSL resource represented by
a member variable of type ``__hlsl_resource_t`` is a rasterizer ordered view.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLIsCounterDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The ``hlsl::is_counter`` attribute specifies that the HLSL resource represented
by a member variable of type ``__hlsl_resource_t`` is a counter buffer.

This attribute is only valid for resource handles, and is an implementation
detail of clang's HLSL implementation. For more information see `HLSL Resource
Types`_

.. _`HLSL Resource Types`: https://clang.llvm.org/docs/HLSL/ResourceTypes.html>
  }];
}

def HLSLGroupSharedAddressSpaceDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
HLSL enables threads of a compute shader to exchange values via shared memory.
HLSL provides barrier primitives such as GroupMemoryBarrierWithGroupSync,
and so on to ensure the correct ordering of reads and writes to shared memory
in the shader and to avoid data races.
Here's an example to declare a groupshared variable.

```c++
groupshared GSData data[5*5*1];
```

The full documentation is available here: <https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-variable-syntax#group-shared>
  }];
}

def HLSLParamQualifierDocs : Documentation {
  let Category = DocCatVariable;
  let Heading = "HLSL Parameter Modifiers";
  let Content = [{
HLSL function parameters are passed by value. Parameter declarations support
three qualifiers to denote parameter passing behavior. The three qualifiers are
`in`, `out` and `inout`.

Parameters annotated with `in` or with no annotation are passed by value from
the caller to the callee.

Parameters annotated with `out` are written to the argument after the callee
returns (Note: arguments values passed into `out` parameters *are not* copied
into the callee).

Parameters annotated with `inout` are copied into the callee via a temporary,
and copied back to the argument after the callee returns.
  }];
}

def HLSLVkExtBuiltinInputDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
Vulkan shaders have `Input` builtins. Those variables are externally
initialized by the driver/pipeline, but each copy is private to the current
lane.

Those builtins can be declared using the `[[vk::ext_builtin_input]]` attribute
like follows:

```c++
[[vk::ext_builtin_input(/* WorkgroupId */ 26)]]
static const uint3 groupid;
```

This variable will be lowered into a module-level variable, with the `Input`
storage class, and the `BuiltIn 26` decoration.

The full documentation for this inline SPIR-V attribute can be found here:
<https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md>
  }];
}

def HLSLVkExtBuiltinOutputDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
Vulkan shaders have `Output` builtins. Those variables are externally
visible to the driver/pipeline, but each copy is private to the current
lane.

Those builtins can be declared using the `[[vk::ext_builtin_output]]`
attribute like follows:

```c++
[[vk::ext_builtin_output(/* Position */ 0)]]
static float4 position;
```

This variable will be lowered into a module-level variable, with the `Output`
storage class, and the `BuiltIn 0` decoration.

The full documentation for this inline SPIR-V attribute can be found here:
<https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md>
  }];
}

def HLSLVkPushConstantDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
Vulkan shaders have `PushConstants`

The `[[vk::push_constant]]` attribute allows you to declare this
global variable as a push constant when targeting Vulkan.
This attribute is ignored otherwise.

This attribute must be applied to the variable, not underlying type.
The variable type must be a struct, per the requirements of Vulkan, "there
must be no more than one push constant block statically used per shader entry
point."
}];
}

def AnnotateDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `annotate` attribute is used to add annotations to declarations or statements,
typically for use by static analysis tools that are not integrated into the
core Clang compiler (e.g., Clang-Tidy checks or out-of-tree Clang-based tools).
It is a counterpart to the `annotate_type` attribute, which serves the same
purpose, but for types.

The attribute takes a mandatory string literal argument specifying the
annotation category and an arbitrary number of optional arguments that provide
additional information specific to the annotation category. The optional
arguments must be constant expressions of arbitrary type.

For example:

```c++
[[clang::annotate("category1", "foo", 1)]] void func(int val [[clang::annotate("category2")]]) {
  [[clang::annotate("category3")]] if (val) {

  }
}
```

  }];
}

def AnnotateTypeDocs : Documentation {
  let Category = DocCatType;
  let Heading = "annotate_type";
  let Content = [{
This attribute is used to add annotations to types, typically for use by static
analysis tools that are not integrated into the core Clang compiler (e.g.,
Clang-Tidy checks or out-of-tree Clang-based tools). It is a counterpart to the
`annotate` attribute, which serves the same purpose, but for declarations.

The attribute takes a mandatory string literal argument specifying the
annotation category and an arbitrary number of optional arguments that provide
additional information specific to the annotation category. The optional
arguments must be constant expressions of arbitrary type.

For example:

```c++
int* [[clang::annotate_type("category1", "foo", 1)]] f(int[[clang::annotate_type("category2")]] *);
```

The attribute does not have any effect on the semantics of the type system,
neither type checking rules, nor runtime semantics. In particular:

- `std::is_same<T, T [[clang::annotate_type("foo")]]>` is true for all types
  `T`.
- It is not permissible for overloaded functions or template specializations
  to differ merely by an `annotate_type` attribute.
- The presence of an `annotate_type` attribute will not affect name
  mangling.
  }];
}

def WeakDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
In supported output formats the `weak` attribute can be used to
specify that a variable or function should be emitted as a symbol with
`weak` (if a definition) or `extern_weak` (if a declaration of an
external symbol) [linkage](https://llvm.org/docs/LangRef.html#linkage-types).

If there is a non-weak definition of the symbol the linker will select
that over the weak. They must have same type and alignment (variables
must also have the same size), but may have a different value.

If there are multiple weak definitions of same symbol, but no non-weak
definition, they should have same type, size, alignment and value, the
linker will select one of them (see also [selectany] attribute).

If the `weak` attribute is applied to a `const` qualified variable
definition that variable is no longer consider a compiletime constant
as its value can change during linking (or dynamic linking). This
means that it can e.g no longer be part of an initializer expression.

```c
const int ANSWER __attribute__ ((weak)) = 42;

/* This function may be replaced link-time */
__attribute__ ((weak)) void debug_log(const char *msg)
{
    fprintf(stderr, "DEBUG: %s\n", msg);
}

int main(int argc, const char **argv)
{
    debug_log ("Starting up...");

    /* This may print something else than "6 * 7 = 42",
       if there is a non-weak definition of "ANSWER" in
       an object linked in */
    printf("6 * 7 = %d\n", ANSWER);

    return 0;
 }
```

If an external declaration is marked weak and that symbol does not
exist during linking (possibly dynamic) the address of the symbol will
evaluate to NULL.

```c
void may_not_exist(void) __attribute__ ((weak));

int main(int argc, const char **argv)
{
    if (may_not_exist) {
        may_not_exist();
    } else {
        printf("Function did not exist\n");
    }
    return 0;
}
```
  }];
}

def FunctionReturnThunksDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The attribute `function_return` can replace return instructions with jumps to
target-specific symbols. This attribute supports 2 possible values,
corresponding to the values supported by the `-mfunction-return=` command
line flag:

- `__attribute__((function_return("keep")))` to disable related transforms.
  This is useful for undoing global setting from `-mfunction-return=` locally
  for individual functions.
- `__attribute__((function_return("thunk-extern")))` to replace returns with
  jumps, while NOT emitting the thunk.

The values `thunk` and `thunk-inline` from GCC are not supported.

The symbol used for `thunk-extern` is target specific:
\* X86: `__x86_return_thunk`

As such, this function attribute is currently only supported on X86 targets.
  }];
}

def ReadOnlyPlacementDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
This attribute is attached to a structure, class or union declaration.

: When attached to a record declaration/definition, it checks if all instances
  of this type can be placed in the read-only data segment of the program. If it
  finds an instance that can not be placed in a read-only segment, the compiler
  emits a warning at the source location where the type was used.

  Examples:
  \* `struct __attribute__((enforce_read_only_placement)) Foo;`
  \* `struct __attribute__((enforce_read_only_placement)) Bar { ... };`

  Both `Foo` and `Bar` types have the `enforce_read_only_placement` attribute.

  The goal of introducing this attribute is to assist developers with writing secure
  code. A `const`-qualified global is generally placed in the read-only section
  of the memory that has additional run time protection from malicious writes. By
  attaching this attribute to a declaration, the developer can express the intent
  to place all instances of the annotated type in the read-only program memory.

  Note 1: The attribute doesn't guarantee that the object will be placed in the
  read-only data segment as it does not instruct the compiler to ensure such
  a placement. It emits a warning if something in the code can be proven to prevent
  an instance from being placed in the read-only data segment.

  Note 2: Currently, clang only checks if all global declarations of a given type 'T'
  are `const`-qualified. The following conditions would also prevent the data to be
  put into read only segment, but the corresponding warnings are not yet implemented.

  1. An instance of type `T` is allocated on the heap/stack.
  2. Type `T` defines/inherits a mutable field.
  3. Type `T` defines/inherits non-constexpr constructor(s) for initialization.
  4. A field of type `T` is defined by type `Q`, which does not bear the
     `enforce_read_only_placement` attribute.
  5. A type `Q` inherits from type `T` and it does not have the
     `enforce_read_only_placement` attribute.
  }];
}

def HLSLVkBindingDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
The `[[vk::binding]]` attribute allows you to explicitly specify the descriptor
set and binding for a resource when targeting SPIR-V. This is particularly
useful when you need different bindings for SPIR-V and DXIL, as the `register`
attribute can be used for DXIL-specific bindings.

The attribute takes two integer arguments: the binding and the descriptor set.
The descriptor set is optional and defaults to 0 if not provided.

```c++
// A structured buffer with binding 23 in descriptor set 102.
[[vk::binding(23, 102)]] StructuredBuffer<float> Buf;

// A structured buffer with binding 14 in descriptor set 0.
[[vk::binding(14)]] StructuredBuffer<float> Buf2;

// A cbuffer with binding 1 in descriptor set 2.
[[vk::binding(1, 2)]] cbuffer MyCBuffer {
  float4x4 worldViewProj;
};
```
  }];
}

def HLSLVkLocationDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
Attribute used for specifying the location number for the stage input/output
variables. Allowed on function parameters, function returns, and struct
fields. This parameter has no effect when used outside of an entrypoint
parameter/parameter field/return value.

This attribute maps to the 'Location' SPIR-V decoration.
  }];
}

def HLSLMatrixLayoutDocs : Documentation {
  let Category = DocCatVariable;
  let Heading = "row_major, column_major";
  let Content = [{
The `row_major` and `column_major` keywords specify the memory layout
of an HLSL matrix type.

- `row_major`: Matrices are stored in memory row-by-row.
- `column_major`: Matrices are stored in memory column-by-column (default).

Example:

```hlsl
row_major float2x2 myMatrix;
```
  }];
}

def WebAssemblyFuncrefDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
Clang supports the `__funcref` attribute for the WebAssembly target.
This attribute may be attached to a function pointer type, where it modifies
its underlying representation to be a WebAssembly `funcref`.
  }];
}

def PreferredTypeDocumentation : Documentation {
  let Category = DocCatField;
  let Label = "langext-preferred_type_documentation";
  let Content = [{
This attribute allows adjusting the type of a bit-field in debug information.
This can be helpful when a bit-field is intended to store an enumeration value,
but has to be specified as having the enumeration's underlying type in order to
facilitate compiler optimizations or bit-field packing behavior. Normally, the
underlying type is what is emitted in debug information, which can make it hard
for debuggers to know to map a bit-field's value back to a particular enumeration.

```c++
enum Colors { Red, Green, Blue };

struct S {
  [[clang::preferred_type(Colors)]] unsigned ColorVal : 2;
  [[clang::preferred_type(bool)]] unsigned UseAlternateColorSpace : 1;
} s = { Green, false };
```

Without the attribute, a debugger is likely to display the value `1` for `ColorVal`
and `0` for `UseAlternateColorSpace`. With the attribute, the debugger may now
display `Green` and `false` instead.

This can be used to map a bit-field to an arbitrary type that isn't integral
or an enumeration type. For example:

```c++
struct A {
  short a1;
  short a2;
};

struct B {
  [[clang::preferred_type(A)]] unsigned b1 : 32 = 0x000F'000C;
};
```

will associate the type `A` with the `b1` bit-field and is intended to display
something like this in the debugger:

```text
Process 2755547 stopped
* thread #1, name = 'test-preferred-', stop reason = step in
    frame #0: 0x0000555555555148 test-preferred-type`main at test.cxx:13:14
   10   int main()
   11   {
   12       B b;
-> 13       return b.b1;
   14   }
(lldb) v -T
(B) b = {
  (A:32) b1 = {
    (short) a1 = 12
    (short) a2 = 15
  }
}
```

Note that debuggers may not be able to handle more complex mappings, and so
this usage is debugger-dependent.
  }];
}

def CleanupDocs : Documentation {
  let Category = DocCatVariable;
  let Content = [{
This attribute allows a function to be run when a local variable goes out of
scope. The attribute takes the identifier of a function with a parameter type
that is a pointer to the type with the attribute.

```c
static void foo (int *) { ... }
static void bar (int *) { ... }
void baz (void) {
  int x __attribute__((cleanup(foo)));
  {
    int y __attribute__((cleanup(bar)));
  }
}
```

The above example will result in a call to `bar` being passed the address of
`y` when `y` goes out of scope, then a call to `foo` being passed the
address of `x` when `x` goes out of scope. If two or more variables share
the same scope, their `cleanup` callbacks are invoked in the reverse order
the variables were declared in. It is not possible to check the return value
(if any) of these `cleanup` callback functions.
}];
}

def CtorDtorDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `constructor` attribute causes the function to be called before entering
`main()`, and the `destructor` attribute causes the function to be called
after returning from `main()` or when the `exit()` function has been
called. Note, `quick_exit()`, `_Exit()`, and `abort()` prevent a function
marked `destructor` from being called.

The constructor or destructor function should not accept any arguments and its
return type should be `void`.

The attributes accept an optional argument used to specify the priority order
in which to execute constructor and destructor functions. The priority is
given as an integer constant expression between 101 and 65535 (inclusive).
Priorities outside of that range are reserved for use by the implementation. A
lower value indicates a higher priority of initialization. Note that only the
relative ordering of values is important. For example:

```c++
__attribute__((constructor(200))) void foo(void);
__attribute__((constructor(101))) void bar(void);
```

`bar()` will be called before `foo()`, and both will be called before
`main()`. If no argument is given to the `constructor` or `destructor`
attribute, they default to the value `65535`.
}];
}

def CoroOnlyDestroyWhenCompleteDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `coro_only_destroy_when_complete` attribute should be marked on a C++ class. The coroutines
whose return type is marked with the attribute are assumed to be destroyed only after the coroutine has
reached the final suspend point.

This is helpful for the optimizers to reduce the size of the destroy function for the coroutines.

For example,

```c++
A foo() {
  dtor d;
  co_await something();
  dtor d1;
  co_await something();
  dtor d2;
  co_return 43;
}
```

The compiler may generate the following pseudocode:

```c++
void foo.destroy(foo.Frame *frame) {
  switch(frame->suspend_index()) {
    case 1:
      frame->d.~dtor();
      break;
    case 2:
      frame->d.~dtor();
      frame->d1.~dtor();
      break;
    case 3:
      frame->d.~dtor();
      frame->d1.~dtor();
      frame->d2.~dtor();
      break;
    default: // coroutine completed or haven't started
      break;
  }

  frame->promise.~promise_type();
  delete frame;
}
```

The `foo.destroy()` function's purpose is to release all of the resources
initialized for the coroutine when it is destroyed in a suspended state.
However, if the coroutine is only ever destroyed at the final suspend state,
the rest of the conditions are superfluous.

The user can use the `coro_only_destroy_when_complete` attributo suppress
generation of the other destruction cases, optimizing the above `foo.destroy` to:

```c++
void foo.destroy(foo.Frame *frame) {
  frame->promise.~promise_type();
  delete frame;
}
```

  }];
}

def CoroReturnTypeAndWrapperDoc : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `[[clang::coro_return_type]]` attribute is used to help static analyzers to recognize
coroutines from the function signatures.

The `coro_return_type` attribute should be marked on a C++ class to mark it as
a **coroutine return type (CRT)**.

A function `R func(P1, .., PN)` has a coroutine return type (CRT) `R` if `R`
is marked by `[[clang::coro_return_type]]` and `R` has a promise type associated to it
(i.e., std::coroutine_traits\<R, P1, .., PN>::promise_type is a valid promise type).

If the return type of a function is a `CRT` then the function must be a coroutine.
Otherwise the program is invalid. It is allowed for a non-coroutine to return a `CRT`
if the function is marked with `[[clang::coro_wrapper]]`.

The `[[clang::coro_wrapper]]` attribute should be marked on a C++ function to mark it as
a **coroutine wrapper**. A coroutine wrapper is a function which returns a `CRT`,
is not a coroutine itself and is marked with `[[clang::coro_wrapper]]`.

Clang will enforce that all functions that return a `CRT` are either coroutines or marked
with `[[clang::coro_wrapper]]`. Clang will enforce this with an error.

From a language perspective, it is not possible to differentiate between a coroutine and a
function returning a CRT by merely looking at the function signature.

Coroutine wrappers, in particular, are susceptible to capturing
references to temporaries and other lifetime issues. This allows to avoid such lifetime
issues with coroutine wrappers.

For example,

```c++
// This is a CRT.
template <typename T> struct [[clang::coro_return_type]] Task {
  using promise_type = some_promise_type;
};

Task<int> increment(int a) { co_return a + 1; } // Fine. This is a coroutine.
Task<int> foo() { return increment(1); } // Error. foo is not a coroutine.

// Fine for a coroutine wrapper to return a CRT.
[[clang::coro_wrapper]] Task<int> foo() { return increment(1); }

void bar() {
  // Invalid. This intantiates a function which returns a CRT but is not marked as
  // a coroutine wrapper.
  std::function<Task<int>(int)> f = increment;
}
```

Note: `a_promise_type::get_return_object` is exempted from this analysis as it is a necessary
implementation detail of any coroutine library.
}];
}

def CodeAlignAttrDocs : Documentation {
  let Category = DocCatVariable;
  let Heading = "clang::code_align";
  let Content = [{
The `clang::code_align(N)` attribute applies to a loop and specifies the byte
alignment for a loop. The attribute accepts a positive integer constant
initialization expression indicating the number of bytes for the minimum
alignment boundary. Its value must be a power of 2, between 1 and 4096
(inclusive).

```c++
void foo() {
  int var = 0;
  [[clang::code_align(16)]] for (int i = 0; i < 10; ++i) var++;
}

void Array(int *array, size_t n) {
  [[clang::code_align(64)]] for (int i = 0; i < n; ++i) array[i] = 0;
}

void count () {
  int a1[10], int i = 0;
  [[clang::code_align(32)]] while (i < 10) { a1[i] += 3; }
}

void check() {
  int a = 10;
  [[clang::code_align(8)]] do {
    a = a + 1;
  } while (a < 20);
}

template<int A>
void func() {
  [[clang::code_align(A)]] for(;;) { }
}
```

  }];
}

def CoroLifetimeBoundDoc : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `[[clang::coro_lifetimebound]]` is a class attribute which can be applied
to a coroutine return type ([coro_return_type, coro_wrapper]) (i.e.
it should also be annotated with `[[clang::coro_return_type]]`).

All parameters of a function are considered to be lifetime bound if the function returns a
coroutine return type (CRT) annotated with `[[clang::coro_lifetimebound]]`.
This lifetime bound analysis can be disabled for a coroutine wrapper or a coroutine by annotating the function
with `[[clang::coro_disable_lifetimebound]]` function attribute .
See documentation of [lifetimebound] for details about lifetime bound analysis.

Reference parameters of a coroutine are susceptible to capturing references to temporaries or local variables.

For example,

```c++
task<int> coro(const int& a) { co_return a + 1; }
task<int> dangling_refs(int a) {
  // `coro` captures reference to a temporary. `foo` would now contain a dangling reference to `a`.
  auto foo = coro(1);
  // `coro` captures reference to local variable `a` which is destroyed after the return.
  return coro(a);
}
```

Lifetime bound static analysis can be used to detect such instances when coroutines capture references
which may die earlier than the coroutine frame itself. In the above example, if the CRT `task` is annotated with
`[[clang::coro_lifetimebound]]`, then lifetime bound analysis would detect capturing reference to
temporaries or return address of a local variable.

Both coroutines and coroutine wrappers are part of this analysis.

```c++
template <typename T> struct [[clang::coro_return_type, clang::coro_lifetimebound]] Task {
  using promise_type = some_promise_type;
};

Task<int> coro(const int& a) { co_return a + 1; }
[[clang::coro_wrapper]] Task<int> coro_wrapper(const int& a, const int& b) {
  return a > b ? coro(a) : coro(b);
}
Task<int> temporary_reference() {
  auto foo = coro(1); // warning: capturing reference to a temporary which would die after the expression.

  int a = 1;
  auto bar = coro_wrapper(a, 0); // warning: `b` captures reference to a temporary.

  co_return co_await coro(1); // fine.
}
[[clang::coro_wrapper]] Task<int> stack_reference(int a) {
  return coro(a); // warning: returning address of stack variable `a`.
}
```

This analysis can be disabled for all calls to a particular function by annotating the function
with function attribute `[[clang::coro_disable_lifetimebound]]`.
For example, this could be useful for coroutine wrappers which accept reference parameters
but do not pass them to the underlying coroutine or pass them by value.

```c++
Task<int> coro(int a) { co_return a + 1; }
[[clang::coro_wrapper, clang::coro_disable_lifetimebound]] Task<int> coro_wrapper(const int& a) {
  return coro(a + 1);
}
void use() {
  auto task = coro_wrapper(1); // use of temporary is fine as the argument is not lifetime bound.
}
```
}];
}

def CoroAwaitElidableDoc : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `[[clang::coro_await_elidable]]` is a class attribute which can be
applied to a coroutine return type. It provides a hint to the compiler to apply
Heap Allocation Elision more aggressively.

When a coroutine function returns such a type, a direct call expression therein
that returns a prvalue of a type attributed `[[clang::coro_await_elidable]]`
is said to be under a safe elide context if one of the following is true:

- it is the immediate right-hand side operand to a co_await expression.
- it is an argument to a `[[clang::coro_await_elidable_argument]]` parameter
  or parameter pack of another direct call expression under a safe elide context.

Do note that the safe elide context applies only to the call expression itself,
and the context does not transitively include any of its subexpressions unless
exceptional rules of `[[clang::coro_await_elidable_argument]]` apply.

The compiler performs heap allocation elision on call expressions under a safe
elide context, if the callee is a coroutine.

Example:

```c++
class [[clang::coro_await_elidable]] Task { ... };

Task foo();
Task bar() {
  co_await foo(); // foo()'s coroutine frame on this line is elidable
  auto t = foo(); // foo()'s coroutine frame on this line is NOT elidable
  co_await t;
}
```

Such elision replaces the heap allocated activation frame of the callee coroutine
with a local variable within the enclosing braces in the caller's stack frame.
The local variable, like other variables in coroutines, may be collected into the
coroutine frame, which may be allocated on the heap. The behavior is undefined
if the caller coroutine is destroyed earlier than the callee coroutine.

}];
}

def CoroAwaitElidableArgumentDoc : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `[[clang::coro_await_elidable_argument]]` is a function parameter attribute.
It works in conjunction with `[[clang::coro_await_elidable]]` to propagate a
safe elide context to a parameter or parameter pack if the function is called
under a safe elide context.

This is sometimes necessary on utility functions used to compose or modify the
behavior of a callee coroutine.

Example:

```c++
template <typename T>
class [[clang::coro_await_elidable]] Task { ... };

template <typename... T>
class [[clang::coro_await_elidable]] WhenAll { ... };

// `when_all` is a utility function that composes coroutines. It does not
// need to be a coroutine to propagate.
template <typename... T>
WhenAll<T...> when_all([[clang::coro_await_elidable_argument]] Task<T> tasks...);

Task<int> foo();
Task<int> bar();
Task<void> example1() {
  // `when_all`, `foo`, and `bar` are all elide safe because `when_all` is
  // under a safe elide context and, thanks to the [[clang::coro_await_elidable_argument]]
  // attribute, such context is propagated to foo and bar.
  co_await when_all(foo(), bar());
}

Task<void> example2() {
  // `when_all` and `bar` are elide safe. `foo` is not elide safe.
  auto f = foo();
  co_await when_all(f, bar());
}


Task<void> example3() {
  // None of the calls are elide safe.
  auto t = when_all(foo(), bar());
  co_await t;
}
```

}];
}

def CountedByDocs : Documentation {
  let Category = DocCatField;
  let Heading = "counted_by";
  let Content = [{
The `counted_by` attribute is applied to a pointer or flexible array member to
indicate that the pointer points to (or the flexible array member contains) at
least the number of *elements* given by the attribute's argument.

This attribute is used by {doc}`-fbounds-safety <BoundsSafety>` to propagate
bounds information on API surfaces without any ABI changes. This attribute is
also used to improve the results of the array bound sanitizer and the
`__builtin_dynamic_object_size` builtin.

Because the size of the pointee type must be known to compute the pointer's
bounds, such a pointer must not be used while its pointee type is incomplete; a
pointer to a forward-declared type is accepted on fields annotated with
`counted_by`, but the type must be completed before the pointer is used. If
the pointee type can never be completed, `counted_by` is rejected and
`sized_by` should be used instead. `void *` is a special case: as a GNU
extension (diagnosed by `-Wgnu-pointer-arith`), `counted_by` is accepted on
it, where it behaves like `sized_by` (the argument is treated as a byte count,
`void` having an assumed size of one byte).

A pointer annotated with `counted_by` must have a count of zero when it is
null. This requirement is currently only enforced when compiling with
{doc}`-fbounds-safety <BoundsSafety>` (see {ref}`Current status of
-fbounds-safety support in upstream Clang <bounds-safety-current-upstream-status>`). Use
`counted_by_or_null` for a pointer that may be null while carrying a nonzero
count.

#### Keeping pointer and count in sync

The `counted_by` attribute establishes a relationship between the annotated
pointer and its count: the pointer must point to at least `count` elements.
Assigning to only one of them can break this relationship.
Without {doc}`-fbounds-safety <BoundsSafety>`, it is the programmer's
responsibility to ensure the pointer and count remain in sync. With
`-fbounds-safety` it is automatically enforced. For example:

```c
struct buffer {
  int *buf __attribute__((counted_by(count)));
  size_t count;
};

void grow(struct buffer *b, size_t new_count) {
  // b->buf isn't updated. The underlying memory pointed to by b->buf might be
  // smaller than new_count which would contradict the counted_by attribute.
  // Compile error with -fbounds-safety but allowed without -fbounds-safety.
  b->count = new_count;
}
```

Updating both together - so that `buf` points to `count` elements - keeps
the attribute true. For example:

```c
void grow(struct buffer *b, size_t new_count) {
  // Allowed by -fbounds-safety
  int *new_buf = malloc(new_count * sizeof(int));
  // -fbounds-safety enforces that the `new_buf` points to at least `new_count`
  // integers at runtime. Without -fbounds-safety nothing enforces this.
  b->buf = new_buf;
  b->count = new_count;
}
```

#### Flexible array members

The `counted_by` attribute may also be applied to the flexible array member of
a structure in C. In this case the argument names the field member holding the
count of elements in the flexible array; that field must be within the same
non-anonymous, enclosing struct as the flexible array member.

This example specifies that the flexible array member `array` has the number
of elements allocated for it in `count`:

```c
struct bar;

struct foo {
  size_t count;
  char other;
  struct bar *array[] __attribute__((counted_by(count)));
};
```

This establishes a relationship between `array` and `count`. Specifically,
`array` must have at least `count` number of elements available. It's the
user's responsibility to ensure that this relationship is maintained through
changes to the structure.

In the following example, the allocated array erroneously has fewer elements
than what's specified by `p->count`. This would result in an out-of-bounds
access not being detected.

```c
#define SIZE_INCR 42

struct foo *p;

void foo_alloc(size_t count) {
  p = malloc(MAX(sizeof(struct foo),
                 offsetof(struct foo, array[0]) + count * sizeof(struct bar *)));
  p->count = count + SIZE_INCR;
}
```

The next example updates `p->count`, but breaks the relationship requirement
that `p->array` must have at least `p->count` number of elements available:

```c
#define SIZE_INCR 42

struct foo *p;

void foo_alloc(size_t count) {
  p = malloc(MAX(sizeof(struct foo),
                 offsetof(struct foo, array[0]) + count * sizeof(struct bar *)));
  p->count = count;
}

void use_foo(int index, int val) {
  p->count += SIZE_INCR + 1; /* 'count' is now larger than the number of elements of 'array'. */
  p->array[index] = val;     /* The sanitizer can't properly check this access. */
}
```

In this example, an update to `p->count` maintains the relationship
requirement:

```c
void use_foo(int index, int val) {
  if (p->count == 0)
    return;
  --p->count;
  p->array[index] = val;
}
```
  }];
}

def SizedByDocs : Documentation {
  let Category = DocCatField;
  let Heading = "sized_by";
  let Content = [{
The `sized_by` attribute is applied to a pointer to indicate that the pointer
points to memory containing at least the number of *bytes* given by the
attribute's argument. It is closely related to `counted_by`; the difference is
that `counted_by` counts the number of *elements* of the pointee type, whereas
`sized_by` counts the number of *bytes*. This makes `sized_by` the natural
choice for `void *` and other byte buffers.

This attribute is used by {doc}`-fbounds-safety <BoundsSafety>` to propagate
bounds information on API surfaces without any ABI changes. This attribute is
also used to improve the results of the array bound sanitizer and the
`__builtin_dynamic_object_size` builtin.

The argument is an expression of integer type, following the same rules as the
argument of `counted_by`. Unlike `counted_by`, `sized_by` cannot be
applied to a C99 flexible array member; it applies to pointers only. For
example:

```c
struct object {
  unsigned long size;
  void *data __attribute__((sized_by(size)));
};
```

A pointer annotated with `sized_by` must have a size of zero when it is null.
This requirement is currently only enforced when compiling with
{doc}`-fbounds-safety <BoundsSafety>` (see {ref}`Current status of
-fbounds-safety support in upstream Clang <bounds-safety-current-upstream-status>`). Use
`sized_by_or_null` for a pointer that may be null while carrying a nonzero
size.

#### Keeping pointer and size in sync

The `sized_by` attribute establishes a relationship between the annotated
pointer and its size: the pointer must point to at least `size` bytes.
Assigning to only one of them can break this relationship.
Without {doc}`-fbounds-safety <BoundsSafety>`, it is the programmer's
responsibility to ensure the pointer and size remain in sync. With
`-fbounds-safety` it is automatically enforced. For example:

```c
struct buffer {
  uint8_t *buf __attribute__((sized_by(size)));
  size_t size;
};

void grow(struct buffer *b, size_t new_size) {
  // b->buf isn't updated. The underlying memory pointed to by b->buf might be
  // smaller than new_size which would contradict the sized_by attribute.
  // Compile error with -fbounds-safety but allowed without -fbounds-safety.
  b->size = new_size;
}
```

Updating both together - so that `buf` points to `size` bytes - keeps
the attribute true. For example:

```c
void grow(struct buffer *b, size_t new_size) {
  // Allowed by -fbounds-safety
  uint8_t *new_buf = malloc(new_size);
  // -fbounds-safety enforces that the `new_buf` points to at least `new_size`
  // bytes at runtime. Without -fbounds-safety nothing enforces this.
  b->buf = new_buf;
  b->size = new_size;
}
```

#### Incomplete and variable-length pointees

`sized_by` is typically applied to `void *` or a pointer to a byte-sized
type, but it may be used with any pointee type. Two situations call for this,
both of which rule out counting fixed-size elements:

First, the pointee type may be incomplete, such as an opaque type. Its element
size is then unavailable, so `counted_by` cannot be used, whereas `sized_by`
bounds the memory in bytes and imposes no completeness requirement.

Second, the buffer may hold variable-length elements, so there is no fixed
element size to count, even though the total byte size is well defined. For
example, a buffer might pack together several structures that each end in a
flexible array member of differing length:

```c
struct var_len {
  int fam_size;
  char data[] __attribute__((counted_by(fam_size)));
};

struct buffer_view {
  int byte_size;
  struct var_len *buf __attribute__((sized_by(byte_size)));
};
```

Here `counted_by` cannot be applied to `buf` because its pointee is a
variable-length structure, but `sized_by` bounds the whole region in bytes;
the region is traversed by advancing a byte offset rather than by indexing
elements.
  }];
}

def CountedByOrNullDocs : Documentation {
  let Category = DocCatField;
  let Heading = "counted_by_or_null";
  let Content = [{
The `counted_by_or_null` attribute is applied to a pointer to indicate that,
if the pointer is non-null, it points to memory containing at least the number
of *elements* given by the attribute's argument. If the pointer is null, the
value of the argument is ignored and the pointer points to zero elements.

The `counted_by_or_null` attribute is identical to `counted_by` except that
it treats null pointers differently and cannot be applied to a flexible array
member. Whereas `counted_by` requires a null pointer to have a count of zero,
`counted_by_or_null` allows the pointer to be null regardless of the value of
the count. This supports the common idiom where a pointer is either null or
points to memory containing at least the given number of elements.

Currently only {doc}`-fbounds-safety <BoundsSafety>` makes use of the
distinction between `counted_by_or_null` and `counted_by` (see
{ref}`Current status of -fbounds-safety support in upstream Clang
<bounds-safety-current-upstream-status>`).
  }];
}

def SizedByOrNullDocs : Documentation {
  let Category = DocCatField;
  let Heading = "sized_by_or_null";
  let Content = [{
The `sized_by_or_null` attribute is applied to a pointer to indicate that, if
the pointer is non-null, it points to memory containing at least the number of
*bytes* given by the attribute's argument. If the pointer is null, the value of
the argument is ignored and the pointer points to zero bytes.

The `sized_by_or_null` attribute is identical to `sized_by` except in how
it treats null pointers. Whereas `sized_by` requires a null pointer to have a
size of zero, `sized_by_or_null` allows the pointer to be null regardless of
the value of the size. This supports the common idiom where a pointer is either
null or points to memory containing at least the given number of bytes.

Currently only {doc}`-fbounds-safety <BoundsSafety>` makes use of the
distinction between `sized_by_or_null` and `sized_by` (see
{ref}`Current status of -fbounds-safety support in upstream Clang
<bounds-safety-current-upstream-status>`).
  }];
}

def ClspvLibclcBuiltinDoc : Documentation {
  let Category = DocCatFunction;
  let Content = [{
Attribute used by [clspv][clspv] (OpenCL-C to Vulkan SPIR-V compiler) to identify functions coming from [libclc][libclc] (OpenCL-C builtin library).

```c
void __attribute__((clspv_libclc_builtin)) libclc_builtin() {}
```

[clspv]: https://github.com/google/clspv
[libclc]: https://libclc.llvm.org
}];
}

def NoTrivialAutoVarInitDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `__declspec(no_init_all)` attribute disables the automatic initialization
that the {option}`-ftrivial-auto-var-init` flag would have applied to locals in
a marked function, or instances of a marked type. Note that this attribute has
no effect for locals that are automatically initialized without the
{option}`-ftrivial-auto-var-init` flag.
}];
}

def DocCatNonBlockingNonAllocating : DocumentationCategory<"Performance Constraint Attributes"> {
  let Content = [{
The `nonblocking`, `blocking`, `nonallocating` and `allocating` attributes can be attached
to function types, including blocks, C++ lambdas, and member functions. The attributes declare
constraints about a function's behavior pertaining to blocking and heap memory allocation.

There are several rules for function types with these attributes, enforced with
compiler warnings:

- When assigning or otherwise converting to a function pointer of `nonblocking` or
  `nonallocating` type, the source must also be a function or function pointer of
  that type, unless it is a null pointer, i.e. the attributes should not be "spoofed". Conversions
  that remove the attributes are transparent and valid.
- An override of a `nonblocking` or `nonallocating` virtual method must also be declared
  with that same attribute (or a stronger one.) An overriding method may add an attribute.
- A redeclaration of a `nonblocking` or `nonallocating` function must also be declared with
  the same attribute (or a stronger one). A redeclaration may add an attribute.

The warnings are controlled by `-Wfunction-effects`, which is disabled by default.

The compiler also diagnoses function calls from `nonblocking` and `nonallocating`
functions to other functions which lack the appropriate attribute.
  }];
}

def NonBlockingDocs : Documentation {
  let Category = DocCatNonBlockingNonAllocating;
  let Heading = "nonblocking";
  let Content = [{
Declares that a function or function type either does or does not block in any way, according
to the optional, compile-time constant boolean argument, which defaults to true. When the argument
is false, the attribute is equivalent to `blocking`.

For the purposes of diagnostics, `nonblocking` is considered to include the
`nonallocating` guarantee and is therefore a "stronger" constraint or attribute.
  }];
}

def NonAllocatingDocs : Documentation {
  let Category = DocCatNonBlockingNonAllocating;
  let Heading = "nonallocating";
  let Content = [{
Declares that a function or function type either does or does not allocate heap memory, according
to the optional, compile-time constant boolean argument, which defaults to true. When the argument
is false, the attribute is equivalent to `allocating`.
  }];
}

def BlockingDocs : Documentation {
  let Category = DocCatNonBlockingNonAllocating;
  let Heading = "blocking";
  let Content = [{
Declares that a function potentially blocks, and prevents any potential inference of `nonblocking`
by the compiler.
  }];
}

def AllocatingDocs : Documentation {
  let Category = DocCatNonBlockingNonAllocating;
  let Heading = "allocating";
  let Content = [{
Declares that a function potentially allocates heap memory, and prevents any potential inference
of `nonallocating` by the compiler.
  }];
}

def NonStringDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `nonstring` attribute can be applied to the declaration of a variable or
a field whose type is a character pointer or character array to specify that
the buffer is not intended to behave like a null-terminated string. This will
silence diagnostics with code like:

```c
char BadStr[3] = "foo"; // No space for the null terminator, diagnosed
__attribute__((nonstring)) char NotAStr[3] = "foo"; // Not diagnosed
```
  }];
}

def ModularFormatDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `modular_format` attribute can be applied to a function that bears the
`format` attribute (or standard library functions) to indicate that the
implementation is "modular", that is, that the implementation is logically
divided into a number of named aspects. When the compiler can determine that
not all aspects of the implementation are needed for a given call, the compiler
may redirect the call to the identifier given as the first argument to the
attribute (the modular implementation function).

The second argument is an implementation name, and the remaining arguments are
aspects of the format string for the compiler to report. The implementation
name is an unevaluated identifier in the C namespace.

The compiler reports that a call requires an aspect by issuing a relocation for
the symbol `<impl_name>_<aspect>` at the point of the call. This arranges for
code and data needed to support the aspect of the implementation to be brought
into the link to satisfy weak references in the modular implemenation function.
If the compiler does not understand an aspect, it must summarily consider any
call to require that aspect.

For example, say `printf` is annotated with
`modular_format(__modular_printf, "__printf", "float")`. Then, a call to
`printf(var, 42)` would be untouched. A call to `printf("%d", 42)` would
become a call to `__modular_printf` with the same arguments, as would
`printf("%f", 42.0)`. The latter would be accompanied with a strong
relocation against the symbol `__printf_float`, which would bring floating
point support for `printf` into the link.

If the attribute appears more than once on a declaration, or across a chain of
redeclarations, it is an error for the attributes to have different arguments,
excepting that the aspects may be in any order.

The following aspects are currently supported:

- `fixed`: The call has a C ISO 18037 fixed-point argument.
- `float`: The call has a floating-point argument.
  }];
}

def OverflowBehaviorDocs : Documentation {
  let Category = DocCatType;
  let Content = [{
The `overflow_behavior` attribute provides fine-grained, type-level control
over how arithmetic operations on an integer type behave on overflow. It may be
applied to a `typedef`, to a variable or data member, or to an integer type
directly, and accepts one of two behaviors as its argument:

- `wrap`: arithmetic on the attributed type wraps on overflow, using two's
  complement semantics. This is equivalent to `-fwrapv` but scoped to the
  attributed type, and works for both signed and unsigned types. UBSan's
  `signed-integer-overflow`, `unsigned-integer-overflow`,
  `implicit-signed-integer-truncation`, and
  `implicit-unsigned-integer-truncation` checks are suppressed for the type.
- `trap`: arithmetic on the attributed type is checked for overflow, enabling
  overflow checks for the type even when `-fwrapv` is in effect globally.

```c++
typedef unsigned int __attribute__((overflow_behavior(trap))) non_wrapping_uint;

non_wrapping_uint add_one(non_wrapping_uint a) {
  return a + 1; // Overflow is checked for this operation.
}

int mul_alot(int n) {
  int __attribute__((overflow_behavior(wrap))) a = n;
  return a * 1337; // Overflow is not checked and is well-defined.
}
```

The keyword spellings `__ob_wrap` and `__ob_trap` are equivalent to
`overflow_behavior(wrap)` and `overflow_behavior(trap)` respectively.

The attribute wholly overrides global flags (`-ftrapv`, `-fwrapv`,
sanitizers, and Sanitizer Special Case Lists) for the attributed type. It can
only be applied to integer types.

This feature is experimental and must be enabled with the `-cc1` option
`-fexperimental-overflow-behavior-types`. For full details on promotion and
conversion rules, pointer semantics, diagnostics, and interaction with
sanitizers, see {doc}`OverflowBehaviorTypes`.
  }];
}

def MSStructDocs : Documentation {
  let Category = DocCatDecl;
  let Content = [{
The `ms_struct` and `gcc_struct` attributes request the compiler to enter a
special record layout compatibility mode which mimics the layout of Microsoft or
Itanium C++ ABI respectively. Obviously, if the current C++ ABI matches the
requested ABI, the attribute does nothing. However, if it does not, annotated
structure or class is laid out in a special compatibility mode, which slightly
changes offsets for fields and bit-fields. The intention is to match the layout
of the requested ABI for structures which only use C features.

Note that the default behavior can be controlled by `-mms-bitfields` and
`-mno-ms-bitfields` switches and via `#pragma ms_struct`.

The primary difference is for bitfields, where the MS variant only packs
adjacent fields into the same allocation unit if they have integral types
of the same size, while the GCC/Itanium variant packs all fields in a bitfield
tightly.
  }];
}

def PersonalityDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
`__attribute__((personality(<routine>)))` is used to specify a personality
routine that is different from the language that is being used to implement the
function. This is a targeted, low-level feature aimed at language runtime
implementors who write runtime support code in C/C++ but need that code to
participate in a foreign language's exception-handling or unwinding model.

A personality routine is a language-specific callback attached to each stack
frame that the unwinder invokes to determine whether that frame handles a given
exception and what cleanup actions to perform. It effectively colors the
language-agnostic unwinding mechanism with language-specific semantics, enabling
different languages to coexist on the same call stack while each interpreting
exceptions according to their own rules.
  }];
}

def SentinelDocs : Documentation {
  let Category = DocCatFunction;
  let Content = [{
The `sentinel` attribute can be applied to variadic functions and pointers to
variadic functions, to diagnose each function call that does not pass a
sentinel value (a null pointer constant) as the last argument to the function
call. The attribute accepts two optional arguments: the first argument is the
position of the expected sentinel value, starting from the last parameter. The
second argument describes whether the last fixed parameter is treated as a
valid sentinel value when set to '1'.
All arguments described above default to '0' when elided.
The attribute is also supported with blocks and in Objective-C.

```c
void foo(const char*, ...) __attribute__((sentinel));
void bar(int, ...) __attribute__((sentinel(1)));
void baz(const char*, const char*, ...) __attribute__((sentinel(0, 1)));

void example() {
  foo("Example", (void*)0);
  foo("Another", "example", NULL);
  foo("Missing", "sentinel"); // Not OK

  bar(1, 2, NULL, 3);         // OK: sentinel value at the 2nd to last position
  bar(1, 2, 3, nullptr, 4);   // OK: `nullptr` is valid in C23
  bar(1, 2, 3, 4, NULL);      // Not OK

  baz("Test", "with", "multiple", "args", NULL);
  baz("One", NULL);           // OK: last fixed parameter is a valid sentinel

  void (*ptr) (int arg, ...) __attribute__ ((__sentinel__));
  ptr(1, 2, 3, NULL);
}
```

```c++
struct Ty {
    int value;

    template<typename T>
    auto&& foo(T&& val, ...) __attribute__((sentinel(1))) {
        return std::forward<T>(val);
    }

    template<class Self>
    auto&& bar(this Self&& self, ...) __attribute__((sentinel(1))) {
        return std::forward<Self>(self).value;
    }
};

void example2() {
    auto sty = Ty{};
    sty.foo(1, nullptr, 3);
    sty.bar(1, nullptr, 3);

    auto lmbd = [](int a, ...) __attribute__((sentinel)) {};
    lmbd(1, 2, nullptr);
}
```
  }];
}