Attributes in Clang

Introduction

This page lists the attributes currently supported by Clang.

AArch64 SME Attributes

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 for more details about the features related to the SME extension.

See Procedure Call Standard for the Arm® 64-bit Architecture (AArch64) for more details about streaming-interface functions and shared/private-ZA interface functions.

__arm_in

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_in

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.

__arm_inout

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_inout

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.

__arm_locally_streaming

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_locally_streaming

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.

__arm_new

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_new

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.

__arm_out

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_out

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.

__arm_preserves

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_preserves

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.

__arm_streaming

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_streaming

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.

__arm_streaming_compatible

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__arm_streaming_compatible

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.

AMD GPU Attributes

amdgpu_flat_work_group_size

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

amdgpu_flat_work_group_size

clang::amdgpu_flat_work_group_size

Yes

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.

amdgpu_max_num_work_groups

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

amdgpu_max_num_work_groups

clang::amdgpu_max_num_work_groups

Yes

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.

amdgpu_num_sgpr

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

amdgpu_num_sgpr

clang::amdgpu_num_sgpr

Yes

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.

amdgpu_num_vgpr

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

amdgpu_num_vgpr

clang::amdgpu_num_vgpr

Yes

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.

amdgpu_waves_per_eu

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

amdgpu_waves_per_eu

clang::amdgpu_waves_per_eu

Yes

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.

Calling Conventions

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.

aarch64_sve_pcs

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

aarch64_sve_pcs

clang::aarch64_sve_pcs

clang::aarch64_sve_pcs

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_vector_pcs

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

aarch64_vector_pcs

clang::aarch64_vector_pcs

clang::aarch64_vector_pcs

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 on the Arm Developer website.

fastcall

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

fastcall

gnu::fastcall

gnu::fastcall

__fastcall
_fastcall

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 on MSDN.

m68k_rtd

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

m68k_rtd

clang::m68k_rtd

clang::m68k_rtd

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.

ms_abi

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ms_abi

gnu::ms_abi

gnu::ms_abi

On non-Windows x86_64 targets, this attribute changes the calling convention of a function to match the default convention used on Windows x86_64. This attribute has no effect on Windows targets or non-x86_64 targets.

pcs

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

pcs

gnu::pcs

gnu::pcs

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”.

preserve_all

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preserve_all

clang::preserve_all

clang::preserve_all

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.

preserve_most

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preserve_most

clang::preserve_most

clang::preserve_most

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.

preserve_none

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preserve_none

clang::preserve_none

clang::preserve_none

Yes

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.

regcall

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

regcall

gnu::regcall

gnu::regcall

__regcall

On x86 targets, this attribute changes the calling convention to __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.

regparm

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

regparm

gnu::regparm

gnu::regparm

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.

riscv::vector_cc, riscv_vector_cc, clang::riscv_vector_cc

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

riscv_vector_cc

riscv::vector_cc
clang::riscv_vector_cc

riscv::vector_cc
clang::riscv_vector_cc

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.

stdcall

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

stdcall

gnu::stdcall

gnu::stdcall

__stdcall
_stdcall

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 on MSDN.

sysv_abi

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

sysv_abi

gnu::sysv_abi

gnu::sysv_abi

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.

thiscall

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

thiscall

gnu::thiscall

gnu::thiscall

__thiscall
_thiscall

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 on MSDN.

vectorcall

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

vectorcall

clang::vectorcall

clang::vectorcall

__vectorcall
_vectorcall

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 on MSDN for more details.

Consumed Annotation Checking

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.

callable_when

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

callable_when

clang::callable_when

Yes

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")))

consumable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

consumable

clang::consumable

Yes

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.

param_typestate

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

param_typestate

clang::param_typestate

Yes

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.

return_typestate

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

return_typestate

clang::return_typestate

Yes

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.

set_typestate

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

set_typestate

clang::set_typestate

Yes

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.

test_typestate

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

test_typestate

clang::test_typestate

Yes

Use __attribute__((test_typestate(tested_state))) to indicate that a method returns true if the object is in the specified state..

Customizing Swift Import

Clang supports additional attributes for customizing how APIs are imported into Swift.

swift_async

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_async

clang::swift_async

clang::swift_async

Yes

The swift_async attribute specifies if and how a particular function or Objective-C method is imported into a swift async method. For instance:

@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.

swift_async_error

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_async_error

clang::swift_async_error

clang::swift_async_error

Yes

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:

@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.

swift_async_name

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_async_name

Yes

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.

@interface URL
+ (void) loadContentsFrom:(URL *)url callback:(void (^)(NSData *))data __attribute__((__swift_async_name__("URL.loadContentsFrom(_:)")))
@end

swift_attr

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_attr

Yes

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.

swift_bridge

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_bridge

The swift_bridge attribute indicates that the declaration to which the attribute appertains is bridged to the named Swift type.

__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.

swift_bridged

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_bridged_typedef

Yes

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).

@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.

swift_error

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_error

Yes

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.

swift_name

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_name

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.

@interface URL
- (void) initWithString:(NSString *)s __attribute__((__swift_name__("URL.init(_:)")))
@end

void __attribute__((__swift_name__("squareRoot()"))) sqrt(double v) {
}

swift_newtype

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_newtype
swift_wrapper

Yes

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.

    // 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)));
    

swift_objc_members

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_objc_members

Yes

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.

swift_private

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

swift_private

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.

Declaration Attributes

Owner

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

gsl::Owner

Yes

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.

Pointer

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

gsl::Pointer

Yes

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.

int f() {
  IntPointer P(IntOwner{}); // P "points into" a temporary IntOwner object
  P.getInt(); // P is dangling
}

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.

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.
}

__single_inheritance, __multiple_inheritance, __virtual_inheritance

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__single_inheritance
__multiple_inheritance
__virtual_inheritance
__unspecified_inheritance

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:

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:

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:

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

asm

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

asm
__asm__

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:

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.

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.

coro_await_elidable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_await_elidable

clang::coro_await_elidable

clang::coro_await_elidable

Yes

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:

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.

coro_await_elidable_argument

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_await_elidable_argument

clang::coro_await_elidable_argument

clang::coro_await_elidable_argument

Yes

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:

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;
}

coro_disable_lifetimebound

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_disable_lifetimebound

clang::coro_disable_lifetimebound

clang::coro_disable_lifetimebound

Yes

The [[clang::coro_lifetimebound]] is a class attribute which can be applied to a coroutine return type (CRT) (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 [[clang::lifetimebound]] for details about lifetime bound analysis.

Reference parameters of a coroutine are susceptible to capturing references to temporaries or local variables.

For example,

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.

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.

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.
}

coro_lifetimebound

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_lifetimebound

clang::coro_lifetimebound

clang::coro_lifetimebound

Yes

The [[clang::coro_lifetimebound]] is a class attribute which can be applied to a coroutine return type (CRT) (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 [[clang::lifetimebound]] for details about lifetime bound analysis.

Reference parameters of a coroutine are susceptible to capturing references to temporaries or local variables.

For example,

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.

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.

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.
}

coro_only_destroy_when_complete

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_only_destroy_when_complete

clang::coro_only_destroy_when_complete

clang::coro_only_destroy_when_complete

Yes

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,

A foo() {
  dtor d;
  co_await something();
  dtor d1;
  co_await something();
  dtor d2;
  co_return 43;
}

The compiler may generate the following pseudocode:

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:

void foo.destroy(foo.Frame *frame) {
  frame->promise.~promise_type();
  delete frame;
}

coro_return_type

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_return_type

clang::coro_return_type

clang::coro_return_type

Yes

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,

// 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.

coro_wrapper

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

coro_wrapper

clang::coro_wrapper

clang::coro_wrapper

Yes

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,

// 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.

deprecated

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

deprecated

gnu::deprecated
deprecated

gnu::deprecated
deprecated

deprecated

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:

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.

empty_bases

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

empty_bases

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.

enum_extensibility

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

enum_extensibility

clang::enum_extensibility

clang::enum_extensibility

Yes

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.

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
}

external_source_symbol

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

external_source_symbol

clang::external_source_symbol

clang::external_source_symbol

Yes

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:

@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:

__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.

flag_enum

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

flag_enum

clang::flag_enum

clang::flag_enum

Yes

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.

grid_constant

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

grid_constant

__grid_constant__

Yes

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.

layout_version

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

layout_version

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.

lto_visibility_public

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

lto_visibility_public

clang::lto_visibility_public

clang::lto_visibility_public

Yes

See LTO Visibility.

managed

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

managed

__managed__

Yes

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 __hipRegisterManagedVariable 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.

novtable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

novtable

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.

ns_error_domain

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ns_error_domain

Yes

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:

#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,
};

objc_boxable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_boxable

clang::objc_boxable

clang::objc_boxable

Yes

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:

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);

objc_direct

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_direct

clang::objc_direct

clang::objc_direct

Yes

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.

objc_direct_members

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_direct_members

clang::objc_direct_members

clang::objc_direct_members

Yes

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.

objc_non_runtime_protocol

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_non_runtime_protocol

clang::objc_non_runtime_protocol

clang::objc_non_runtime_protocol

Yes

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.

objc_nonlazy_class

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_nonlazy_class

clang::objc_nonlazy_class

clang::objc_nonlazy_class

Yes

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.

objc_runtime_name

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_runtime_name

clang::objc_runtime_name

clang::objc_runtime_name

Yes

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:

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

objc_runtime_visible

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_runtime_visible

clang::objc_runtime_visible

clang::objc_runtime_visible

Yes

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.

objc_subclassing_restricted

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_subclassing_restricted

clang::objc_subclassing_restricted

clang::objc_subclassing_restricted

Yes

This attribute can be added to an Objective-C @interface declaration to ensure that this class cannot be subclassed.

preferred_name

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preferred_name

clang::preferred_name

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:

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.

randomize_layout, no_randomize_layout

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_randomize_layout

gnu::no_randomize_layout

gnu::no_randomize_layout

Yes

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:

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:

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.

randomize_layout, no_randomize_layout

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

randomize_layout

gnu::randomize_layout

gnu::randomize_layout

Yes

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:

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:

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.

selectany

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

selectany

gnu::selectany

gnu::selectany

selectany

This attribute appertains to a global symbol, causing it to have a weak definition ( linkonce ), allowing the linker to select any definition.

For more information see gcc documentation or msvc documentation.

transparent_union

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

transparent_union

gnu::transparent_union

gnu::transparent_union

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++.

trivial_abi

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

trivial_abi

clang::trivial_abi

Yes

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:

// 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.

If a type is trivial for the purpose of calls, it is assumed to be trivially relocatable for the purpose of __is_trivially_relocatable.

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

using_if_exists

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

using_if_exists

clang::using_if_exists

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:

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.

weak

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

weak

gnu::weak

gnu::weak

Yes

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.

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.

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.

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;
}

Field Attributes

counted_by

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

counted_by

clang::counted_by

clang::counted_by

Clang supports the counted_by attribute on the flexible array member of a structure in C. The argument for the attribute is the name of a field member holding the count of elements in the flexible array. This information can be used to improve the results of the array bound sanitizer and the __builtin_dynamic_object_size builtin. The count field member 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:

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.

#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:

#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:

void use_foo(int index, int val) {
  if (p->count == 0)
    return;
  --p->count;
  p->array[index] = val;
}

counted_by_or_null

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

counted_by_or_null

clang::counted_by_or_null

clang::counted_by_or_null

Clang supports the counted_by attribute on the flexible array member of a structure in C. The argument for the attribute is the name of a field member holding the count of elements in the flexible array. This information can be used to improve the results of the array bound sanitizer and the __builtin_dynamic_object_size builtin. The count field member 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:

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.

#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:

#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:

void use_foo(int index, int val) {
  if (p->count == 0)
    return;
  --p->count;
  p->array[index] = val;
}

no_unique_address

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_unique_address
msvc::no_unique_address

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:

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.

preferred_type

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preferred_type

clang::preferred_type

clang::preferred_type

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.

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:

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:

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.

sized_by

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

sized_by

clang::sized_by

clang::sized_by

Clang supports the counted_by attribute on the flexible array member of a structure in C. The argument for the attribute is the name of a field member holding the count of elements in the flexible array. This information can be used to improve the results of the array bound sanitizer and the __builtin_dynamic_object_size builtin. The count field member 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:

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.

#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:

#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:

void use_foo(int index, int val) {
  if (p->count == 0)
    return;
  --p->count;
  p->array[index] = val;
}

sized_by_or_null

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

sized_by_or_null

clang::sized_by_or_null

clang::sized_by_or_null

Clang supports the counted_by attribute on the flexible array member of a structure in C. The argument for the attribute is the name of a field member holding the count of elements in the flexible array. This information can be used to improve the results of the array bound sanitizer and the __builtin_dynamic_object_size builtin. The count field member 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:

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.

#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:

#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:

void use_foo(int index, int val) {
  if (p->count == 0)
    return;
  --p->count;
  p->array[index] = val;
}

Function Attributes

#pragma omp declare simd

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

`` declare simd``

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:

#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:

simdlen(length)
linear(argument-list[:constant-linear-step])
aligned(argument-list[:alignment])
uniform(argument-list)
inbranch
notinbranch

#pragma omp declare target

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

`` declare target``

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:

#pragma omp declare target new-line
declarations-definition-seq
#pragma omp end declare target new-line

or

#pragma omp declare target (extended-list) new-line

or

#pragma omp declare target clause[ [,] clause ... ] new-line

where clause is one of the following:

to(extended-list)
link(list)
device_type(host | nohost | any)

#pragma omp declare variant

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

`` declare variant``

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:

#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:

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)}:

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.

SV_DispatchThreadID

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

SV_DispatchThreadID

The SV_DispatchThreadID semantic, when applied to an input parameter, specifies a data binding to map the global thread offset within the Dispatch call (per dimension of the group) to the specified parameter. When applied to a field of a struct, the data binding is specified to the field when the struct is used as a parameter type. The semantic on the field is ignored when not used as a parameter. This attribute is only supported in compute shaders.

The full documentation is available here: https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/sv-dispatchthreadid

SV_GroupIndex

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

SV_GroupIndex

The SV_GroupIndex semantic, when applied to an input parameter, specifies a data binding to map the group index to the specified parameter. This attribute is only supported in compute shaders.

The full documentation is available here: https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/sv-groupindex

WaveSize

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

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:

``[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:

``[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

_Noreturn

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

_Noreturn

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.

__funcref

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__funcref

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.

abi_tag

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

abi_tag

gnu::abi_tag

Yes

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.

acquire_capability, acquire_shared_capability

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

acquire_capability
acquire_shared_capability
exclusive_lock_function
shared_lock_function

clang::acquire_capability
clang::acquire_shared_capability

Marks a function as acquiring a capability.

alloc_align

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

alloc_align

gnu::alloc_align

gnu::alloc_align

Use __attribute__((alloc_align(<alignment>)) on a function declaration to specify that the return value of the function (which must be a pointer type) is at least as aligned as the value of the indicated parameter. 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.

// The returned pointer has the alignment specified by the first parameter.
void *a(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.

alloc_size

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

alloc_size

gnu::alloc_size

gnu::alloc_size

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

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);
}

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.

allocator

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

allocator

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.

This attribute does not affect optimizations in any way, unlike GCC’s __attribute__((malloc)).

always_inline, __force_inline

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

always_inline

gnu::always_inline
clang::always_inline

gnu::always_inline
clang::always_inline

__forceinline

Yes

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.

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:

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 GCC Common Function Attribute docs, and the GCC Inline docs.

artificial

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

artificial

gnu::artificial

gnu::artificial

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.

assert_capability, assert_shared_capability

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

assert_capability
assert_shared_capability

clang::assert_capability
clang::assert_shared_capability

Marks a function that dynamically tests whether a capability is held, and halts the program if it is not held.

assume

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

omp::assume

Yes

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 __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:

"omp_no_openmp"
"omp_no_openmp_routines"
"omp_no_parallelism"

The OpenMP standard defines the meaning of OpenMP assumptions (“omp_XYZ” is spelled “XYZ” in the OpenMP 5.1 Standard).

assume_aligned

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

assume_aligned

gnu::assume_aligned

gnu::assume_aligned

Yes

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.

// 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.

availability

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

availability

clang::availability

clang::availability

Yes

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:

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 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 shadermodel

Supported environment names for the ShaderModel platform:

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

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, 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:

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.,:

@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:

@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.

#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:

#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:

#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:

#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 @available

btf_decl_tag

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

btf_decl_tag

clang::btf_decl_tag

clang::btf_decl_tag

Yes

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.

callback

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

callback

clang::callback

clang::callback

Yes

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.

__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);

carries_dependency

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

carries_dependency

carries_dependency

Yes

The carries_dependency attribute specifies dependency propagation into and out of functions.

When specified on a function or Objective-C method, the carries_dependency attribute means that the return value carries a dependency out of the function, so that the implementation need not constrain ordering upon return from that function. Implementations of the function and its caller may choose to preserve dependencies instead of emitting memory ordering instructions such as fences.

Note, this attribute does not change the meaning of the program, but may result in generation of more efficient code.

cf_consumed

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cf_consumed

clang::cf_consumed

clang::cf_consumed

Yes

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.

cf_returns_not_retained

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cf_returns_not_retained

clang::cf_returns_not_retained

clang::cf_returns_not_retained

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.

cf_returns_retained

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cf_returns_retained

clang::cf_returns_retained

clang::cf_returns_retained

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.

cfi_canonical_jump_table

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cfi_canonical_jump_table

clang::cfi_canonical_jump_table

clang::cfi_canonical_jump_table

Yes

Use __attribute__((cfi_canonical_jump_table)) on a function declaration to make the function’s CFI jump table canonical. See the CFI documentation for more details.

clang::builtin_alias, clang_builtin_alias

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

clang_builtin_alias

clang::builtin_alias

clang::builtin_alias

Yes

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.

clang_arm_builtin_alias

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

__clang_arm_builtin_alias

clang::__clang_arm_builtin_alias

clang::__clang_arm_builtin_alias

Yes

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.

clspv_libclc_builtin

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

clspv_libclc_builtin

clang::clspv_libclc_builtin

clang::clspv_libclc_builtin

Yes

Attribute used by clspv (OpenCL-C to Vulkan SPIR-V compiler) to identify functions coming from libclc (OpenCL-C builtin library).

void __attribute__((clspv_libclc_builtin)) libclc_builtin() {}

cmse_nonsecure_entry

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cmse_nonsecure_entry

Yes

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 for more information.

code_seg

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

code_seg

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) on MSDN.

cold

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cold

gnu::cold

gnu::cold

Yes

__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))).

constructor

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

constructor

gnu::constructor

gnu::constructor

Yes

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:

__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.

convergent

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

convergent

clang::convergent

clang::convergent

Yes

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.

In languages designed for SPMD/SIMT programming model, e.g. OpenCL or CUDA, the call instructions of a function with this attribute must be executed by all work items or threads in a work group or sub group.

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:

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

cpu_dispatch

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cpu_dispatch

clang::cpu_dispatch

clang::cpu_dispatch

cpu_dispatch

Yes

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:

// 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.

// 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.

cpu_specific

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

cpu_specific

clang::cpu_specific

clang::cpu_specific

cpu_specific

Yes

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:

// 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.

// 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.

destructor

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

destructor

gnu::destructor

gnu::destructor

Yes

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:

__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.

diagnose_as_builtin

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

diagnose_as_builtin

clang::diagnose_as_builtin

clang::diagnose_as_builtin

Yes

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.

__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.

__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.

diagnose_if

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

diagnose_if

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:

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:

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).

disable_sanitizer_instrumentation

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

disable_sanitizer_instrumentation

clang::disable_sanitizer_instrumentation

clang::disable_sanitizer_instrumentation

Yes

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.

disable_tail_calls

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

disable_tail_calls

clang::disable_tail_calls

clang::disable_tail_calls

Yes

The disable_tail_calls attribute instructs the backend to not perform tail call optimization inside the marked function.

For example:

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.

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.
  }
};

enable_if

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

enable_if

Yes

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.

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:

__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.

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:

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:

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.

enforce_tcb

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

enforce_tcb

clang::enforce_tcb

clang::enforce_tcb

Yes

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

enforce_tcb_leaf

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

enforce_tcb_leaf

clang::enforce_tcb_leaf

clang::enforce_tcb_leaf

Yes

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

error, warning

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

error
warning

gnu::error
gnu::warning

gnu::error
gnu::warning

Yes

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.

__attribute__((warning("oh no"))) void dontcall();
void foo() {
  if (someCompileTimeAssertionThatsTrue)
    dontcall(); // Warning

  dontcall(); // Warning

  if (someCompileTimeAssertionThatsFalse)
    dontcall(); // No Warning
  sizeof(dontcall()); // No Warning
}

exclude_from_explicit_instantiation

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

exclude_from_explicit_instantiation

clang::exclude_from_explicit_instantiation

clang::exclude_from_explicit_instantiation

Yes

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:

// 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:

// 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.

export_name

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

export_name

clang::export_name

clang::export_name

Yes

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.

flatten

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

flatten

gnu::flatten

gnu::flatten

Yes

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.

force_align_arg_pointer

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

force_align_arg_pointer

gnu::force_align_arg_pointer

gnu::force_align_arg_pointer

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).

__attribute__ ((force_align_arg_pointer))
void f () {
  ...
}

format

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

format

gnu::format

gnu::format

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 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:

    __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:

    __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:

__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:

__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.

function_return

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

function_return

gnu::function_return

gnu::function_return

Yes

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.

gnu_inline

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

gnu_inline

gnu::gnu_inline

gnu::gnu_inline

Yes

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, -std=c94, or -fgnu89-inline.

guard

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

guard

clang::guard

clang::guard

guard

Yes

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>

hot

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

hot

gnu::hot

gnu::hot

Yes

__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))).

hybrid_patchable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

hybrid_patchable

clang::hybrid_patchable

clang::hybrid_patchable

hybrid_patchable

Yes

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.

ifunc

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ifunc

gnu::ifunc

gnu::ifunc

Yes

__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. Targets other than ELF and Mach-O currently do not support this attribute.

import_module

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

import_module

clang::import_module

clang::import_module

Yes

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.

import_name

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

import_name

clang::import_name

clang::import_name

Yes

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.

internal_linkage

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

internal_linkage

clang::internal_linkage

clang::internal_linkage

Yes

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.

interrupt (ARM)

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

interrupt

gnu::interrupt

gnu::interrupt

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.

interrupt (AVR)

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

interrupt

gnu::interrupt

gnu::interrupt

Yes

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.

interrupt (MIPS)

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

interrupt

gnu::interrupt

gnu::interrupt

Yes

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”.

interrupt (RISC-V)

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

interrupt

gnu::interrupt

gnu::interrupt

Yes

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 user, supervisor, and machine. If there is no parameter, then it defaults to machine.

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.

interrupt (X86)

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

interrupt

gnu::interrupt

gnu::interrupt

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:

__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:

__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.

lifetimebound

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

lifetimebound

clang::lifetimebound

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 structs) 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:

#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:

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, but does not affect whether temporary objects have their lifetimes extended.

long_call, far

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

long_call
far

gnu::long_call
gnu::far

gnu::long_call
gnu::far

Yes

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.

malloc

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

malloc

gnu::malloc

gnu::malloc

restrict

Yes

The malloc attribute indicates that the function acts like a system memory allocation function, returning a pointer to allocated storage disjoint from the storage for any other object accessible to the caller.

micromips

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

micromips

gnu::micromips

gnu::micromips

Yes

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.

mig_server_routine

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

mig_server_routine

clang::mig_server_routine

clang::mig_server_routine

Yes

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]].

min_vector_width

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

min_vector_width

clang::min_vector_width

clang::min_vector_width

Yes

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.

minsize

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

minsize

clang::minsize

clang::minsize

Yes

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.

no_builtin

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_builtin

clang::no_builtin

clang::no_builtin

Yes

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.

// 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;
}

no_caller_saved_registers

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_caller_saved_registers

gnu::no_caller_saved_registers

gnu::no_caller_saved_registers

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:

__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.

no_profile_instrument_function

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_profile_instrument_function

gnu::no_profile_instrument_function

gnu::no_profile_instrument_function

Yes

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.

no_sanitize

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_sanitize

clang::no_sanitize

clang::no_sanitize

Yes

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 Controlling Code Generation for a full list of supported sanitizer flags.

no_sanitize_address, no_address_safety_analysis

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_address_safety_analysis
no_sanitize_address
no_sanitize_thread
no_sanitize_memory

gnu::no_address_safety_analysis
gnu::no_sanitize_address
gnu::no_sanitize_thread
clang::no_sanitize_memory

gnu::no_address_safety_analysis
gnu::no_sanitize_address
gnu::no_sanitize_thread
clang::no_sanitize_memory

Yes

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.

no_sanitize_memory

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_address_safety_analysis
no_sanitize_address
no_sanitize_thread
no_sanitize_memory

gnu::no_address_safety_analysis
gnu::no_sanitize_address
gnu::no_sanitize_thread
clang::no_sanitize_memory

gnu::no_address_safety_analysis
gnu::no_sanitize_address
gnu::no_sanitize_thread
clang::no_sanitize_memory

Yes

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.

no_sanitize_thread

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_address_safety_analysis
no_sanitize_address
no_sanitize_thread
no_sanitize_memory

gnu::no_address_safety_analysis
gnu::no_sanitize_address
gnu::no_sanitize_thread
clang::no_sanitize_memory

gnu::no_address_safety_analysis
gnu::no_sanitize_address
gnu::no_sanitize_thread
clang::no_sanitize_memory

Yes

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.

no_speculative_load_hardening

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_speculative_load_hardening

clang::no_speculative_load_hardening

clang::no_speculative_load_hardening

Yes

This attribute can be applied to a function declaration in order to indicate

that Speculative Load Hardening 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 -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:

__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);
}

no_split_stack

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_split_stack

gnu::no_split_stack

gnu::no_split_stack

Yes

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.

no_stack_protector, safebuffers

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

no_stack_protector

clang::no_stack_protector
gnu::no_stack_protector

clang::no_stack_protector
gnu::no_stack_protector

safebuffers

Yes

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.

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.

noalias

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

noalias

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.

nocf_check

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nocf_check

gnu::nocf_check

gnu::nocf_check

Yes

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).

noconvergent

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

noconvergent

clang::noconvergent

clang::noconvergent

noconvergent

Yes

This attribute prevents a function from being treated as convergent, which means that optimizations can only move calls to that function to control-equivalent blocks. If a statement is marked as noconvergent and contains calls, it also prevents those calls from being treated as convergent. In other words, those calls are not restricted to only being moved to control-equivalent blocks.

In languages following SPMD/SIMT programming model, e.g., CUDA/HIP, function declarations and 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.

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

__device__ int example(void) {
  float x;
  [[clang::noconvergent]] x = bar(x);
}

nodiscard, warn_unused_result

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

warn_unused_result

nodiscard
clang::warn_unused_result
gnu::warn_unused_result

nodiscard
gnu::warn_unused_result

Yes

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.

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.

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
}

noduplicate

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

noduplicate

clang::noduplicate

clang::noduplicate

Yes

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:

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:

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

where the call to “nodupfunc” is duplicated and sunk into the two branches of the condition.

noinline

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

noinline

gnu::noinline
clang::noinline
msvc::noinline

gnu::noinline
clang::noinline
msvc::noinline

noinline

__noinline__

Yes

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)).

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

nomicromips

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nomicromips

gnu::nomicromips

gnu::nomicromips

Yes

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.

noreturn, _Noreturn

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

noreturn

noreturn
_Noreturn

Yes

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>.

not_tail_called

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

not_tail_called

clang::not_tail_called

clang::not_tail_called

Yes

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:

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:

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:

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.
}

nothrow

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nothrow

gnu::nothrow

gnu::nothrow

nothrow

Yes

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.

nouwtable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nouwtable

clang::nouwtable

clang::nouwtable

Yes

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.

ns_consumed

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ns_consumed

clang::ns_consumed

clang::ns_consumed

Yes

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.

ns_consumes_self

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ns_consumes_self

clang::ns_consumes_self

clang::ns_consumes_self

Yes

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.

ns_returns_autoreleased

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ns_returns_autoreleased

clang::ns_returns_autoreleased

clang::ns_returns_autoreleased

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.

ns_returns_not_retained

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ns_returns_not_retained

clang::ns_returns_not_retained

clang::ns_returns_not_retained

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.

ns_returns_retained

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

ns_returns_retained

clang::ns_returns_retained

clang::ns_returns_retained

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.

numthreads

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

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

objc_method_family

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_method_family

clang::objc_method_family

clang::objc_method_family

Yes

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:

- (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).

objc_requires_super

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

objc_requires_super

clang::objc_requires_super

clang::objc_requires_super

Yes

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:

- (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:

- (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:

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

optnone

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

optnone

clang::optnone

clang::optnone

Yes

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.

os_consumed

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

os_consumed

clang::os_consumed

clang::os_consumed

Yes

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.

os_consumes_this

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

os_consumes_this

clang::os_consumes_this

clang::os_consumes_this

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.

os_returns_not_retained

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

os_returns_not_retained

clang::os_returns_not_retained

clang::os_returns_not_retained

Yes

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.

os_returns_retained

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

os_returns_retained

clang::os_returns_retained

clang::os_returns_retained

Yes

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.

os_returns_retained_on_non_zero

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

os_returns_retained_on_non_zero

clang::os_returns_retained_on_non_zero

clang::os_returns_retained_on_non_zero

Yes

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.

os_returns_retained_on_zero

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

os_returns_retained_on_zero

clang::os_returns_retained_on_zero

clang::os_returns_retained_on_zero

Yes

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.

overloadable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

overloadable

clang::overloadable

clang::overloadable

Yes

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:

#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.,

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:

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:

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:

// 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).

packoffset

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

packoffset

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:

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

patchable_function_entry

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

patchable_function_entry

gnu::patchable_function_entry

gnu::patchable_function_entry

Yes

__attribute__((patchable_function_entry(N,M))) is used to generate M NOPs before the function entry and N-M NOPs after the function entry. This attribute takes precedence over the command line option -fpatchable-function-entry=N,M. M defaults to 0 if omitted.

This attribute is only supported on aarch64/aarch64-be/loongarch32/loongarch64/riscv32/riscv64/i386/x86-64/ppc/ppc64 targets. For ppc/ppc64 targets, AIX is still not supported.

preserve_access_index

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preserve_access_index

clang::preserve_access_index

clang::preserve_access_index

Yes

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().

preserve_static_offset

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

preserve_static_offset

clang::preserve_static_offset

clang::preserve_static_offset

Yes

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:

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;.

register

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

register

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:

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

reinitializes

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

reinitializes

clang::reinitializes

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:

template <class T>
class Container {
public:
  ...
  [[clang::reinitializes]] void Clear();
  ...
};

release_capability, release_shared_capability

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

release_capability
release_shared_capability
release_generic_capability
unlock_function

clang::release_capability
clang::release_shared_capability
clang::release_generic_capability
clang::unlock_function

Marks a function as releasing a capability.

retain

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

retain

gnu::retain

gnu::retain

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.

shader

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

The shader type attribute applies to HLSL shader entry functions to identify the shader type for the entry function. The syntax is:

``[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.

short_call, near

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

short_call
near

gnu::short_call
gnu::near

gnu::short_call
gnu::near

Yes

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.

signal

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

signal

gnu::signal

gnu::signal

Yes

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.

speculative_load_hardening

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

speculative_load_hardening

clang::speculative_load_hardening

clang::speculative_load_hardening

Yes

This attribute can be applied to a function declaration in order to indicate

that Speculative Load Hardening 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 -mno-speculative-load-hardening is specified.

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.

strict_gs_check

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

strict_gs_check

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.

__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.

sycl_kernel

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

sycl_kernel

clang::sycl_kernel

clang::sycl_kernel

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:

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:

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.

sycl_kernel_entry_point

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

sycl_kernel_entry_point

clang::sycl_kernel_entry_point

clang::sycl_kernel_entry_point

Yes

The sycl_kernel_entry_point attribute facilitates 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 specifies a class type 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 void return type.

  • Is not a non-static member function, constructor, or destructor.

  • Is not a C variadic function.

  • Is not a coroutine.

  • Is not defined as deleted or as defaulted.

  • 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.

namespace sycl {
class handler {
  template<typename KernelNameType, typename KernelType>
  [[ clang::sycl_kernel_entry_point(KernelNameType) ]]
  static void kernel_entry_point(KernelType kernel) {
    kernel();
  }

public:
  template<typename KernelNameType, typename KernelType>
  void single_task(KernelType kernel) {
    // Call kernel_entry_point() to trigger generation of an offload
    // kernel entry point.
    kernel_entry_point<KernelNameType>(kernel);
    // Call functions appropriate for the desired offload backend
    // (OpenCL, CUDA, HIP, Level Zero, etc...).
  }
};
} // namespace sycl

A SYCL kernel 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. A SYCL kernel invocation function is responsible for copying the provided SYCL kernel object to an offload device and initiating a call to it. The SYCL kernel object and its data members constitute the parameters of an offload kernel.

A SYCL kernel 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 be copied to device memory as if by memcpy(). Additionally, some offload backends, OpenCL for example, require objects of some of these types to be passed as individual arguments to the offload kernel.

An offload kernel consists of an entry point function that declares the parameters of the offload kernel and the set of all functions and variables that are directly or indirectly used by the entry point function.

A SYCL kernel invocation function invokes a SYCL kernel on a device by 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. Deconstructing the SYCL kernel object, if necessary, to produce the set of offload kernel arguments required by the offload kernel entry point.

  3. Copying the offload kernel arguments to device memory.

  4. Initiating execution of the offload kernel entry point.

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

  1. Reconstituting the SYCL kernel object, if necessary, using the offload kernel parameters.

  2. Calling the operator() member function of the (reconstituted) SYCL kernel object.

The sycl_kernel_entry_point attribute automates generation of an offload kernel entry point that performs those latter tasks. The parameters and body of a function declared with the sycl_kernel_entry_point attribute specify a pattern from which the parameters and body of the entry point function are derived. Consider the following call to a SYCL kernel invocation function.

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. It has two data members corresponding to the captures of sout and s. Since one of these data members corresponds to a special SYCL type that must be passed individually as an offload kernel parameter, it is necessary to decompose the SYCL kernel object into its constituent parts; the offload kernel will have two kernel parameters. Given a SYCL implementation that uses a sycl_kernel_entry_point attributed function like the one shown above, an offload kernel entry point function will be generated that looks approximately as follows.

void sycl-kernel-caller-for-KN(sycl::stream sout, S s) {
  kernel-type kernel = { sout, s );
  kernel();
}

There are a few items worthy of note:

  1. The name of the generated function incorporates the SYCL kernel name, KN, that was passed as the KernelNameType template parameter to kernel_entry_point() and provided as the argument to the sycl_kernel_entry_point attribute. 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 kernel_entry_point(). Lambda types cannot be declared and initialized using the aggregate initialization syntax used above, but the intended behavior should be clear.

  3. S is a device copyable type that does not directly or indirectly contain a data member of a SYCL special type. It therefore does not need to be decomposed into its constituent members to be passed as a kernel argument.

  4. The depiction of the sycl::stream parameter as a single self contained kernel parameter is an oversimplification. SYCL special types may require additional decomposition such that the generated function might have three or more parameters depending on how the SYCL library implementation defines these types.

  5. The call to kernel_entry_point() has no effect other than to trigger emission of the entry point function. The statments that make up the body of the function are not executed when the function is called; they are only used in the generation of the entry point function.

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.

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. Function parameter decomposition and reconstitution is performed for all function parameters. The function must abide by the language feature restrictions described in section 5.4, “Language restrictions for device functions” in the SYCL 2020 specification.

target

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

target

gnu::target

gnu::target

Yes

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:

__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.

target_clones

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

target_clones

gnu::target_clones

gnu::target_clones

Yes

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:

__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:

__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.

target_version

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

target_version

gnu::target_version

gnu::target_version

Yes

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.

try_acquire_capability, try_acquire_shared_capability

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

try_acquire_capability
try_acquire_shared_capability

clang::try_acquire_capability
clang::try_acquire_shared_capability

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).

unsafe_buffer_usage

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

unsafe_buffer_usage

clang::unsafe_buffer_usage

clang::unsafe_buffer_usage

Yes

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 does not suppress -Wunsafe-buffer-usage inside the function to which it is attached. These warnings still need to be addressed.

    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:

    [[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:

    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:

    [[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.

    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.

used

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

used

gnu::used

gnu::used

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.

xray_always_instrument, xray_never_instrument, xray_log_args

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

xray_always_instrument
xray_never_instrument

clang::xray_always_instrument
clang::xray_never_instrument

clang::xray_always_instrument
clang::xray_never_instrument

Yes

__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.

xray_always_instrument, xray_never_instrument, xray_log_args

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

xray_log_args

clang::xray_log_args

clang::xray_log_args

Yes

__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.

zero_call_used_regs

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

zero_call_used_regs

gnu::zero_call_used_regs

gnu::zero_call_used_regs

Yes

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 (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.

Handle Attributes

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.

acquire_handle

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

acquire_handle

clang::acquire_handle

clang::acquire_handle

Yes

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.

// 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")));

release_handle

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

release_handle

clang::release_handle

clang::release_handle

Yes

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.

zx_status_t zx_handle_close(zx_handle_t handle [[clang::release_handle("tag")]]);

use_handle

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

use_handle

clang::use_handle

clang::use_handle

Yes

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.

zx_status_t zx_port_wait(zx_handle_t handle [[clang::use_handle("zircon")]],
                         zx_time_t deadline,
                         zx_port_packet_t* packet);

Nullability Attributes

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:

// 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:

@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.

_Nonnull

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

_Nonnull

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:

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.

_Null_unspecified

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

_Null_unspecified

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.

_Nullable

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

_Nullable

The _Nullable nullability qualifier indicates that a value of the _Nullable pointer type can be null. For example, given:

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:

class _Nullable ArenaPointer { ... };

ArenaPointer _Nonnull x = ...;
ArenaPointer _Nullable y = nullptr;

_Nullable_result

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

_Nullable_result

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:

-(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.

nonnull

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nonnull

gnu::nonnull

gnu::nonnull

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) 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:

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:

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:

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.

returns_nonnull

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

returns_nonnull

gnu::returns_nonnull

gnu::returns_nonnull

Yes

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:

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

OpenCL Address Spaces

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).

__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.

[[clang::opencl_global_device]], [[clang::opencl_global_host]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_global_device

clang::opencl_global_device

clang::opencl_global_device

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:

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”).

[[clang::opencl_global_device]], [[clang::opencl_global_host]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_global_host

clang::opencl_global_host

clang::opencl_global_host

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:

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”).

__constant, constant, [[clang::opencl_constant]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_constant

clang::opencl_constant

clang::opencl_constant

__constant
constant

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.

__generic, generic, [[clang::opencl_generic]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_generic

clang::opencl_generic

clang::opencl_generic

__generic
generic

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.

__global, global, [[clang::opencl_global]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_global

clang::opencl_global

clang::opencl_global

__global
global

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.

__local, local, [[clang::opencl_local]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_local

clang::opencl_local

clang::opencl_local

__local
local

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.

__private, private, [[clang::opencl_private]]

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

opencl_private

clang::opencl_private

clang::opencl_private

__private
private

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.

Performance Constraint Attributes

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.

allocating

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

allocating

clang::allocating

clang::allocating

Declares that a function potentially allocates heap memory, and prevents any potential inference of nonallocating by the compiler.

blocking

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

blocking

clang::blocking

clang::blocking

Declares that a function potentially blocks, and prevents any potential inference of nonblocking by the compiler.

nonallocating

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nonallocating

clang::nonallocating

clang::nonallocating

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.

nonblocking

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

nonblocking

clang::nonblocking

clang::nonblocking

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.

Statement Attributes

#pragma clang loop

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

`` loop``
unroll
nounroll
unroll_and_jam
nounroll_and_jam

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 language extensions for details.

#pragma unroll, #pragma nounroll

Supported Syntaxes

GNU

C++11

C23

__declspec

Keyword

#pragma

HLSL Annotation

#pragma clang attribute

`` loop``
unroll
nounroll
unroll_and_jam
nounroll_and_jam

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 loo