OpenCL Support

Clang has complete support of OpenCL C versions from 1.0 to 3.0. Support for OpenCL 3.0 is in experimental phase (OpenCL 3.0).

Clang also supports the C++ for OpenCL kernel language.

There are also other new and experimental features available.

Details about usage of clang for OpenCL can be found in Clang Compiler User’s Manual.

Missing features or with limited support

Internals Manual

This section acts as internal documentation for OpenCL features design as well as some important implementation aspects. It is primarily targeted at the advanced users and the toolchain developers integrating frontend functionality as a component.

OpenCL Metadata

Clang uses metadata to provide additional OpenCL semantics in IR needed for backends and OpenCL runtime.

Each kernel will have function metadata attached to it, specifying the arguments. Kernel argument metadata is used to provide source level information for querying at runtime, for example using the clGetKernelArgInfo call.

Note that -cl-kernel-arg-info enables more information about the original kernel code to be added e.g. kernel parameter names will appear in the OpenCL metadata along with other information.

The IDs used to encode the OpenCL’s logical address spaces in the argument info metadata follows the SPIR address space mapping as defined in the SPIR specification section 2.2

OpenCL Specific Options

In addition to the options described in Clang Compiler User’s Manual there are the following options specific to the OpenCL frontend.

All the options in this section are frontend-only and therefore if used with regular clang driver they require frontend forwarding, e.g. -cc1 or -Xclang.


Adds most of builtin types and function declarations during compilations. By default the OpenCL headers are not loaded by the frontend and therefore certain builtin types and most of builtin functions are not declared. To load them automatically this flag can be passed to the frontend (see also the section on the OpenCL Header):

$ clang -Xclang -finclude-default-header

Alternatively the internal header opencl-c.h containing the declarations can be included manually using -include or -I followed by the path to the header location. The header can be found in the clang source tree or installation directory.

$ clang -I<path to clang sources>/lib/Headers/opencl-c.h
$ clang -I<path to clang installation>/lib/clang/<llvm version>/include/opencl-c.h/opencl-c.h

In this example it is assumed that the kernel code contains #include <opencl-c.h> just as a regular C include.

Because the header is very large and long to parse, PCH (Precompiled Header and Modules Internals) and modules (Modules) can be used internally to improve the compilation speed.

To enable modules for OpenCL:

$ clang --target=spir-unknown-unknown -c -emit-llvm -Xclang -finclude-default-header -fmodules -fimplicit-module-maps -fmodules-cache-path=<path to the generated module>

Another way to circumvent long parsing latency for the OpenCL builtin declarations is to use mechanism enabled by -fdeclare-opencl-builtins flag that is available as an alternative feature.


In addition to regular header includes with builtin types and functions using -finclude-default-header, clang supports a fast mechanism to declare builtin functions with -fdeclare-opencl-builtins. This does not declare the builtin types and therefore it has to be used in combination with -finclude-default-header if full functionality is required.

Example of Use:

$ clang -Xclang -fdeclare-opencl-builtins

Overrides the target address space map with a fake map. This allows adding explicit address space IDs to the bitcode for non-segmented memory architectures that do not have separate IDs for each of the OpenCL logical address spaces by default. Passing -ffake-address-space-map will add/override address spaces of the target compiled for with the following values: 1-global, 2-constant, 3-local, 4-generic. The private address space is represented by the absence of an address space attribute in the IR (see also the section on the address space attribute).

$ clang -cc1 -ffake-address-space-map

OpenCL builtins

Clang builtins

There are some standard OpenCL functions that are implemented as Clang builtins:

Fast builtin function declarations

The implementation of the fast builtin function declarations (available via the -fdeclare-opencl-builtins option) consists of the following main components:

  • A TableGen definitions file This contains a compact representation of the supported builtin functions. When adding new builtin function declarations, this is normally the only file that needs modifying.

  • A Clang TableGen emitter defined in ClangOpenCLBuiltinEmitter.cpp. During Clang build time, the emitter reads the TableGen definition file and generates This generated file contains various tables and functions that capture the builtin function data from the TableGen definitions in a compact manner.

  • OpenCL specific code in SemaLookup.cpp. When Sema::LookupBuiltin encounters a potential builtin function, it will check if the name corresponds to a valid OpenCL builtin function. If so, all overloads of the function are inserted using InsertOCLBuiltinDeclarationsFromTable and overload resolution takes place.

OpenCL Extensions and Features

Clang implements various extensions to OpenCL kernel languages.

New functionality is accepted as soon as the documentation is detailed to the level sufficient to be implemented. There should be an evidence that the extension is designed with implementation feasibility in consideration and assessment of complexity for C/C++ based compilers. Alternatively, the documentation can be accepted in a format of a draft that can be further refined during the implementation.

Implementation guidelines

This section explains how to extend clang with the new functionality.

Parsing functionality

If an extension modifies the standard parsing it needs to be added to the clang frontend source code. This also means that the associated macro indicating the presence of the extension should be added to clang.

The default flow for adding a new extension into the frontend is to modify OpenCLExtensions.def, containing the list of all extensions and optional features supported by the frontend.

This will add the macro automatically and also add a field in the target options clang::TargetOptions::OpenCLFeaturesMap to control the exposure of the new extension during the compilation.

Note that by default targets like SPIR-V, SPIR or X86 expose all the OpenCL extensions. For all other targets the configuration has to be made explicitly.

Note that the target extension support performed by clang can be overridden with -cl-ext command-line flags.

Library functionality

If an extension adds functionality that does not modify standard language parsing it should not require modifying anything other than header files and detailed in OpenCL builtins. Most commonly such extensions add functionality via libraries (by adding non-native types or functions) parsed regularly. Similar to other languages this is the most common way to add new functionality.

Clang has standard headers where new types and functions are being added, for more details refer to the section on the OpenCL Header. The macros indicating the presence of such extensions can be added in the standard header files conditioned on target specific predefined macros or/and language version predefined macros (see feature/extension preprocessor macros defined in opencl-c-base.h).


Some extensions alter standard parsing dynamically via pragmas.

Clang provides a mechanism to add the standard extension pragma OPENCL EXTENSION by setting a dedicated flag in the extension list entry of OpenCLExtensions.def. Note that there is no default behavior for the standard extension pragmas as it is not specified (for the standards up to and including version 3.0) in a sufficient level of detail and, therefore, there is no default functionality provided by clang.

Pragmas without detailed information of their behavior (e.g. an explanation of changes it triggers in the parsing) should not be added to clang. Moreover, the pragmas should provide useful functionality to the user. For example, such functionality should address a practical use case and not be redundant i.e. cannot be achieved using existing features.

Note that some legacy extensions (published prior to OpenCL 3.0) still provide some non-conformant functionality for pragmas e.g. add diagnostics on the use of types or functions. This functionality is not guaranteed to remain in future releases. However, any future changes should not affect backward compatibility.

Address spaces attribute

Clang has arbitrary address space support using the address_space(N) attribute, where N is an integer number in the range specified in the Clang source code. This addresses spaces can be used along with the OpenCL address spaces however when such addresses spaces converted to/from OpenCL address spaces the behavior is not governed by OpenCL specification.

An OpenCL implementation provides a list of standard address spaces using keywords: private, local, global, and generic. In the AST and in the IR each of the address spaces will be represented by unique number provided in the Clang source code. The specific IDs for an address space do not have to match between the AST and the IR. Typically in the AST address space numbers represent logical segments while in the IR they represent physical segments. Therefore, machines with flat memory segments can map all AST address space numbers to the same physical segment ID or skip address space attribute completely while generating the IR. However, if the address space information is needed by the IR passes e.g. to improve alias analysis, it is recommended to keep it and only lower to reflect physical memory segments in the late machine passes. The mapping between logical and target address spaces is specified in the Clang’s source code.

C++ for OpenCL Implementation Status

Clang implements language versions 1.0 and 2021 published in the official release of C++ for OpenCL Documentation.

Limited support of experimental C++ libraries is described in the experimental features.

GitHub issues for this functionality are typically prefixed with ‘[C++4OpenCL]’ - click here to view the full bug list.

Missing features or with limited support

OpenCL C 3.0 Usage

OpenCL C 3.0 language standard makes most OpenCL C 2.0 features optional. Optional functionality in OpenCL C 3.0 is indicated with the presence of feature-test macros (list of feature-test macros is here). Command-line flag -cl-ext can be used to override features supported by a target.

For cases when there is an associated extension for a specific feature (fp64 and 3d image writes) user should specify both (extension and feature) in command-line flag:

$ clang -cl-std=CL3.0 -cl-ext=+cl_khr_fp64,+__opencl_c_fp64 ...
$ clang -cl-std=CL3.0 -cl-ext=-cl_khr_fp64,-__opencl_c_fp64 ...

OpenCL C 3.0 Implementation Status

The following table provides an overview of features in OpenCL C 3.0 and their implementation status.





Command line interface

New value for -cl-std flag


Predefined macros

New version macro


Predefined macros

Feature macros


Feature optionality

Generic address space

done and

Feature optionality

Builtin function overloads with generic address space


Feature optionality

Program scope variables in global memory


Feature optionality

3D image writes including builtin functions

done (frontend)

Feature optionality

read_write images including builtin functions

done (frontend) and, (functions)

Feature optionality

C11 atomics memory scopes, ordering and builtin function


Feature optionality

Blocks and Device-side kernel enqueue including builtin functions


Feature optionality

Pipes including builtin functions

done (frontend) and (functions)

Feature optionality

Work group collective builtin functions


Feature optionality

Image types and builtin functions

done (frontend) and (functions)

Feature optionality

Double precision floating point type


New functionality

RGBA vector components


New functionality

Subgroup functions


New functionality

Atomic mem scopes: subgroup, all devices including functions


Experimental features

Clang provides the following new WIP features for the developers to experiment and provide early feedback or contribute with further improvements. Feel free to contact us on the Discourse forums (Clang Frontend category) or file a GitHub issue.

C++ libraries for OpenCL

There is ongoing work to support C++ standard libraries from LLVM’s libcxx in OpenCL kernel code using C++ for OpenCL mode.

It is currently possible to include type_traits from C++17 in the kernel sources when the following clang extensions are enabled __cl_clang_function_pointers and __cl_clang_variadic_functions, see Clang Language Extensions for more details. The use of non-conformant features enabled by the extensions does not expose non-conformant behavior beyond the compilation i.e. does not get generated in IR or binary. The extension only appear in metaprogramming mechanism to identify or verify the properties of types. This allows to provide the full C++ functionality without a loss of portability. To avoid unsafe use of the extensions it is recommended that the extensions are disabled directly after the header include.

Example of Use:

The example of kernel code with type_traits is illustrated here.

#pragma OPENCL EXTENSION __cl_clang_function_pointers : enable
#pragma OPENCL EXTENSION __cl_clang_variadic_functions : enable
#include <type_traits>
#pragma OPENCL EXTENSION __cl_clang_function_pointers : disable
#pragma OPENCL EXTENSION __cl_clang_variadic_functions : disable

using sint_type = std::make_signed<unsigned int>::type;

__kernel void foo() {
  static_assert(!std::is_same<sint_type, unsigned int>::value);

The possible clang invocation to compile the example is as follows:

$ clang -I<path to libcxx checkout or installation>/include test.clcpp

Note that type_traits is a header only library and therefore no extra linking step against the standard libraries is required. See full example in Compiler Explorer.

More OpenCL specific C++ library implementations built on top of libcxx are available in libclcxx project.