Add hipstdpar support to BabelStream #195
Conversation
src/std-indices/model.cmake
Outdated
| @@ -50,4 +66,13 @@ macro(setup) | |||
| register_definitions(USE_ONEDPL) | |||
| register_link_library(oneDPL) | |||
| endif () | |||
| if (CLANG_OFFLOAD) | |||
There was a problem hiding this comment.
I'm not totally sold on the name of this definition as CLANG is not specific to AMD, but the CLANG_FLAGS set below are definitely specific. Can we rename this please to something appropriate?
There was a problem hiding this comment.
Would replacing CLANG_OFFLOAD with AMDGPU_TARGET_OFFLOAD and CLANG_FLAGS with AMDGPU_TARGET_OFFLOAD_FLAGS be acceptable?
There was a problem hiding this comment.
Hi @tomdeakin, please confirm if the above suggestion would work or recommend something better. I can make the changes accordingly.
|
It's great to see hipstdpar working, so let's work to get this merged in. Thanks for the contributions. |
Change definition names to be more specific to AMD GPUs
|
Hi @tomdeakin, @afanfa and I have made the changes requested. Please check and approve if everything looks okay. |
|
Added this PR with some fixes to #202 . |
This PR adds support for offload to AMD GPUs using the
par_unseqexecution policy in C++ standard parallelism algorithms. To trigger the GPU offload of all parallel algorithms, the--hipstdparcompilation flag must be provided. For GPU targets other than the current default ofgfx906, the--offload-arch=<arch_string>option must also be provided at compile time.When using ROCm 6.1.0, the compilation commands may look like the following if compiling for an AMD Instinct MI200 series GPU:
Remember to set the environment variable to enable address translation and page migration (where applicable) when running
std-data-streamorstd-indices-stream: