[Driver][SYCL][NewOffload] Fix arch settings for nvptx and amd #14340
Conversation
When compiling for -fsycl-targets values of nvptx64-nvidia-cuda and amdgcn-amd-gpu, the default arch behaviors were not applied to the compilation. Updates to do the following: - Add default of sm_50 for nvptx64 if not provided - Emit diagnostic if no arch provided for amd - Parse -Xsycl-backend-target for offload-arch values
mdtoguchi
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
asudarsa
left a comment
asudarsa
left a comment
There was a problem hiding this comment.
I tested the changes for the new offloading model on several failing SYCL E2E tests on an AMD machine and the fix resolves all of those failures. Test added looks fine.
Thanks much for the fix.
|
@intel/llvm-gatekeepers, This should be ready for merge - please take a look. |
|
We should really be not running CI testing for comment updates...:-( |
|
@mdtoguchi this is awaiting approval from @srividya-sundaram |
|
We have approval from a driver team reviewer, This change unblocks development by fixing a large number of test failures. So, I ill merge it now. @srividya-sundaram, please feel free to reach out to me or @mdtoguchi for any post-commit changes if required. Thanks |
|
Sorry, I missed the notification for approval request. |
5 participants
When compiling for -fsycl-targets values of nvptx64-nvidia-cuda and amdgcn-amd-gpu, the default arch behaviors were not applied to the compilation. Updates to do the following: