[SYCL][Driver][HIP] Do not define __CUDA_ARCH__ for HIP-AMD targets #15443
Conversation
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
|
Requires #15521 to get merged to rebase on those changes for the |
GeorgeWeb
force-pushed
the
georgi/hip-undef-cuda-arch
branch
from
95a29e6 to
7062717
Compare
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
force-pushed
the
georgi/hip-undef-cuda-arch
branch
from
84bff0a to
d31852f
Compare
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
force-pushed
the
georgi/hip-undef-cuda-arch
branch
from
d31852f to
45b86ef
Compare
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
GeorgeWeb
force-pushed
the
georgi/hip-undef-cuda-arch
branch
from
6c5e501 to
bf72790
Compare
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
| @@ -309,7 +309,6 @@ void AMDGPUTargetInfo::getTargetDefines(const LangOptions &Opts, | |||
| Twine("\"") + Twine(CanonName) + Twine("\"")); | |||
| Builder.defineMacro("__amdgcn_target_id__", | |||
| Twine("\"") + Twine(*getTargetID()) + Twine("\"")); | |||
| Builder.defineMacro("__CUDA_ARCH__", "0"); | |||
There was a problem hiding this comment.
NOTE:
This was previously added to avoid setting it to a non-zero value but is not needed anymore after the changes in this PR completely disable the definition of the macro for SYCL offload. Additionally, the upstream llvm-project doesn't have this line https://github.com/llvm/llvm-project/blob/main/clang/lib/Basic/Targets/AMDGPU.cpp#L311.
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
GeorgeWeb
had a problem deploying
to
WindowsCILock
GitHub Actions
Error
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
|
The Update: Merged latest sycl branch that includes the above fix so all is passing now. |
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
GeorgeWeb
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
|
@intel/llvm-gatekeepers This should be good merge now. Thanks! |
|
@GeorgeWeb The AMD runner is not doing too hot so we won't have E2E HIP testing for a bit (I'm working on it ASAP), are you confident this is sufficiently tested in compile-time tests? Thanks |
Good reminder. I forgot that was the reason I didn't ping for merging on Friday. I've run local testing last week but I'd rather it pass the precommit here. It's not a patch of high importance at the moment. |
|
@GeorgeWeb We don't have plans do to HIP testing in precommit for a while because the runner is in such bad shape, but we will have it in post commit when I can fix the runner. If you tested locally and are confident it should work I'm happy to merge the patch and wait and see, just wanted to check your confidence level. |
|
I fixed the HIP AMD runner so I'll merge this now |
This commit updates NVIDIA CCCL to version 2.7.0 and starts following the official repository rather than my fork. In order to make this work, I had to incorporate a temporary workaround for intel/llvm#15544 until we start using a release with intel/llvm#15443. Closes acts-project#660.
Fixes #15544