[CI] Run build on Ubuntu 22, pre-commit if CUDA adapter changes. #17757
Conversation
uditagarwal97
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
uditagarwal97
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
uditagarwal97
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
| @@ -52,6 +52,22 @@ jobs: | |||
| changes: ${{ needs.detect_changes.outputs.filters }} | |||
| e2e_binaries_artifact: sycl_e2e_bin_default | |||
|
|
|||
| # If a PR changes CUDA adapter, run the build on Ubuntu 22.04 as well. | |||
| # Ubuntu 22.04 container has CUDA 12.1 installed while Ubuntu 24.0 image | |||
| # has CUDA 12.6.1 installed. | |||
There was a problem hiding this comment.
If I get it right, patches with changes in CUDA adaptor will be tested with both CUDA 12.6.1 and CUDA 12.1 (at least build + LIT?).
In addition, I saw CUDA 12.8 is installed on (at least) one of Windows runners.
Do intentionally use different CUDA versions in different workflows (or different machines)?
@npmiller, FYI.
There was a problem hiding this comment.
If I get it right, patches with changes in CUDA adaptor will be tested with both CUDA 12.6.1 and CUDA 12.1 (at least build + LIT?).
Yes, for Linux.
In addition, I saw CUDA 12.8 is installed on (at least) one of Windows runners.
Do intentionally use different CUDA versions in different workflows (or different machines)?
@sarnex Since you installed CUDA on Windows runners, do we plan to have CUDA 12.8 uniformly on all Windows runners? CodePlay's UR CUDA adapter workflow uses CUDA 12.4, so I'm not sure if CUDA 12.8 is even extensively tested
There was a problem hiding this comment.
yeah im installing cuda 12.8 on all windows runners, but we dont have any nvidia windows gpu machines, so its basically build only testing
|
I'm seeing a fail that I think might be related to this change, this error state https://github.com/intel/llvm/actions/runs/14263534349/job/39981394795?pr=17571#step:17:323 is reached based on the definition of |
We decided that in this circumstance, it would be simplest to warn the user that the property will not be used instead of throwing an error. See this PR which changes to a warning: #17863 |
…7863) Warn users over setting the memory pool maximum size property instead of error. This should aid CI which is affected by the memory_pool.cpp e2e test which expects to be able to set this property despite DPC++ being built with a CUDA version that is too old. See this CI error: #17757 (comment) This PR relies on #17095 to be merged for the printing of the warning to be enabled.
Warn users over setting the memory pool maximum size property instead of error. This should aid CI which is affected by the memory_pool.cpp e2e test which expects to be able to set this property despite DPC++ being built with a CUDA version that is too old. See this CI error: intel/llvm#17757 (comment) This PR relies on intel/llvm#17095 to be merged for the printing of the warning to be enabled.
Our Ubuntu 22.04 container has CUDA 12.1 installed while Ubuntu 24.04 image has CUDA 12.6.1 installed.
If a PR changes CUDA adapter, we should test the change with both CUDA versions.