Enable split build feature on all workflows#1225
Conversation
| # available so we can't pip install -r requirements.txt. Keep | ||
| # this list in sync with OffloadTest/test/requirements.txt. | ||
| pip-install: lit pyyaml | ||
| - name: Install lit and pyyaml |
There was a problem hiding this comment.
From here I can verify that the sccache step is running as expected.
But it seems like this step isn't and it still uses the user path?
There was a problem hiding this comment.
Here is an example where this step runs:
https://github.com/llvm/offload-test-suite/actions/runs/26776253602/job/78937729286?pr=1225#step:8:138
I believe the linked pipeline you're looking at doesn't have the split build-feature enabled.
From what I can gather, the pipelines without my changes, triggered by PR matrix and configured as exec tests are run. And, the workflows I changed are individually run with the changes.
Execution Testing is the prefix for the unchanged runs, it looks like, and Windows is the prefix for the runs with my changes.
There was a problem hiding this comment.
I see thanks. Okay that makes more sense.
| # track hit rate / cache size across runs and diagnose cold-cache | ||
| # blowups on new builders. Always-runs so we still get stats on | ||
| # build failure. | ||
| - name: sccache stats |
There was a problem hiding this comment.
nit:
| - name: sccache stats | |
| - name: Dump sccache Stats |
The names seem to be verbs and capitalized
| # available so we can't pip install -r requirements.txt. Keep | ||
| # this list in sync with OffloadTest/test/requirements.txt. | ||
| pip-install: lit pyyaml | ||
| - name: Install lit and pyyaml |
There was a problem hiding this comment.
I see thanks. Okay that makes more sense.
…05/offload-test-suite into regularly_test_splitbuild
The infrastructure seems to be in place for github to split jobs into build and test jobs and distribute build jobs across compatible runners. This PR does a bit of quality improvement, but also actually enables the split build feature on all workflows.
Separately, this PR adds an sccache step so that we can track the hit rate, and ensures that clang is built in the build LLVM stage and not during the install step.
Assisted by: Github Copilot.
Manifests the implementation of: #1081