[Inductor Cutlass backend] Set INDUCTOR_TEST_DISABLE_FRESH_CACHE in test setup #124574
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/124574
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (3 Unrelated Failures)As of commit 69dcc1d with merge base failed to retrieve merge base, please contact dev infra: FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
| # interacts badly with persistent subprocesses during | ||
| # autotuning. So we need to disable automatic cache refresh | ||
| # before calling setUp() on the parent class. | ||
| os.environ["INDUCTOR_TEST_DISABLE_FRESH_CACHE"] = "1" |
There was a problem hiding this comment.
Should we revert this after the call to super().setUp()? Otherwise, the env var will stick around and may affect other tests running in the same process in CI?
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: 3 mandatory check(s) failed. The first few are:
Dig deeper by viewing the failures on hud |
|
@pytorchbot merge -f "Failing profiler tests are known to be broken" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
…emm_choices (#124575) Clean up CUTLASSGemmTemplate.add_cutlass_gemm_choices, removing code that became unneccessary by removing EVT-based epilogue fusion. Test Plan: Already covered by test_cutlass_backend.py Pull Request resolved: #124575 Approved by: https://github.com/jansel ghstack dependencies: #121497, #123930, #123932, #121734, #124107, #124574
…est setup (pytorch#124574) The diff pytorch#122661 introduces a new automatic cache refresh mechanism during all inductor-derived test cases. But this refresh mechanism seems not to work properly across process boundaries, specifically when using autotune_in_subproc, which many tests in test_cutlass_backend.py rely on. Solution: Set the env var INDUCTOR_TEST_DISABLE_FRESH_CACHE=1 early during test setup within test_cutlass_backend.py Test Plan: This is a change to unit tests only. Pull Request resolved: pytorch#124574 Approved by: https://github.com/aakhundov ghstack dependencies: pytorch#121497, pytorch#123930, pytorch#123932, pytorch#121734, pytorch#124107
…emm_choices (pytorch#124575) Clean up CUTLASSGemmTemplate.add_cutlass_gemm_choices, removing code that became unneccessary by removing EVT-based epilogue fusion. Test Plan: Already covered by test_cutlass_backend.py Pull Request resolved: pytorch#124575 Approved by: https://github.com/jansel ghstack dependencies: pytorch#121497, pytorch#123930, pytorch#123932, pytorch#121734, pytorch#124107, pytorch#124574
3 participants
Stack from ghstack (oldest at bottom):
The diff #122661 introduces a new automatic cache refresh mechanism during all inductor-derived test cases.
But this refresh mechanism seems not to work properly across process boundaries, specifically when using autotune_in_subproc, which many tests in test_cutlass_backend.py rely on.
Solution: Set the env var INDUCTOR_TEST_DISABLE_FRESH_CACHE=1
early during test setup within test_cutlass_backend.py
Test Plan:
This is a change to unit tests only.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @muchulee8 @ColinPeppler @amjames @desertfire @chauhang