[PGO] no code state update on dynamic=False #155961
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/155961
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 662ff18 with merge base 54976bc ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This pull request was exported from Phabricator. Differential Revision: D76630499 |
facebook-github-bot
force-pushed
the
export-D76630499
branch
from
4b1c793 to
3efbc7b
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76630499 |
pianpwk
force-pushed
the
export-D76630499
branch
from
3efbc7b to
62f8082
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76630499 |
pianpwk
force-pushed
the
export-D76630499
branch
from
62f8082 to
4f00810
Compare
Summary: When tensor size changes are detected on `dynamic=False`, overwrites the PGO state with the newest static shapes to reflect the latest frame state, instead of updating automatic dynamic. A longer term solution, if we move to shared PGO state between multiple jobs, would be to update automatic dynamic, but avoid suggesting/logging the whitelist (compiling with `dynamic=False` should already override any dynamic PGO that's read, so we're fine there). This way if any particular job runs with `dynamic=False`, it won't statically overwrite the entire PGO state if it's shared with many other jobs. Test Plan: test/dynamo/test_pgo.py Rollback Plan: Reviewed By: bobrenjc93 Differential Revision: D76630499
facebook-github-bot
force-pushed
the
export-D76630499
branch
from
4f00810 to
3dcdf3f
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76630499 |
pianpwk
force-pushed
the
export-D76630499
branch
from
3dcdf3f to
1d3a931
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76630499 |
pianpwk
force-pushed
the
export-D76630499
branch
from
1d3a931 to
ef8540e
Compare
Summary: When tensor size changes are detected on `dynamic=False`, overwrites the PGO state with the newest static shapes to reflect the latest frame state, instead of updating automatic dynamic. A longer term solution, if we move to shared PGO state between multiple jobs, would be to update automatic dynamic, but avoid suggesting/logging the whitelist (compiling with `dynamic=False` should already override any dynamic PGO that's read, so we're fine there). This way if any particular job runs with `dynamic=False`, it won't statically overwrite the entire PGO state if it's shared with many other jobs. Test Plan: test/dynamo/test_pgo.py Rollback Plan: Reviewed By: bobrenjc93 Differential Revision: D76630499
facebook-github-bot
force-pushed
the
export-D76630499
branch
from
ef8540e to
662ff18
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76630499 |
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
Merge failedReason: This PR has internal changes and must be landed via Phabricator! Please try reimporting/rexporting the PR! Details for Dev Infra teamRaised by workflow job |
|
@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: 1 jobs have failed, first few of them are: Meta Internal-Only Changes Check Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge -i |
Merge startedYour change will be merged while ignoring the following 1 checks: Meta Internal-Only Changes Check Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
4 participants
Summary:
When tensor size changes are detected on
dynamic=False, overwrites the PGO state with the newest static shapes to reflect the latest frame state, instead of updating automatic dynamic.A longer term solution, if we move to shared PGO state between multiple jobs, would be to update automatic dynamic, but avoid suggesting/logging the whitelist (compiling with
dynamic=Falseshould already override any dynamic PGO that's read, so we're fine there). This way if any particular job runs withdynamic=False, it won't statically overwrite the entire PGO state if it's shared with many other jobs.Test Plan:
test/dynamo/test_pgo.py
Rollback Plan:
Differential Revisi,on: D76630499
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames