-
Notifications
You must be signed in to change notification settings - Fork 748
Arm backend: test without output re-order workaround #15826
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
By default outputs are re-ordered to correct order during TOSA lowering. However this is seen as a workaround as it should not be needed. Furthermore the output issue is not easily reproduced, rather it seems to happen randomly. Therefore we add a test case without the workaround, which is currently passing. In case it won't pass without the workaround at some point, the new changes might give some hints on why the workaround is needed and how to fix it. In case it continues to pass, we may switch the default and potentially even remove the workaround. Change-Id: I4c65cfaebe0a662d43c339c64a370b308674d1b5
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15826
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (4 Unrelated Failures)As of commit fe878a4 with merge base 3374ff8 ( BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
Timeout problem trunk / test-arm-backend (test_pytest_ops_ethosu_fvp) / linux-job (push)Cancelled after 121m |
|
Seem to be happening after #14109 was merged, trying to revert that PR |
|
See comment for more context of workaround: #13454 (comment) |
By default outputs are re-ordered to correct order during TOSA lowering. However this is seen as a workaround as it should not be needed. Furthermore the output issue is not easily reproduced, rather it seems to happen randomly. Therefore we add a test case without the workaround, which is currently passing.
In case it won't pass without the workaround at some point, the new changes might give some hints on why the workaround is needed and how to fix it.
In case it continues to pass, we may switch the default and potentially even remove the workaround.
cc @freddan80 @per @zingo @oscarandersson8218 @digantdesai