Skip to content
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

Fix Post From-Device Copy Processing Issue #1796

Merged
merged 3 commits into from
Jan 16, 2025
Merged

Conversation

ctodTT
Copy link
Contributor

@ctodTT ctodTT commented Jan 15, 2025

This change fixes an issue with processing output tensors after they are pulled off the device when checking golden-ness. The resulting tensor had the first 32 bytes overwritten with seemingly uninitialized memory, resulting in nondeterministic failures during similarity checks. Now, the unadulterated tensor is returned.

This change also fixes a minor bug relating to the binding of ceil, as well as includes a convenience change to .gitignore, ignoring top level .mlir files often used during debugging.
Fixes #1795

Copy link
Contributor

@vprajapati-tt vprajapati-tt left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ttir_builder related changes LGTM.

Copy link
Contributor

@nsmithtt nsmithtt left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks!

`ceil` was accidentally bound to `torch.log`. Now it is bound to
`torch.ceil` as expected
This change fixes an issue with processing output tensors after they are
pulled off the device when checking golden-ness. The resulting tensor
had the first 32 bytes overwritten with seemingly uninitialized memory,
resulting in nondeterministic failures during similarity checks. Now,
the unadulterated tensor is returned.

Fixes #1795
@ctodTT ctodTT force-pushed the ctod/fix-device-copy-cast branch from 0b18496 to e61cd66 Compare January 16, 2025 19:21
@ctodTT ctodTT enabled auto-merge (squash) January 16, 2025 19:21
@ctodTT ctodTT merged commit 9278f5d into main Jan 16, 2025
20 checks passed
@ctodTT ctodTT deleted the ctod/fix-device-copy-cast branch January 16, 2025 21:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

FIx Nondeterministic Failures of PCC
5 participants