PR #21022: [XLA:CPU][oneDNN] Modify addend shape for small Convolution + Bias contraction #3080
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR #21022: [XLA:CPU][oneDNN] Modify addend shape for small Convolution + Bias contraction
Imported from GitHub PR openxla/xla#21022
For small constant biases, XLA may perform constant folding, which can alter the shape of the bias passed to oneDNN. This PR removes any extraneous trivial dimensions from the bias that is passed to the oneDNN library. It also adds a test in all applicable dtypes to test the functionality.
Copybara import of the project:
--
5ce5ba690d5acd1ebedca1484483639b2f3b0be1 by Akhil Goel [email protected]:
Modify conv bias shape
Merging this change closes #21022
FUTURE_COPYBARA_INTEGRATE_REVIEW=openxla/xla#21022 from Intel-tensorflow:akhil/conv_bias_fix 5ce5ba690d5acd1ebedca1484483639b2f3b0be1