Added exp FP32 FWD/BWD oneDNN kernel and optimized other oneDNN grad kernels #38624
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
| Thanks for your contribution! |
Aganlengzi reviewed Jan 4, 2022
Contributor Author
Contributor Author
| Hi @Aganlengzi, could you please continue your review? |
Member
| LGTM 👍 |
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
4 participants
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 Outdated suggestions cannot be applied.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 types
New features
PR changes
OPs
Describe
Added exp FP32 FWD/BWD oneDNN kernel and optimized other oneDNN grad kernels by allowing the usage of Out tensor instead of X tensor in some activation grad kernels. New version is working faster, because in some cases the computations are a lot simpler, f.e. for exp activation:
forward equation:
out = exp(x)
grad equation:
dx = dout * exp(x)
optimized grad equation(using out instead of x):
dx = dout * out
Simple multiplication is a lot faster than a multiplication followed by calculating the exponential function and same logic applies to other kernels that use "use_dst_for_bwd" versions of oneDNN kernels. Our new kernels are working up to 10% faster than old ones.