aarch64: Remove computation cache from conv, matmul, inner product #357
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
nikhil-arm
reviewed
Aug 29, 2025
| @@ -35,7 +35,7 @@ struct inner_product_forward_params { | |||
| struct inner_product_forward | |||
| : public dnnl::inner_product_forward, | |||
| #ifdef __aarch64__ | |||
There was a problem hiding this comment.
Will we ever hit a case when oneDNN is not built with ACL on Arm platform and this code would crash. Same for other instances where this is used
nikhil-arm
reviewed
Aug 29, 2025
include/ideep/lru_cache.hpp
Outdated
| class dummy_computation_cache { | ||
| public: | ||
| static inline dummy_lru t_store() { | ||
| return dummy_lru(); |
There was a problem hiding this comment.
Returning by value? Can we verify that no code is binding by ref or expects similar semantic to real cache.
Suggested change
| return dummy_lru(); | |
| static dummy_lru store; | |
| return store; |
nikhil-arm
reviewed
Aug 29, 2025
| param.pd = primitive_desc( | ||
| aengine, src_desc, weights.get_desc(), dst_desc, op_attr); | ||
| } | ||
| param.primitive = std::move(super(param.pd)); |
There was a problem hiding this comment.
is move redundant on temporary data? I am doubtful of my comment and needs suggestions from author
conv, matmul and inner product have been converted to being stateless in ACL, thus there is no longer a need for the computation cache. This has been replaced with a placeholder cache as the cache is still expected to exist. The placeholder cache can be removed once deconv is also converted to stateless. No change in performance.
Ryo-not-rio
force-pushed
the
remove-cache
branch
from
b1eb174 to
58508ec
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
conv, matmul and inner product have been converted to being stateless in ACL, thus there is no longer a need for the computation cache. This has been replaced with a dummy cache as the cache is still expected to exist. The dummy cache can be removed once deconv is also converted to stateless.