implement bufferization interfaces for lattigo#2976
Open
copybara-service[bot] wants to merge 1 commit into
Open
implement bufferization interfaces for lattigo#2976copybara-service[bot] wants to merge 1 commit into
copybara-service[bot] wants to merge 1 commit into
Conversation
Collaborator
|
@j2kun I'd love a brief walkthrough of this during the meeting tomorrow! |
copybara-service
Bot
force-pushed
the
test_918605118
branch
4 times, most recently
from
f7078aa to
c8f6057
Compare
This implements just enough bufferization to make MNIST work in a reasonable time frame. Handling the remaining ops will come in a followup PR. Before this change, MNIST with looped kernels would take 7 minutes per inference. Now it takes 3 minutes per inference. Fixing AllocToInPlace (#2975) will resolve the remaining latency issues PiperOrigin-RevId: 918605118
copybara-service
Bot
force-pushed
the
test_918605118
branch
from
c8f6057 to
b3e4695
Compare
Collaborator
|
@AlexanderViand if you feel like reviewing this one it is ready. I don't think there's anything groundbreaking going on, but the change from dot_product_8f to use linalg.dot is because the HECO passes are not strictly compatible with the current layout system (in particular, for dot_product they use a target slot of 7, when the layout assigned by the default layout pipeline treats the output as being in slot 0). I think we plan to deprecate those passes anyway, so that is not a major issue. |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
implement bufferization interfaces for lattigo
This implements just enough bufferization to make MNIST work in a reasonable
time frame. Handling the remaining ops will come in a followup PR.
Before this change, MNIST with looped kernels would take 7 minutes per
inference. Now it takes 3 minutes per inference. Fixing AllocToInPlace (#2975) will resolve the remaining latency issues