Optimizes weights kernel #30
Merged
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
Updates fork
Merge master into fork
Update weights branch with latest version of code See merge request daclark/gauxc-mirror!3
|
I updated the PR with some clean up changes and minor optimizations. Right now it is failing 371 of the assertions in the weight unit test, however the differences are all relatively small. Removing the reciprocal optimization seems to resolve the issue. |
|
@dmclark17 How small is small? I'm more than happy to update the UT checks if its a worth while optimization |
|
The maximum absolute difference 1.4e-5 and the average absolute difference being 4.8e-8. The maximum percent difference is 0.4% and the average percent difference is 0.03%. The reciprocal optimization brings the total runtime of the weights kernel from 18.9s to 14.2s for a ubiquitin simulation on a V100. |
wavefunction91
approved these changes
Dec 18, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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Outdated suggestions cannot be applied.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.
This changes the level of parallelism to assign a warp to each point instead of a thread.
For the iParent portion, a warp computes the contribution from 32 jCenters concurrently. They then do a warp reduction to compute the final contribution. This changes the early exit scheme since the check effectively only happens every 32 jCenters.
For the pairs portion, each thread in the warp computes the contribution from 1 iCenter at a time as the warp synchronously iterates over jCenters. The threads are able to independently exit early and start on a new iCenter.