Repository navigation
Fuse the three jacobian dot products into one horizontal reduction. - #2151
Merged
Merged
Conversation
SolveVelocityConstraintGetTotalLambda computed three independent Vec3::Dot calls and summed the scalar results. Since Dot(a,b) = sum_xyz(a*b), the sum of the dots equals a single reduction of the component-wise sum of products, so the two redundant horizontal reductions can be removed from the latency-bound critical path. Accumulate component-wise into a Vec3 and call ReduceSum() once. Performance uplift about 3% for Pyramid test with 300 iterations. Cross-platform determinism is preserved but results changed (fewer roundings), so the determinism check hashes have been regenerated.
Owner
|
Clever, thanks! |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
SolveVelocityConstraintGetTotalLambda computed three independent Vec3::Dot calls and summed the scalar results. Since Dot(a,b) = sum_xyz(a*b), the sum of the dots equals a single reduction of the component-wise sum of products, so the two redundant horizontal reductions can be removed from the latency-bound critical path. Accumulate component-wise into a Vec3 and call ReduceSum() once.
Performance uplift about 3% for Pyramid test with 300 iterations.
Cross-platform determinism is preserved but results changed (fewer roundings), so the determinism check hashes have been regenerated.