Improve the performance of computing the destination for each shred #1131
Conversation
This file contains 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
ptaffet-jump
force-pushed
the
ptaffet/shred-perf-improvements
branch
2 times, most recently
from
a69f46e to
c377173
Compare
Switch from using treap to using a light-weight radix-9 tree inspired by a B-tree. Almost entirely branch-free because branch mispredictions were killer.
ptaffet-jump
force-pushed
the
ptaffet/shred-perf-improvements
branch
from
c377173 to
357b4d7
Compare
mmcgee-jump
approved these changes
Jan 31, 2024
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Outdated suggestions cannot be applied.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.
Performance investigations showed that computing the destination for each shred was extremely expensive compared to computing the shred data itself. With a lot of effort (and including the recent Chacha20 improvements), performance increased from 28.3 us/shred to 8.3 us/shred on my workstation.
Shred footprint changed, and that is causing some tests to fail. I should also rewrite the explanation to update it to radix 9, and switch the right_sums to left_sums which are much more intuitive, since the performance win from right sums didn't materialize.