Choose fastest bit reversal method at runtime #60
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
…ecting the fastest one
…rsions, same as Dif one
…ents and include it in the planner
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #60 +/- ##
==========================================
- Coverage 99.82% 99.43% -0.40%
==========================================
Files 13 14 +1
Lines 2261 2289 +28
==========================================
+ Hits 2257 2276 +19
- Misses 4 13 +9 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Collaborator
Author
|
I've re-run the benchmarks against latest main on zen4 on the same versions of the compiler and all dependencies. The largest regression is 4% while many smaller sizes improve by a lot, some over 2x. I think this is good to go. benchmarks on desktop zen4 vs main on commit 0f47ea1There is more we can do here. I'd like to integrate the LUT variant of COBRA from #47. We could also add constraints on the applicable sizes for each transform so that we wouldn't need to bundle all unrolled variants into one function, not have to measure COBRA for smaller sizes, etc. |
Collaborator
Author
|
I've messed around with COBRA block sizes. On Zen4, switching block size from 128 to 64 significantly helps I'll add the LUT variant from #47 and add variants for different block sizes. |
Closed
Collaborator
Author
|
closing in favor of #62 - CO-BRAVO is fastest for all benchmarked sizes |
This was referenced Jan 24, 2026
Closed
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Outdated suggestions cannot be applied.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.
Helps x86 a lot by switching away from unrolled impls. Fixes #49
Also papers over whatever memory subsystem quirk we're hitting at f32/8388608 and f64/4194304, possibly something to do with cache associativity.
preliminary benchmarks from zen4
On Apple M4 this is big improvement for f32/32768 which I assume is also due to some implementations hitting cache associativity issues or a similar hardware quirk. No equivalent change for f64/16384, oddly enough. Also +7% gains for all f64 up to and including 1024, no change on larger sizes.
TODO:
cobra_apply#47