perf: revert tile-count pass to single-splat-per-thread#8603
Merged
mvaligursky merged 1 commit intomainfrom Apr 15, 2026
Merged
perf: revert tile-count pass to single-splat-per-thread#8603mvaligursky merged 1 commit intomainfrom
mvaligursky merged 1 commit intomainfrom
Conversation
The multi-splat-per-thread (N=8) strategy from #8594 caused significant register pressure, degrading occupancy on both NVIDIA and Apple GPUs. Benchmarks at 17M splats show tile-count is 2.1x faster on RTX 2070 and 1.7x faster on M4 with N=1.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Reverts the multi-splat-per-thread (N=8) tile-count strategy introduced in #8594, which caused significant register pressure and performance regression on NVIDIA GPUs.
Changes:
SPLATS_PER_THREADconstant and all register arrays from the tile-count shaderPerformance:
Benchmarked at 17M splats:
Tile-count pass is 2.1x faster on NVIDIA and 1.7x faster on M4 with N=1. The register array approach increased register pressure which reduced occupancy on both architectures, with NVIDIA being more severely affected.