perf(gpu): more efficient column evaluation claims at the end of GKR - #2754
Merged
Conversation
Remove unused imports, debug println!s, and the commented-out fallback verification in evaluate_traces. Drop the hard-coded one-column zero padding between preprocessed and main and the matching preprocessed_cols skip in round_batch_evaluations: evaluate_traces now returns a contiguous [non-zero preprocessed][non-zero main] layout, and the caller's evals_so_far is already at the boundary after the preprocessed loop (it advances by num_polys only for non-zero entries, mirroring what evaluate_traces writes). This works generally — the previous layout only happened to align when there was exactly one added padding column and no poly_size==0 preprocessed entries. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
|
tamirhemo
approved these changes
May 13, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Before, the CUDA prover used repeated calls to the "jagged fix last variable" function/kernel to produce its evaluation claims on the columns needed at the end of GKR. This would take 15-17ms per shard.
In this PR, we use pointer arithmetic and the tensor dot kernels to do the same work. This takes around 3ms per shard now.