perf: localize SIMD to profitable scan paths - #235
Open
BridgeAR wants to merge 1 commit into
Open
Conversation
SIMD setup dominates short scans, and whole-module `-msimd128` also changes scalar hot paths. This keeps short and unrelated scans scalar, then uses SIMD for long line comments and templates. Non-SIMD builds remain byte-identical to v3. Long line-comment scans improve 9-65%, and long templates improve 66.8% against the SHA-identical scalar control on Node 18 arm64. The complete 3.06 MB sample corpus shows no aggregate regression across Node 18-26. Two isolated-process runs each produced a 0.94% equal-version geomean improvement. The SIMD build is 563 bytes smaller than #231.
BridgeAR
marked this pull request as ready for review
August 10, 2026 18:40
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
SIMD setup dominates short scans, and whole-module
-msimd128also changes scalar hot paths. This keeps short and unrelated scans scalar, then uses SIMD for long line comments and templates. Non-SIMD builds remain byte-identical to v3.Long line-comment scans improve 9-65%, and long templates improve 66.8% against the SHA-identical scalar control on Node 18 arm64. The complete 3.06 MB sample corpus shows no aggregate regression across Node 18-26. Two isolated-process runs each produced a 0.94% equal-version geomean improvement. The SIMD build is 563 bytes smaller than #231.