Repository navigation
Conversation
The SVE2 histogram accumulates counts in U16 lanes, so a symbol that appears 65,536 or more times in one call wraps, possibly to zero. That yields wrong probabilities, missed RLE detection, and corrupt Huffman output. Split large inputs into chunks of at most 65,520 bytes, a multiple of 16 that also leaves room for the zero padding of the final vector load. Merge each chunk's counts into U32 totals, tracking the largest count and max symbol across the whole input. Add a fuzzer test that checks exact counts, largest count and max symbol for every byte value and for sizes around the chunk and 16-bit boundaries, through both HIST_count_wksp and HIST_countFast_wksp. It also checks that entries past maxSymbolValue are not written and that an out-of-range symbol in a later chunk returns an error.
This branch has not been deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
#4711 flags an issue in SVE2 histogramming that this PR fixes.
The SVE2 histogram accumulates counts in U16 lanes, so a symbol that appears 65,536 or more times in one call wraps, possibly to zero. That yields wrong probabilities, missed RLE detection, and corrupt Huffman output.
Split large inputs into chunks of at most 65,520 bytes, a multiple of 16 that also leaves room for the zero padding of the final vector load. Merge each chunk's counts into U32 totals, tracking the largest count and max symbol across the whole input.
Add a fuzzer test that checks exact counts, largest count and max symbol for every byte value and for sizes around the chunk and 16-bit boundaries, through both HIST_count_wksp and HIST_countFast_wksp. It also checks that entries past maxSymbolValue are not written and that an out-of-range symbol in a later chunk returns an error.