Skip to content

Optimize 1-bit estimator tail path#49

Open
Flitieter wants to merge 1 commit intoVectorDB-NTU:mainfrom
Flitieter:pr/include-optimized-estimator-cleanup
Open

Optimize 1-bit estimator tail path#49
Flitieter wants to merge 1 commit intoVectorDB-NTU:mainfrom
Flitieter:pr/include-optimized-estimator-cleanup

Conversation

@Flitieter
Copy link
Copy Markdown

This PR introduces a tail-padded AVX-512 path:
Full 512-dimensional blocks are processed with 512-bit SIMD loads and popcount, while the final partial block is stored compactly and handled with masked AVX-512 loads. This avoids falling back to scalar-style tail processing and keeps the tail path vectorized.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant