jerry-024 opened a new pull request, #76: URL: https://github.com/apache/paimon-vector-index/pull/76
## Summary Use the sparse transposed 8-bit IVF-PQ scan path for filter densities up to 25% instead of 12.5%. Local x86 benchmarks show this avoids unnecessary dense scans at the 18.75% and 25% crossover points. Depends on #75. Review the final optimization diff from commit `1357928` to `4eeea12`; once #75 merges, this PR diff will contain only that change. ## Changes - Change the transposed sparse-scan divisor from 8 to 4. - Cover the new 25% boundary and adjacent filtered-scan cases. - Add a deterministic 8-bit IVF-PQ filter-density benchmark. ## Testing - [x] `cargo fmt --all -- --check` - [x] `cargo test -p paimon-vindex-core` on x86_64: 449 passed, 0 failed, 1 ignored; integration suites passed - [x] Linux x86_64 JNI release build with glibc 2.17 target - [x] Interleaved benchmark medians: 18.75% 20.110 -> 11.213 ms (-44.2%); 25% 19.594 -> 12.733 ms (-35.0%); unchanged-density regressions stayed below 3% ## Notes The benchmark validates the generic x86 crossover decision. AMD Genoa-specific SIMD work is intentionally excluded and requires measurement on the production CPU family. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
