jpountz commented on PR #14896: URL: https://github.com/apache/lucene/pull/14896#issuecomment-3090202052
Nightly benchmarks caught up with the change: https://benchmarks.mikemccandless.com/2025.07.17.18.04.07.html. There seems to be a few small speedups (AndHighHigh, OrHighHigh) and a few small slowdowns (FilteredOrStopWords, FilteredOr2Terms2StopWords, FilteredOrHighMed). In general, the slowdowns are tiny, so I don't think we need to revert, this would likely still help when AVX512 is available? But the fact that filtered queries are impacted suggests that the vectorized impl performs slower when all scores are greater than or equal to the threshold (this is because the filter is applied as a leading ConstantScoreQuery with a score of 0 on these queries). Maybe we can do something about it? Either by detecting the case when the leading clause produces 0 scores since there's nothing to filter, or/and by looking optimizing vectorized filtering in this case? -- 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: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
