iprithv opened a new pull request, #16002: URL: https://github.com/apache/lucene/pull/16002
## Description This changes `MaxScoreBulkScorer` to sometimes evaluate essential clauses before applying a filter. When the filter cost is substantially higher than the sum of essential-clause costs, the scorer now collects essential hits into a buffer, removes hits that do not match the filter, and then scores non-essential clauses. The heuristic is intentionally conservative: post-filtering is only used when `filter.cost > 16 * sum(essential costs)`. This keeps the existing leap-frog path for marginal cases where extra buffering can offset the benefit. ## Benchmark | essentialInterval | filterInterval | baseline ops/s | current ops/s | delta | |---:|---:|---:|---:|---:| | 100 | 1 | 107.60 | 108.08 | +0.4% | | 100 | 10 | 514.12 | 584.55 | +13.7% | | 100 | 1000 | 19321.47 | 19664.42 | +1.8% | | 1000 | 1 | 102.01 | 111.26 | +9.1% | | 1000 | 10 | 586.99 | 593.03 | +1.0% | | 1000 | 1000 | 26120.80 | 27941.56 | +7.0% | -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
