gf2121 opened a new pull request, #14751: URL: https://github.com/apache/lucene/pull/14751
This tries to speed up `BlockMaxConjunctionBulkScorer` by: * Score with a default window when lead clause does not have specialized `shallowAdvance`. * Stop doc first once dynamic pruning kicks in. * Increase batch size of `Scorer#nextDocsAndScores` to some reduce per-batch overhead. ``` TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff p-value FilteredAnd2Terms2StopWords 135.81 (4.4%) 131.67 (4.0%) -3.0% ( -11% - 5%) 0.042 AndHighOrMedMed 33.03 (3.0%) 32.22 (5.7%) -2.5% ( -10% - 6%) 0.128 Term 1191.15 (6.7%) 1170.11 (6.6%) -1.8% ( -14% - 12%) 0.453 TermMonthSort 4725.92 (3.9%) 4650.17 (4.9%) -1.6% ( -9% - 7%) 0.303 FilteredOrHighMed 93.59 (4.2%) 92.22 (4.5%) -1.5% ( -9% - 7%) 0.344 FilteredOrMany 7.91 (2.1%) 7.80 (3.7%) -1.4% ( -7% - 4%) 0.187 AndMedOrHighHigh 38.74 (1.8%) 38.21 (4.7%) -1.4% ( -7% - 5%) 0.275 FilteredOr2Terms2StopWords 110.68 (3.9%) 109.31 (4.0%) -1.2% ( -8% - 6%) 0.377 And2Terms2StopWords 159.62 (6.2%) 157.69 (5.4%) -1.2% ( -12% - 11%) 0.554 CountOrHighMed 155.40 (2.7%) 153.56 (3.5%) -1.2% ( -7% - 5%) 0.282 DismaxTerm 1116.22 (4.1%) 1103.38 (4.7%) -1.2% ( -9% - 7%) 0.458 Prefix3 184.48 (1.7%) 182.37 (4.1%) -1.1% ( -6% - 4%) 0.304 FilteredTerm 164.72 (3.7%) 162.90 (4.7%) -1.1% ( -9% - 7%) 0.453 FilteredOr3Terms 95.66 (4.6%) 94.61 (4.2%) -1.1% ( -9% - 8%) 0.481 FilteredOrHighHigh 29.19 (3.6%) 28.88 (4.7%) -1.1% ( -9% - 7%) 0.474 TermDTSort 258.92 (1.5%) 256.39 (3.4%) -1.0% ( -5% - 3%) 0.290 OrHighMed 193.96 (2.3%) 192.10 (4.9%) -1.0% ( -8% - 6%) 0.481 CountOrMany 11.92 (3.7%) 11.81 (4.5%) -1.0% ( -8% - 7%) 0.508 FilteredPhrase 23.25 (2.9%) 23.03 (3.7%) -0.9% ( -7% - 5%) 0.430 FilteredPrefix3 170.12 (1.4%) 168.57 (4.3%) -0.9% ( -6% - 4%) 0.418 CountOrHighHigh 88.41 (2.1%) 87.61 (4.0%) -0.9% ( -6% - 5%) 0.421 Respell 79.83 (3.0%) 79.13 (3.8%) -0.9% ( -7% - 6%) 0.466 CombinedTerm 21.16 (1.4%) 20.98 (2.6%) -0.8% ( -4% - 3%) 0.271 FilteredOrStopWords 18.93 (3.9%) 18.79 (4.8%) -0.8% ( -9% - 8%) 0.622 TermDayOfYearSort 332.18 (1.0%) 329.75 (2.7%) -0.7% ( -4% - 3%) 0.315 DismaxOrHighMed 121.49 (2.9%) 120.62 (4.3%) -0.7% ( -7% - 6%) 0.579 AndHighHigh 50.99 (4.3%) 50.63 (6.0%) -0.7% ( -10% - 10%) 0.699 CountFilteredOrHighMed 47.32 (6.0%) 47.00 (6.7%) -0.7% ( -12% - 12%) 0.762 Fuzzy1 99.32 (3.1%) 98.66 (4.5%) -0.7% ( -8% - 7%) 0.628 CountAndHighHigh 83.97 (1.8%) 83.44 (3.5%) -0.6% ( -5% - 4%) 0.512 TermTitleSort 174.98 (2.5%) 173.98 (3.8%) -0.6% ( -6% - 5%) 0.613 CountFilteredOrHighHigh 38.61 (4.9%) 38.39 (5.8%) -0.6% ( -10% - 10%) 0.767 Fuzzy2 91.66 (2.8%) 91.18 (4.0%) -0.5% ( -7% - 6%) 0.665 Wildcard 119.03 (2.3%) 118.41 (4.4%) -0.5% ( -7% - 6%) 0.676 CountFilteredOrMany 10.51 (3.4%) 10.46 (4.6%) -0.5% ( -8% - 7%) 0.718 Phrase 16.45 (3.7%) 16.37 (4.8%) -0.5% ( -8% - 8%) 0.738 AndHighMed 130.52 (2.5%) 129.91 (4.1%) -0.5% ( -6% - 6%) 0.693 CountAndHighMed 121.46 (3.2%) 120.96 (3.5%) -0.4% ( -6% - 6%) 0.728 CountFilteredPhrase 21.74 (3.0%) 21.65 (3.5%) -0.4% ( -6% - 6%) 0.721 DismaxOrHighHigh 76.77 (4.4%) 76.75 (5.6%) -0.0% ( -9% - 10%) 0.984 Or2Terms2StopWords 155.03 (6.1%) 155.30 (4.9%) 0.2% ( -10% - 11%) 0.931 OrHighHigh 49.00 (3.6%) 49.10 (5.5%) 0.2% ( -8% - 9%) 0.906 CountTerm 14454.86 (4.9%) 14487.18 (7.1%) 0.2% ( -11% - 12%) 0.917 OrMany 9.80 (6.5%) 9.83 (5.5%) 0.2% ( -11% - 13%) 0.912 And3Terms 158.70 (6.4%) 159.42 (4.8%) 0.5% ( -10% - 12%) 0.819 AndStopWords 15.66 (8.8%) 15.81 (6.7%) 1.0% ( -13% - 18%) 0.719 FilteredAnd3Terms 185.25 (2.4%) 187.10 (2.9%) 1.0% ( -4% - 6%) 0.284 Or3Terms 147.30 (6.7%) 149.09 (5.6%) 1.2% ( -10% - 14%) 0.579 OrStopWords 17.85 (8.7%) 18.08 (7.8%) 1.3% ( -13% - 19%) 0.657 OrHighRare 224.34 (10.6%) 228.20 (13.4%) 1.7% ( -20% - 28%) 0.688 CombinedOrHighMed 51.80 (2.7%) 53.00 (4.2%) 2.3% ( -4% - 9%) 0.066 CombinedOrHighHigh 10.77 (3.5%) 11.16 (4.2%) 3.6% ( -3% - 11%) 0.008 FilteredAndHighMed 71.19 (5.9%) 74.13 (4.2%) 4.1% ( -5% - 15%) 0.023 FilteredAndHighHigh 22.12 (6.3%) 23.08 (4.9%) 4.3% ( -6% - 16%) 0.030 FilteredAndStopWords 18.27 (5.7%) 19.68 (4.9%) 7.7% ( -2% - 19%) 0.000 CombinedAndHighHigh 10.23 (1.9%) 11.26 (4.1%) 10.1% ( 3% - 16%) 0.000 CombinedAndHighMed 47.30 (1.7%) 53.66 (4.2%) 13.5% ( 7% - 19%) 0.000 ``` -- 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