jpountz commented on PR #14080:
URL: https://github.com/apache/lucene/pull/14080#issuecomment-2551572454
On my Apple M3:
```
TaskQPS baseline StdDevQPS
my_modified_version StdDev Pct diff p-value
CountOrHighMed 211.78 (1.9%) 174.58
(0.9%) -17.6% ( -19% - -15%) 0.000
IntNRQ 201.91 (9.6%) 195.67
(2.9%) -3.1% ( -14% - 10%) 0.218
OrHighRare 519.00 (11.4%) 503.13
(11.4%) -3.1% ( -23% - 22%) 0.450
TermDTSort 362.51 (4.8%) 357.37
(5.4%) -1.4% ( -11% - 9%) 0.430
FilteredTerm 256.14 (3.1%) 252.99
(2.5%) -1.2% ( -6% - 4%) 0.215
Or3Terms 243.43 (5.0%) 240.50
(3.8%) -1.2% ( -9% - 8%) 0.447
Fuzzy2 158.21 (2.2%) 156.61
(2.2%) -1.0% ( -5% - 3%) 0.195
Fuzzy1 198.82 (2.8%) 196.82
(2.7%) -1.0% ( -6% - 4%) 0.305
CountTerm 21457.63 (4.3%) 21241.75
(5.2%) -1.0% ( -10% - 8%) 0.550
FilteredOrStopWords 67.81 (5.7%) 67.14
(5.0%) -1.0% ( -11% - 10%) 0.602
OrMany 28.30 (6.0%) 28.02
(5.9%) -1.0% ( -12% - 11%) 0.640
CountPhrase 7.04 (6.1%) 6.97
(5.7%) -1.0% ( -12% - 11%) 0.641
OrStopWords 49.35 (4.2%) 48.89
(4.5%) -0.9% ( -9% - 8%) 0.548
DismaxOrHighHigh 186.55 (5.0%) 185.05
(4.4%) -0.8% ( -9% - 9%) 0.629
CountFilteredPhrase 46.58 (3.1%) 46.21
(3.1%) -0.8% ( -6% - 5%) 0.470
AndMedOrHighHigh 80.49 (6.2%) 79.90
(7.1%) -0.7% ( -13% - 13%) 0.756
Or2Terms2StopWords 274.89 (3.8%) 272.96
(3.7%) -0.7% ( -7% - 7%) 0.597
FilteredOrHighHigh 93.49 (5.8%) 92.88
(5.1%) -0.7% ( -10% - 10%) 0.733
FilteredOrMany 20.68 (6.5%) 20.54
(5.6%) -0.6% ( -12% - 12%) 0.763
PKLookup 384.44 (5.0%) 382.07
(4.5%) -0.6% ( -9% - 9%) 0.716
Term 860.96 (5.5%) 855.84
(4.9%) -0.6% ( -10% - 10%) 0.746
FilteredOr2Terms2StopWords 252.12 (4.6%) 250.66
(4.1%) -0.6% ( -8% - 8%) 0.709
FilteredOrHighMed 247.09 (5.1%) 245.88
(4.3%) -0.5% ( -9% - 9%) 0.768
Wildcard 189.41 (4.1%) 188.52
(3.9%) -0.5% ( -8% - 7%) 0.742
FilteredOr3Terms 254.42 (5.6%) 253.28
(4.8%) -0.4% ( -10% - 10%) 0.807
FilteredPhrase 47.58 (1.9%) 47.37
(2.2%) -0.4% ( -4% - 3%) 0.548
DismaxOrHighMed 277.65 (3.7%) 276.50
(3.0%) -0.4% ( -6% - 6%) 0.729
FilteredAnd2Terms2StopWords 274.15 (4.8%) 273.02
(5.5%) -0.4% ( -10% - 10%) 0.821
And2Terms2StopWords 279.32 (2.6%) 278.35
(4.1%) -0.3% ( -6% - 6%) 0.777
CountOrHighHigh 113.83 (1.3%) 113.46
(1.2%) -0.3% ( -2% - 2%) 0.445
Respell 145.07 (2.6%) 144.63
(2.1%) -0.3% ( -4% - 4%) 0.719
OrHighMed 308.19 (4.1%) 307.31
(3.9%) -0.3% ( -8% - 8%) 0.840
Prefix3 342.80 (2.8%) 342.03
(2.1%) -0.2% ( -4% - 4%) 0.797
FilteredAnd3Terms 294.41 (4.1%) 294.08
(4.6%) -0.1% ( -8% - 8%) 0.942
AndHighOrMedMed 70.53 (6.0%) 70.47
(6.0%) -0.1% ( -11% - 12%) 0.964
CombinedOrHighHigh 27.17 (4.8%) 27.16
(5.4%) -0.1% ( -9% - 10%) 0.977
OrHighHigh 75.74 (4.1%) 75.75
(3.9%) 0.0% ( -7% - 8%) 0.996
DismaxTerm 867.09 (7.5%) 867.55
(6.7%) 0.1% ( -13% - 15%) 0.983
TermDayOfYearSort 2896.80 (2.4%) 2898.74
(1.7%) 0.1% ( -3% - 4%) 0.928
FilteredAndHighHigh 78.46 (5.9%) 78.58
(6.2%) 0.1% ( -11% - 12%) 0.947
FilteredAndStopWords 52.07 (7.7%) 52.15
(8.2%) 0.1% ( -14% - 17%) 0.958
And3Terms 253.63 (2.7%) 254.13
(3.4%) 0.2% ( -5% - 6%) 0.854
AndStopWords 39.57 (3.5%) 39.66
(4.2%) 0.2% ( -7% - 8%) 0.872
FilteredAndHighMed 179.72 (3.9%) 180.22
(4.3%) 0.3% ( -7% - 8%) 0.845
CombinedOrHighMed 109.83 (4.7%) 110.18
(4.1%) 0.3% ( -8% - 9%) 0.837
TermMonthSort 450.03 (5.5%) 452.74
(5.5%) 0.6% ( -9% - 12%) 0.756
CombinedAndHighMed 49.10 (5.3%) 49.57
(5.3%) 1.0% ( -9% - 12%) 0.607
AndHighMed 193.76 (3.6%) 195.66
(2.6%) 1.0% ( -5% - 7%) 0.377
AndHighHigh 75.00 (4.0%) 75.91
(2.4%) 1.2% ( -4% - 7%) 0.299
Phrase 23.13 (4.0%) 23.41
(4.0%) 1.2% ( -6% - 9%) 0.389
TermTitleSort 148.32 (3.1%) 150.21
(4.8%) 1.3% ( -6% - 9%) 0.375
CombinedAndHighHigh 14.74 (5.5%) 14.94
(5.1%) 1.4% ( -8% - 12%) 0.469
TermGroup100 38.08 (3.8%) 38.78
(3.2%) 1.8% ( -5% - 9%) 0.146
TermGroup10K 35.30 (4.4%) 36.04
(4.1%) 2.1% ( -6% - 11%) 0.155
TermGroup1M 34.63 (3.8%) 35.50
(3.2%) 2.5% ( -4% - 9%) 0.043
TermBGroup1M1P 58.89 (4.4%) 60.45
(3.3%) 2.6% ( -4% - 10%) 0.053
TermBGroup1M 42.22 (4.2%) 43.41
(3.8%) 2.8% ( -4% - 11%) 0.046
CombinedTerm 51.19 (5.6%) 53.04
(2.0%) 3.6% ( -3% - 11%) 0.016
CountFilteredOrHighHigh 81.73 (1.7%) 89.18
(1.0%) 9.1% ( 6% - 12%) 0.000
CountFilteredOrMany 9.53 (1.7%) 11.24
(1.0%) 17.9% ( 14% - 21%) 0.000
CountFilteredOrHighMed 99.37 (1.3%) 123.02
(0.9%) 23.8% ( 21% - 26%) 0.000
CountAndHighMed 206.17 (1.5%) 263.24
(2.5%) 27.7% ( 23% - 32%) 0.000
CountAndHighHigh 67.21 (1.2%) 118.73
(2.1%) 76.7% ( 72% - 80%) 0.000
```
`CountOrHighMed` has a slowdown because we evaluate counts on disjunctions
via their intersection counts sometimes. I'll see if I can tune the heuristics
to make it faster.
I also ran all queries from https://tantivy-search.github.io/bench/ and this
change was often a big speedup (up to multiple times) and sometimes a small
slowdown (< 10%).
--
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
