original-brownbear opened a new pull request, #13932:
URL: https://github.com/apache/lucene/pull/13932
Generate cleaner code for PForUtil that has no dead parameters.
Also:
PForUtil instances always create their own `ForUtil`, so we can inline that
into the field declaration. Also, we can save cycles for accessing the input on
PostingsDecodingUtil.
Surprisingly, the combination of these cleanups yields a small but
statistically fully visible speedup that the compiler isn't able to get to on
its own it seems. For wikimedium with 40 task repeats and 40 JVMs I get:
```
TaskQPS baseline StdDevQPS
my_modified_version StdDev Pct diff p-value
OrHighNotLow 1031.61 (5.0%) 1039.33
(4.8%) 0.7% ( -8% - 11%) 0.494
TermDTSort 182.64 (6.8%) 184.65
(5.8%) 1.1% ( -10% - 14%) 0.437
HighTermDayOfYearSort 357.56 (6.4%) 361.83
(7.4%) 1.2% ( -11% - 16%) 0.440
MedTermDayTaxoFacets 33.24 (2.7%) 33.65
(2.7%) 1.2% ( -4% - 6%) 0.042
HighTermTitleSort 62.22 (2.9%) 63.03
(3.0%) 1.3% ( -4% - 7%) 0.048
OrHighMed 343.75 (5.2%) 348.33
(4.3%) 1.3% ( -7% - 11%) 0.213
OrNotHighLow 1597.43 (5.1%) 1620.18
(5.2%) 1.4% ( -8% - 12%) 0.219
BrowseRandomLabelSSDVFacets 3.34 (5.1%) 3.39
(4.0%) 1.5% ( -7% - 11%) 0.151
LowTerm 653.68 (5.1%) 663.51
(5.0%) 1.5% ( -8% - 12%) 0.186
PKLookup 241.00 (1.8%) 244.97
(2.1%) 1.6% ( -2% - 5%) 0.000
AndHighMedDayTaxoFacets 46.25 (2.1%) 47.10
(2.4%) 1.8% ( -2% - 6%) 0.000
OrHighNotMed 610.05 (5.7%) 621.53
(5.6%) 1.9% ( -8% - 14%) 0.138
HighTermMonthSort 1488.88 (4.7%) 1517.64
(5.3%) 1.9% ( -7% - 12%) 0.085
AndHighHighDayTaxoFacets 25.76 (3.5%) 26.26
(3.7%) 2.0% ( -5% - 9%) 0.015
Prefix3 507.94 (5.0%) 517.90
(4.9%) 2.0% ( -7% - 12%) 0.075
BrowseDateTaxoFacets 5.16 (9.7%) 5.26
(9.1%) 2.1% ( -15% - 23%) 0.326
BrowseMonthSSDVFacets 4.45 (8.4%) 4.54
(4.6%) 2.1% ( -10% - 16%) 0.167
AndHighHigh 112.67 (4.3%) 115.05
(3.8%) 2.1% ( -5% - 10%) 0.019
Wildcard 400.35 (4.0%) 408.87
(3.9%) 2.1% ( -5% - 10%) 0.017
OrHighLow 591.02 (2.9%) 603.62
(3.3%) 2.1% ( -3% - 8%) 0.002
BrowseDateSSDVFacets 1.26 (4.8%) 1.29
(6.7%) 2.2% ( -8% - 14%) 0.098
AndHighMed 271.86 (3.8%) 277.78
(3.2%) 2.2% ( -4% - 9%) 0.006
LowIntervalsOrdered 170.07 (5.5%) 173.79
(4.5%) 2.2% ( -7% - 12%) 0.052
AndHighLow 1876.44 (4.3%) 1918.82
(4.7%) 2.3% ( -6% - 11%) 0.026
MedTerm 640.37 (7.2%) 655.21
(6.3%) 2.3% ( -10% - 17%) 0.125
BrowseDayOfYearTaxoFacets 5.23 (9.8%) 5.36
(8.5%) 2.4% ( -14% - 22%) 0.247
BrowseRandomLabelTaxoFacets 4.39 (4.1%) 4.50
(4.4%) 2.4% ( -5% - 11%) 0.011
OrHighNotHigh 340.64 (5.8%) 349.08
(5.0%) 2.5% ( -7% - 14%) 0.041
MedPhrase 828.87 (3.5%) 849.87
(4.8%) 2.5% ( -5% - 11%) 0.007
HighIntervalsOrdered 38.15 (5.4%) 39.14
(4.6%) 2.6% ( -7% - 13%) 0.022
HighPhrase 150.20 (3.3%) 154.11
(4.5%) 2.6% ( -4% - 10%) 0.003
MedSpanNear 106.92 (5.3%) 109.71
(3.7%) 2.6% ( -6% - 12%) 0.011
OrHighMedDayTaxoFacets 1.94 (5.9%) 2.00
(5.0%) 2.6% ( -7% - 14%) 0.031
OrNotHighMed 393.61 (4.4%) 404.39
(4.0%) 2.7% ( -5% - 11%) 0.004
Fuzzy2 66.62 (3.5%) 68.49
(2.5%) 2.8% ( -3% - 9%) 0.000
LowPhrase 27.05 (5.4%) 27.82
(4.8%) 2.9% ( -6% - 13%) 0.012
HighSloppyPhrase 29.60 (6.8%) 30.45
(3.8%) 2.9% ( -7% - 14%) 0.018
MedIntervalsOrdered 35.98 (3.7%) 37.02
(4.0%) 2.9% ( -4% - 11%) 0.001
Respell 44.27 (2.0%) 45.56
(1.9%) 2.9% ( 0% - 6%) 0.000
OrHighHigh 100.77 (5.2%) 103.85
(6.2%) 3.1% ( -7% - 15%) 0.017
OrNotHighHigh 530.63 (4.1%) 547.05
(3.7%) 3.1% ( -4% - 11%) 0.000
Fuzzy1 114.15 (3.7%) 118.04
(2.4%) 3.4% ( -2% - 9%) 0.000
LowSpanNear 43.15 (5.5%) 44.63
(4.7%) 3.4% ( -6% - 14%) 0.003
HighTermTitleBDVSort 16.19 (3.8%) 16.75
(3.8%) 3.4% ( -3% - 11%) 0.000
IntNRQ 73.96 (5.4%) 76.51
(8.6%) 3.4% ( -9% - 18%) 0.032
HighSpanNear 20.01 (5.4%) 20.74
(5.1%) 3.6% ( -6% - 14%) 0.002
MedSloppyPhrase 19.62 (4.5%) 20.33
(3.9%) 3.6% ( -4% - 12%) 0.000
BrowseDayOfYearSSDVFacets 4.56 (9.3%) 4.73
(10.5%) 3.7% ( -14% - 26%) 0.092
HighTerm 596.70 (7.0%) 619.46
(7.8%) 3.8% ( -10% - 19%) 0.021
LowSloppyPhrase 24.74 (4.6%) 25.72
(5.8%) 4.0% ( -6% - 15%) 0.001
BrowseMonthTaxoFacets 11.38 (37.0%) 12.99
(32.5%) 14.1% ( -40% - 132%) 0.070
```
--
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]
