jpountz opened a new pull request, #12361:
URL: https://github.com/apache/lucene/pull/12361

   We currently use block-max maxscore for top-level disjunctions, implemented 
as 
   a scorer. Since we only use it for top-level disjunctions, we could actually
   implement it as a bulk scorer, which helps save some overhead. luceneutil
   reports the following numbers on wikimedium10m:
   
   ```
                               TaskQPS baseline      StdDevQPS 
my_modified_version      StdDev                Pct diff p-value
                       HighSpanNear        9.15      (3.7%)        9.03      
(3.5%)   -1.4% (  -8% -    6%) 0.224
                            Prefix3      434.41      (2.2%)      429.16      
(2.1%)   -1.2% (  -5% -    3%) 0.080
               MedTermDayTaxoFacets       37.33      (6.3%)       36.88      
(6.6%)   -1.2% ( -13% -   12%) 0.558
                         AndHighLow     1315.31      (3.1%)     1299.90      
(3.9%)   -1.2% (  -7% -    6%) 0.294
                        MedSpanNear       42.42      (2.5%)       41.96      
(2.3%)   -1.1% (  -5% -    3%) 0.153
             OrHighMedDayTaxoFacets        5.66      (4.9%)        5.60      
(4.9%)   -1.1% ( -10% -    9%) 0.488
                   HighSloppyPhrase       16.72      (3.5%)       16.57      
(5.4%)   -0.9% (  -9% -    8%) 0.539
                           Wildcard      129.90      (3.9%)      128.92      
(3.8%)   -0.8% (  -8% -    7%) 0.537
                         HighPhrase       68.61      (5.5%)       68.10      
(4.4%)   -0.7% ( -10% -    9%) 0.637
                          MedPhrase       27.46      (3.9%)       27.26      
(3.5%)   -0.7% (  -7% -    6%) 0.538
        BrowseRandomLabelSSDVFacets       14.70      (7.4%)       14.61      
(7.7%)   -0.7% ( -14% -   15%) 0.779
                            LowTerm      816.63      (5.6%)      811.34      
(5.0%)   -0.6% ( -10% -   10%) 0.699
              BrowseMonthSSDVFacets       20.41      (1.1%)       20.28      
(2.0%)   -0.6% (  -3% -    2%) 0.207
                          LowPhrase       43.61      (3.4%)       43.35      
(3.0%)   -0.6% (  -6% -    6%) 0.561
                             Fuzzy1      135.81      (1.2%)      135.42      
(1.5%)   -0.3% (  -3% -    2%) 0.504
                        LowSpanNear      114.09      (1.7%)      113.78      
(1.8%)   -0.3% (  -3% -    3%) 0.626
                             Fuzzy2       71.78      (1.1%)       71.60      
(1.0%)   -0.2% (  -2% -    1%) 0.454
           AndHighHighDayTaxoFacets       31.06      (2.2%)       30.98      
(2.3%)   -0.2% (  -4% -    4%) 0.730
                             IntNRQ       88.61      (5.8%)       88.46      
(5.7%)   -0.2% ( -11% -   12%) 0.926
                  HighTermMonthSort     3779.27      (3.8%)     3775.75      
(3.3%)   -0.1% (  -6% -    7%) 0.934
            AndHighMedDayTaxoFacets       58.44      (1.8%)       58.42      
(1.9%)   -0.0% (  -3% -    3%) 0.948
                            Respell       80.73      (1.6%)       80.82      
(1.4%)    0.1% (  -2% -    3%) 0.815
                            MedTerm      731.12      (6.8%)      732.37      
(7.1%)    0.2% ( -12% -   15%) 0.938
                           PKLookup      236.79      (4.4%)      237.48      
(4.6%)    0.3% (  -8% -    9%) 0.838
                         TermDTSort      181.53      (2.7%)      182.14      
(2.1%)    0.3% (  -4% -    5%) 0.661
              HighTermDayOfYearSort      422.38      (3.2%)      423.81      
(3.6%)    0.3% (  -6% -    7%) 0.752
                    LowSloppyPhrase       46.81      (2.7%)       46.98      
(3.0%)    0.3% (  -5% -    6%) 0.696
                         AndHighMed      342.09      (4.6%)      343.63      
(3.8%)    0.4% (  -7% -    9%) 0.737
                        AndHighHigh       46.06      (6.6%)       46.28      
(5.8%)    0.5% ( -11% -   13%) 0.809
               HighTermTitleBDVSort       23.23      (3.6%)       23.34      
(3.2%)    0.5% (  -6% -    7%) 0.650
                           HighTerm      685.44      (7.3%)      689.42      
(7.5%)    0.6% ( -13% -   16%) 0.804
                  HighTermTitleSort      156.76      (5.8%)      157.96      
(5.6%)    0.8% ( -10% -   12%) 0.671
               HighIntervalsOrdered       25.11      (5.2%)       25.32      
(5.2%)    0.8% (  -9% -   11%) 0.607
                       OrNotHighLow     1803.79      (3.7%)     1819.26      
(3.5%)    0.9% (  -6% -    8%) 0.452
                LowIntervalsOrdered       62.41      (3.9%)       63.01      
(3.7%)    1.0% (  -6% -    8%) 0.423
                       OrNotHighMed      456.34      (3.5%)      460.92      
(4.3%)    1.0% (  -6% -    9%) 0.419
                       OrHighNotLow      365.78      (8.4%)      369.70      
(9.0%)    1.1% ( -15% -   20%) 0.698
                      OrNotHighHigh      272.99      (6.9%)      276.13      
(7.7%)    1.2% ( -12% -   16%) 0.618
                      OrHighNotHigh      438.11      (6.5%)      443.93      
(7.3%)    1.3% ( -11% -   16%) 0.543
                       OrHighNotMed      371.40      (7.4%)      376.34      
(8.4%)    1.3% ( -13% -   18%) 0.595
                    MedSloppyPhrase        6.47      (4.8%)        6.56      
(4.7%)    1.4% (  -7% -   11%) 0.357
               BrowseDateSSDVFacets        5.50      (9.0%)        5.61     
(10.0%)    2.0% ( -15% -   23%) 0.509
                MedIntervalsOrdered       29.98      (5.3%)       30.58      
(5.5%)    2.0% (  -8% -   13%) 0.242
          BrowseDayOfYearSSDVFacets       19.88      (8.8%)       20.62     
(14.4%)    3.7% ( -17% -   29%) 0.322
              BrowseMonthTaxoFacets       26.63     (20.6%)       27.75     
(15.2%)    4.2% ( -26% -   50%) 0.462
                          OrHighLow      555.47      (3.1%)      579.50      
(2.4%)    4.3% (  -1% -   10%) 0.000
        BrowseRandomLabelTaxoFacets       31.81     (23.1%)       33.32     
(19.2%)    4.8% ( -30% -   61%) 0.477
               BrowseDateTaxoFacets       38.87     (24.8%)       40.81     
(18.7%)    5.0% ( -30% -   64%) 0.471
          BrowseDayOfYearTaxoFacets       39.19     (24.8%)       41.52     
(18.8%)    5.9% ( -30% -   65%) 0.394
                          OrHighMed      138.70      (3.4%)      149.08      
(3.4%)    7.5% (   0% -   14%) 0.000
                         OrHighHigh       44.38      (3.3%)       51.20      
(3.8%)   15.4% (   8% -   23%) 0.000
   ```
   
   OrHighHigh, OrHighMed and orHighLow all get a speedup with this change.


-- 
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

Reply via email to