[jira] [Commented] (LUCENE-8180) Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit
[ https://issues.apache.org/jira/browse/LUCENE-8180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16442629#comment-16442629 ] Adrien Grand commented on LUCENE-8180: -- (Note: I had to patch luceneutil to use the JAR file of lucene-core rather than the classes folder https://github.com/mikemccand/luceneutil/pull/14) > Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit > > > Key: LUCENE-8180 > URL: https://issues.apache.org/jira/browse/LUCENE-8180 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Adrien Grand >Priority: Trivial > Labels: newdev > Attachments: LUCENE-8180.patch > > > Using Arrays.mismatch with a fixed-size array full of zeros might help find > the next long that is not 0 faster. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8180) Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit
[ https://issues.apache.org/jira/browse/LUCENE-8180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16442627#comment-16442627 ] Adrien Grand commented on LUCENE-8180: -- It doesn't seem to. I tried to test this change but there is a slight slowdown with wikibigall: {noformat} TaskQPS baseline StdDev QPS patch StdDev Pct diff Prefix3 107.21 (6.9%) 100.80 (5.7%) -6.0% ( -17% -7%) Wildcard 73.71 (6.9%) 70.43 (6.0%) -4.4% ( -16% -9%) IntNRQ 28.69 (10.1%) 27.82 (9.0%) -3.0% ( -20% - 17%) OrHighHigh 32.25 (3.2%) 32.12 (3.3%) -0.4% ( -6% -6%) OrHighMed 101.35 (2.5%) 101.03 (2.2%) -0.3% ( -4% -4%) HighSpanNear5.51 (5.5%)5.50 (5.4%) -0.2% ( -10% - 11%) MedPhrase 14.55 (1.7%) 14.52 (1.6%) -0.2% ( -3% -3%) AndHighMed 215.89 (3.0%) 215.59 (2.9%) -0.1% ( -5% -5%) MedSloppyPhrase 23.26 (3.3%) 23.23 (3.3%) -0.1% ( -6% -6%) MedSpanNear 79.38 (5.1%) 79.29 (5.1%) -0.1% ( -9% - 10%) LowPhrase 67.40 (1.4%) 67.33 (1.4%) -0.1% ( -2% -2%) AndCommon 154.32 (4.2%) 154.21 (4.2%) -0.1% ( -8% -8%) LowSpanNear 39.95 (3.1%) 39.92 (3.2%) -0.1% ( -6% -6%) HighPhrase 60.46 (1.7%) 60.42 (1.6%) -0.1% ( -3% -3%) LowSloppyPhrase 706.30 (2.0%) 705.95 (1.9%) -0.1% ( -3% -3%) AndHighHigh 37.60 (2.6%) 37.59 (2.8%) -0.0% ( -5% -5%) OrHighLow 1380.85 (4.3%) 1380.85 (3.6%) 0.0% ( -7% -8%) HighSloppyPhrase1.87 (7.9%)1.87 (7.7%) 0.1% ( -14% - 17%) OrCommon 217.99 (5.1%) 218.29 (5.1%) 0.1% ( -9% - 10%) HighTerm 433.84 (6.7%) 434.54 (6.5%) 0.2% ( -12% - 14%) LowTerm 2344.87 (4.6%) 2349.37 (4.7%) 0.2% ( -8% -9%) HighTermMonthSort 55.69 (11.1%) 55.83 (11.2%) 0.3% ( -19% - 25%) Respell 211.40 (2.6%) 211.98 (2.5%) 0.3% ( -4% -5%) MedTerm 1125.82 (5.6%) 1129.18 (5.4%) 0.3% ( -10% - 11%) HighTermDayOfYearSort 98.79 (9.2%) 99.16 (8.4%) 0.4% ( -15% - 19%) Fuzzy2 111.64 (12.4%) 112.17 (11.3%) 0.5% ( -20% - 27%) Fuzzy1 226.93 (7.3%) 228.67 (8.0%) 0.8% ( -13% - 17%) AndHighLow 1720.51 (5.1%) 1741.37 (5.2%) 1.2% ( -8% - 12%) {noformat} > Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit > > > Key: LUCENE-8180 > URL: https://issues.apache.org/jira/browse/LUCENE-8180 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Adrien Grand >Priority: Trivial > Labels: newdev > > Using Arrays.mismatch with a fixed-size array full of zeros might help find > the next long that is not 0 faster. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8180) Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit
[ https://issues.apache.org/jira/browse/LUCENE-8180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16374747#comment-16374747 ] Uwe Schindler commented on LUCENE-8180: --- Hi, this is interesting. How is java.util.Bitset internally implemented in Java 9. Does it use that method? > Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit > > > Key: LUCENE-8180 > URL: https://issues.apache.org/jira/browse/LUCENE-8180 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Adrien Grand >Priority: Trivial > Labels: newdev > > Using Arrays.mismatch with a fixed-size array full of zeros might help find > the next long that is not 0 faster. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8180) Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit
[ https://issues.apache.org/jira/browse/LUCENE-8180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16370402#comment-16370402 ] Adrien Grand commented on LUCENE-8180: -- I was mostly thinking about the moderately sparse case (eg. about 1/30th of bits set) on a large index with matches that are not uniformly spread across the doc ID space. I can't tell how common it is but I wouldn't be surprised that it was not so uncommon, and in that case there could be some long runs of zeros. > Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit > > > Key: LUCENE-8180 > URL: https://issues.apache.org/jira/browse/LUCENE-8180 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Adrien Grand >Priority: Trivial > Labels: newdev > > Using Arrays.mismatch with a fixed-size array full of zeros might help find > the next long that is not 0 faster. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (LUCENE-8180) Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit
[ https://issues.apache.org/jira/browse/LUCENE-8180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16370082#comment-16370082 ] Robert Muir commented on LUCENE-8180: - I'm not sure this would help unless the bitset was extraordinarily sparse, in which case fixedbitset isn't a great representation, and processing is going to be fast regardless because of the sparsity? > Explore using (Future)Arrays.mismatch for FixedBitSet.nextSetBit > > > Key: LUCENE-8180 > URL: https://issues.apache.org/jira/browse/LUCENE-8180 > Project: Lucene - Core > Issue Type: Improvement >Reporter: Adrien Grand >Priority: Trivial > Labels: newdev > > Using Arrays.mismatch with a fixed-size array full of zeros might help find > the next long that is not 0 faster. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org