[
https://issues.apache.org/jira/browse/LUCENE-5856?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robert Muir updated LUCENE-5856:
--------------------------------
Attachment: LUCENE-5856.patch
simple patch.
i tried to figure out a way to benchmark, so i used matchalldocs+filters of
different densities (and forced fixedbitset).
Gives me numbers like:
{noformat}
Task QPS trunk StdDev QPS patch StdDev
Pct diff
F50.0 14.76 (1.2%) 15.67 (0.1%)
6.1% ( 4% - 7%)
F100.0 20.28 (2.4%) 21.61 (0.0%)
6.6% ( 3% - 9%)
F99.0 20.05 (2.4%) 21.40 (0.1%)
6.7% ( 4% - 9%)
F95.0 19.21 (2.2%) 20.54 (0.2%)
6.9% ( 4% - 9%)
F90.0 18.40 (2.1%) 19.71 (0.3%)
7.1% ( 4% - 9%)
F75.0 16.39 (1.5%) 17.60 (0.6%)
7.4% ( 5% - 9%)
F20.0 27.04 (1.0%) 29.09 (0.1%)
7.6% ( 6% - 8%)
F10.0 42.09 (1.0%) 46.16 (0.5%)
9.7% ( 8% - 11%)
F5.0 56.77 (0.8%) 64.46 (0.6%)
13.6% ( 12% - 15%)
F2.0 70.43 (0.5%) 83.48 (0.7%)
18.5% ( 17% - 19%)
F1.0 76.42 (0.3%) 92.60 (0.7%)
21.2% ( 20% - 22%)
F0.5 80.01 (0.2%) 98.15 (0.6%)
22.7% ( 21% - 23%)
{noformat}
I suppose F0.5 looks the best, because it has less noise from priority queue
checks and other things going on.
> remove useless & 0x3f from *BitSet.get and company
> --------------------------------------------------
>
> Key: LUCENE-5856
> URL: https://issues.apache.org/jira/browse/LUCENE-5856
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Robert Muir
> Attachments: LUCENE-5856.patch
>
>
> java specification says:
> {quote}
> If the promoted type of the left-hand operand is long, then only the six
> lowest-order bits of the right-hand operand are used as the shift distance.
> It is as if the right-hand operand were subjected to a bitwise logical AND
> operator & (ยง15.22.1) with the mask value 0x3f (0b111111). The shift distance
> actually used is therefore always in the range 0 to 63, inclusive.
> {quote}
> and x86 works the same way.
> if we remove this, we just see less instructions with printassembly...
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
