[jira] [Updated] (LUCENE-7105) squeeze a little more perf out of LatLonPoint.newDistanceQuery

["Robert Muir (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/LUCENE-7105?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Robert Muir updated LUCENE-7105:
--------------------------------
    Attachment: LUCENE-7105.patch

Here is a patch. Its a 10.5% QPS improvement on Mike's London benchmark. I also see similar improvements doing queries around my house in geonames.

Changes:
* visit(int, byte[]) does range checks on individual points. This is cheap, removes "false positives" to speed up two-phase intersection, and BKD has read the data already anyway.
* range checks in visit(int, byte[]) and compare(min, max) just work on the binary form.
* 2nd phase iterator uses the two-stage haversin from the distance sort when possible to reject candidates faster.


> squeeze a little more perf out of LatLonPoint.newDistanceQuery
> --------------------------------------------------------------
>
>                 Key: LUCENE-7105
>                 URL: https://issues.apache.org/jira/browse/LUCENE-7105
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Robert Muir
>         Attachments: LUCENE-7105.patch
>
>
> This query can make use of some of the same optimizations we applied to the distance sort.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org