[
https://issues.apache.org/jira/browse/LUCENE-9273?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17056823#comment-17056823
]
Ignacio Vera commented on LUCENE-9273:
--------------------------------------
With this change, there is a speed up of LatLonShape query its over points
because we are now only decoding two points instead of six:
||Approach||Shape||M hits/sec dev||M hits/sec base||M hits/sec diff||QPS
dev||QPS base||QPS diff||Hit count dev||Hit count base||Hit count diff||
|shapes|polyRussia|9.32|8.51|10%|2.66|2.42|10%|3508846|3508846|0%|
|shapes|polyMedium|3.59|3.27|10%|44.01|40.12|10%|2693559|2693559|0%|
|shapes|poly 10|41.24|39.14|5%|26.08|24.75|5%|355809475|355809475|0%|
|shapes|box|42.86|41.63|3%|43.62|42.36|3%|221118844|221118844|0%|
|shapes|distance|42.94|41.80|3%|25.23|24.56|3%|382961957|382961957|0%|
When indexing polygons, increase is very noticeable for intersect, within and
disjoint:
||Approach||Shape||M hits/sec dev||M hits/sec base||M hits/sec diff||QPS
dev||QPS base||QPS diff||Hit count dev||Hit count base||Hit count diff||
|point|intersects|0.00|0.00|-2%|319.39|324.38|-2%|2644|2644|0%|
|box|intersects|6.04|5.95|2%|41.11|40.47|2%|33081264|33081264|0%|
|distance|intersects|4.88|4.37|12%|17.15|15.36|12%|64062400|64062400|0%|
|poly 10|intersects|4.08|3.17|29%|15.56|12.09|29%|59064569|59064569|0%|
|polyMedium|intersects|0.32|0.21|50%|19.81|13.19|50%|528812|528812|0%|
|polyRussia|intersects|1.00|0.64|56%|4.07|2.61|56%|244848|244848|0%|
|point|contains|0.00|0.00|-1%|297.75|299.89|-1%|2644|2644|0%|
|box|contains|0.00|0.00|1%|37.70|37.31|1%|484|484|0%|
|distance|contains|0.00|0.00|1%|17.70|17.48|1%|406|406|0%|
|poly 10|contains|0.00|0.00|-1%|14.80|14.97|-1%|402|402|0%|
|polyMedium|contains|0.00|0.00|0%|15.47|15.47|0%|147|147|0%|
|point|within|0.00|0.00|0%|329.08|335.04|-2%|0|0|0%|
|box|within|0.51|0.43|18%|3.49|2.95|18%|32911251|32911251|0%|
|distance|within|0.91|0.76|20%|3.20|2.67|20%|63868270|63868270|0%|
|poly 10|within|0.81|0.77|4%|3.08|2.95|4%|58873224|58873224|0%|
|polyMedium|within|0.05|0.05|12%|3.39|3.03|12%|522739|522739|0%|
|polyRussia|within|0.61|0.51|18%|2.49|2.10|18%|244661|244661|0%|
|point|disjoint|264.00|264.97|-0%|20.05|20.13|-0%|2962178156|2962178156|0%|
|box|disjoint|184.56|184.75|-0%|14.18|14.19|-0%|2929099536|2929099536|0%|
|distance|disjoint|124.87|117.81|6%|9.69|9.15|6%|2898118400|2898118400|0%|
|poly 10|disjoint|119.71|101.56|18%|9.28|7.87|18%|2903116231|2903116231|0%|
|polyMedium|disjoint|135.92|109.35|24%|10.34|8.32|24%|433924372|433924372|0%|
|polyRussia|disjoint|49.06|31.78|54%|3.80|2.46|54%|12920400|12920400|0%|
> Speed up geometry queries by specialising Component2D spatial operations
> ------------------------------------------------------------------------
>
> Key: LUCENE-9273
> URL: https://issues.apache.org/jira/browse/LUCENE-9273
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Ignacio Vera
> Priority: Major
> Time Spent: 10m
> Remaining Estimate: 0h
>
> This is a follow-up from an observation of [~jpountz] where it notice that
> regardless of the spatial operation we are executing (e.g Intersects), we are
> always calling the method component2D#relateTriangle which it would be less
> expensive if we have an specialise method for intersects.
> The other frustrating thing is that regardless of the type of triangle we are
> dealing with, we are decoding all points of the triangle. In addicting most
> of the implementation of component2D#relateTriangle contain code that check
> the type of triangle to then call specialise methods.
> In this issue it is proposed to replace the method component2D#relateTriangle
> by the following methods:
> component2D#intersectsTriangle
> component2D#intersectsLine
> component2D#containsTriangle
> component2D#containsLine
> For consistency we add as well the methods:
> component2D#withinPoint
> component2D#withinLine
> Finally, the resolution of the triangle type his added to the decoding of the
> triangle.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
