zml1206 commented on code in PR #5750:
URL: https://github.com/apache/incubator-gluten/pull/5750#discussion_r1618195390
##########
backends-velox/src/main/scala/org/apache/gluten/backendsapi/velox/VeloxBackend.scala:
##########
@@ -376,13 +376,8 @@ object VeloxBackendSettings extends BackendSettingsApi {
} else {
t match {
// OPPRO-266: For Velox backend, build right and left are both
supported for
- // LeftOuter and LeftSemi.
- // FIXME Hongze 22/12/06
- // HashJoin.scala in shim was not always loaded by class loader.
- // The file should be removed and we temporarily disable the
improvement
- // introduced by OPPRO-266 by commenting out the following
prerequisite
- // condition.
-// case LeftOuter | LeftSemi => true
+ // LeftOuter.
+ case LeftOuter => true
Review Comment:
I was planning to remove `preferredBuildSide` in the next PR because it is
unnecessary and not as accurate as AQE. Like vanilla spark, its performance
improvement is compared to sort merge join, for left outer and join outer, if
both are shuffle hash joins, choosing a small table will not improve
performance.
I don’t know the logic of velox’s support for small table build. Shuffle
hash join is forced to be used here,
what I'm worried about is that for shuffle hash join of LeftSemi, build left
is much slower than build right.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
