kaivalnp commented on PR #14874:
URL: https://github.com/apache/lucene/pull/14874#issuecomment-3071054127
To look into compiler inlining, I added
`-XX:CompileCommand=PrintInlining,*PanamaVectorUtilSupport.*` to the startup
arguments and saw:
```
@ 113 jdk.incubator.vector.AbstractSpecies::loopBound (9 bytes) force
inline by annotation
@ 5 jdk.incubator.vector.VectorIntrinsics::roundDown (23 bytes) force
inline by annotation
@ 125
org.apache.lucene.internal.vectorization.PanamaVectorUtilSupport::dotProductBody256
(110 bytes) failed to inline: callee is too large
```
..in a run _without_ `-reindex`
However if I performed a run _with_ `-reindex`, I saw lines like:
```
@ 113 jdk.incubator.vector.AbstractSpecies::loopBound (9 bytes) force
inline by annotation
@ 5 jdk.incubator.vector.VectorIntrinsics::roundDown (23 bytes) force
inline by annotation
@ 125
org.apache.lucene.internal.vectorization.PanamaVectorUtilSupport::dotProductBody256
(110 bytes) inline (hot)
```
..which means it is being inlined if the dot product function is hot enough
---
So, I tried to turn on explicit inlining using
`-XX:CompileCommand=inline,*PanamaVectorUtilSupport.*` in some runs:
| recall | latency | inlining | reindex |
| ------ | ------- | -------- | ------- |
| 0.962 | 2.540 | no | no |
| 0.962 | 1.210 | yes | no |
| 0.962 | 1.829 | no | yes |
| 0.962 | 1.808 | yes | yes |
I was able to squeeze performance in the run _without_ `-reindex` but not if
we create an index, I'll try to dig deeper on why this is happening..
--
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: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
