kaivalnp commented on PR #14874:
URL: https://github.com/apache/lucene/pull/14874#issuecomment-3070041519
FYI I was trying to see the compiled code for the dot product function using
`-XX:CompileCommand=print,*PanamaVectorUtilSupport.dotProductBody256` and saw
that performance improved on adding that flag (no reindexing in any run)
This PR without the flag:
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.962 1.711 1.711 1.000 100000 100 50 64
250 no 0.00 Infinity 0.12 1
77.44 292.969 292.969 HNSW
```
This PR with the flag:
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.962 1.171 1.172 1.001 100000 100 50 64
250 no 0.00 Infinity 0.12 1
77.44 292.969 292.969 HNSW
```
`main` without the flag:
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.962 2.503 2.502 0.999 100000 100 50 64
250 no 0.00 Infinity 0.12 1
77.44 292.969 292.969 HNSW
```
`main` with the flag:
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.962 1.194 1.192 0.999 100000 100 50 64
250 no 0.00 Infinity 0.12 1
77.44 292.969 292.969 HNSW
```
Perhaps the flag is forcing the function to be optimized by the JVM?
In this case, it should denote the latency in a long-running application?
(when the function is fully optimized by the compiler..)
If so, I don't see a lot of value in merging this PR (there isn't a large
improvement) -- but I'd love to make our benchmarks more robust and report
something representative of a long-running application! (these
non-deterministic compiler optimizations are too trappy)
--
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
