mcimadamore commented on PR #912: URL: https://github.com/apache/lucene/pull/912#issuecomment-1325173861
> > More investigation is required to understand where the difference comes from. > > After playing a bit more with the benchmark, it looks like 1ns is coming from the various alignment checks we need to perform - whereas ~0.5ns is coming from the segment liveness check. Of course, since this is a single call, the cost for these checks cannot be amortized. We can probably improve things a little (esp. when there's no alignment involved), but the numbers above reflect the fact that the memory segment implementation has more checks, so at the nanosecond scale, everything matters of course. With few optimization (to skip some checks when unaligned layouts are used, as well as to optimize other checks), I can get down to this (*): ``` Benchmark (ELEM_SIZE) Mode Cnt Score Error Units SegmentCopy.segment_copy 5 avgt 30 4.985 ± 0.030 ns/op SegmentCopy.segment_copy 10 avgt 30 4.853 ± 0.144 ns/op SegmentCopy.segment_copy 50 avgt 30 5.926 ± 0.066 ns/op SegmentCopy.segment_copy 100 avgt 30 8.272 ± 0.203 ns/op SegmentCopy.segment_copy 500 avgt 30 34.983 ± 0.226 ns/op SegmentCopy.segment_copy 1000 avgt 30 31.683 ± 0.161 ns/op ``` (*) To reduce the cost of the liveness check I had to use a shared session, as the confinement check seems to be the main offender there. That said, before we go deeper in this kind of micro-optimizations, it would be better to understand if this is really the underlying cause behind the performance regression. -- 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
