rmuir commented on PR #12311: URL: https://github.com/apache/lucene/pull/12311#issuecomment-1555915665
Thanks for re-benchmarking @ChrisHegarty ! It has been a few years and an older JDK version since this stuff was developed. I had in mind to do a couple more runs eventually: * try it on aarch64 mac, i think it has no SVE but at least has NEON. at least make sure perf isn't horrible or something. * see if the manual loop unrolling is still necessary. it is a bit ugly/surprising we have to do this for performance. it is great you have AVX-512, I don't have it. I was worried about using `SPECIES_PREFERRED` and clock throttling, but I think lucene doesn't need to do anything about this? A user can pass `UseAVX=2` to the JVM if they want to avoid AVX-512 for that reason... i think -- 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
