rmuir commented on PR #12737: URL: https://github.com/apache/lucene/pull/12737#issuecomment-1789905123
OK. sounds good to me @uschindler . I agree the user should be in control. And I'd like to avoid us adding tons of flags for this stuff unless we need it for testing. that's just my opinion. So i really like respecting JVM flags for stuff like UseFMA too. But existing UseFMA has bad defaults in the JDK, that is my problem. Sure you get a little extra precision "for free" if you are doing trivial stuff in a serial loop. but if you are trying to max out performance, it only makes sense for intel cpus from what i've seen so far. When trying to max out the throughput on intel, you need FMA on. When its amd or arm neon, you need to use mul+add to keep everything busier. And I don't recommend any user TURN OFF UseFMA for any reason, ever. Chances are they are doing more than just using lucene library in their application and this could have disastrous consequences and slowdowns. So I agree with a specific sysprop just for this case. It is a unique situation. -- 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
