uschindler commented on PR #12311: URL: https://github.com/apache/lucene/pull/12311#issuecomment-1557873260
> > > btw, i think there's a real bug in the SPECIES_PREFERRED stuff that makes testing such degenerate cases _really difficult_. You should be able to just pass `-XX:MaxVectorSize=8` or `-XX:UseAVX=0` or similar to test it out: but this won't change SPECIES_PREFERRED, only make it dog slow. its like it gets the wrong information from the compiler. > > > the only way i know right now is to spin up QEMU and take cpu flags away :) > > > > > > Let's open a bug report. I can do it in the openjdk bug tracker. > > For testing, it would certainly be incredibly useful to have downward configurability over the PREFERRED species. Not only for testing. It is a bug. If somebody for example setups `UseAVX=2` like Elasticsearch does/did by default, it still configures the preferred species to 512 bits. As Hotspot can't handle this, it will use the fallback code and that's slow like hell. Actually we checked the code. It looks like the preferred species are dynamically initialized, bit the VM option is not taken into account. I think it call the wrong macro when it gets maximum lane size. -- 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
