rmuir commented on PR #12694: URL: https://github.com/apache/lucene/pull/12694#issuecomment-1770900939
This means the only way you can do this correctly, is to remove all 16-bit multiplications and all use of `short` completely and go straight from 8-bit to 32-bit with ZERO_EXTEND_B2I. It means suffering downclocking on avx-512 or shortening vectors in half. It means much slower ARM performance. If it gives better search results and it is worth the tradeoff, that is fine. I just want you to be aware of the tradeoffs because the benchmarks you have posted I think are unrealistic. -- 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