[jira] [Commented] (LUCENE-9837) try to improve performance of VectorUtil.dotProduct

["Michael Sokolov (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/LUCENE-9837?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17300952#comment-17300952 ] 

Michael Sokolov commented on LUCENE-9837:
-----------------------------------------

Nice work! 25% improvement for typical use case is great.

In future, I think we may also eventually want to provide a single byte/dimension version of vectors for use with HNSW since the loss of precision gets washed out in the KNN use case where you only care about relative distances and not the precise values of these distance operations. Probably the perf (certainly the index size) can be improved by doing that

> try to improve performance of VectorUtil.dotProduct
> ---------------------------------------------------
>
>                 Key: LUCENE-9837
>                 URL: https://issues.apache.org/jira/browse/LUCENE-9837
>             Project: Lucene - Core
>          Issue Type: Improvement
>            Reporter: Robert Muir
>            Priority: Major
>          Time Spent: 50m
>  Remaining Estimate: 0h
>
> This is the king of cpu usage for the nightly benchmark. Let's see if we can optimize it a bit.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org