kaivalnp opened a new issue, #16556: URL: https://github.com/apache/lucene/issues/16556
### Description Off-heap float vector bulk-scoring was introduced in #14980, and showed some nice performance improvements! (vectors are scored in chunks of a specific machine-dependent size to leverage SIMD instructions, and if I understand correctly, this optimization amortizes the cost of loading chunks of the query vector _across N documents_, instead of loading it _once per document_). There is currently an open issue to enable off-heap bulk-scoring for other vectors (byte vectors and quantized versions): #15155. I wonder if we should move the bulk-scoring APIs one level lower (from [`RandomVectorScorer`](https://github.com/apache/lucene/blob/2822b28b5ab042672d6cb569589128e5dcafa648/lucene/core/src/java/org/apache/lucene/util/hnsw/RandomVectorScorer.java#L37-L54) to [`VectorUtilSupport`](https://github.com/apache/lucene/blob/2822b28b5ab042672d6cb569589128e5dcafa648/lucene/core/src/java/org/apache/lucene/internal/vectorization/VectorUtilSupport.java#L25)) so that the native vectorization provider (#15508) can also benefit from it? (which I think was added to make use of specific instructions that the JVM cannot compile to) -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
