mccullocht commented on PR #15903: URL: https://github.com/apache/lucene/pull/15903#issuecomment-4238589564
@shbhar for performance some things to consider: * The core cost at 8 bits is going to be int8 dot product in both cases. OSQ has a more complicated/expensive correction to the dot product but I doubt it's 30% slower than the TQ correction. Vector incubator code paths are very touchy and it's possible that if you swapped the dot product implementations either way (your dot product on OSQ or lucene dot product on TQ) that performance would even out. It's hard to say unless they use the same interface and support the same set of vector sizes. * Consider microbenchmarking quantization. Quantization for the query will be repeated across all segments and in real workloads if your quantizer is significantly slower it may show up. Keep in mind that OSQ quantization is not SIMD accelerated today but could easily be. -- 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]
