jmalkin commented on issue #20: URL: https://github.com/apache/datasketches-python/issues/20#issuecomment-1865127100
I completed the conversion, but density showed a significant performance regression. Investigating, it seems like the culprit was too many vector conversions in ways that pybind11 can avoid through the use of thinly-wrapped numpy arrays -- something Nanobind doesn't do (in favor of a more source-agnostic model). To address that I'm just wrapping the std::vector.data() in a numpy object whenever we need to call the kernel. It limits us to numpy kernels (which was the case before) but with that change I slightly beat pybind11 times. Vector of KLL sees a slight performance regression. I have not found a solution there. -- 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]
