viirya commented on PR #5822: URL: https://github.com/apache/datafusion-comet/pull/5822#issuecomment-5611284554
Thanks for checking this so thoroughly and testing the singleton fast path. I agree that the conclusion should identify that opportunity directly. Updated in `6274853a3`. I added the non-null empty-map control and re-ran all 41 cases. Forward normalization measured 9.8 µs for empty maps versus 1,042.3 µs for singletons per 8,192-row batch. The conclusion now names skipping singleton `slice` / `sort_to_indices` calls as a concrete optimization target and cites your fast-path experiment separately. I also qualified the empty/singleton comparison: the gap includes different index and output sizes, so it is not a pure measurement of the sort call. This run shows some between-stage drift as well, which is now disclosed rather than treating H, N and C as an exact additive decomposition. All 41 smoke cases and release Clippy for both benchmark targets pass. Production code remains unchanged so the optimization can be evaluated separately. -- 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]
