pitrou opened a new issue, #45681: URL: https://github.com/apache/arrow/issues/45681
### Describe the enhancement requested Currently, the sum function and related functions (such as mean, variance...) use pairwise summation to reduce catastrophic error accumulation when summing floating-point numbers. However, pairwise summation requires a non-trivial amount of temporary state (`O(log2 n)` FP numbers) while [enhanced Kuhan summation](https://en.wikipedia.org/wiki/Kahan_summation_algorithm#Further_enhancements) would require only a single FP number of temporary state. We should investigate whether replacing pairwise summation with enhanced Kahan summation produces results of the same (or higher) quality, with no performance regression. It would then open the door to using Kahan summation in the hash-aggregate variants of those functions, which currently use naive summation. ### Component(s) C++ -- 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: issues-unsubscr...@arrow.apache.org.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
