hhr293 commented on PR #13363: URL: https://github.com/apache/gluten/pull/13363#issuecomment-6288133487
> Thanks. Will this optimize bring negative impact on avx2 code path? Thanks for pointing this out. On AVX2-only CPUs, the resolver selects the default clone, which retains the original AVX2 instructions but adds a function call. I tested this by forcing the default clone (GCC 13, AVX2 baseline). Full 256-value blocks showed no regression, while 64-value blocks had a 2–5% overhead. Smaller tail blocks showed larger relative overhead, but occur at most once per column chunk. If needed, we can select the inline scalar path for small blocks based on block size, while keeping the optimized path for larger blocks. -- 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]
