mapleFU commented on issue #43687: URL: https://github.com/apache/arrow/issues/43687#issuecomment-2290861706
I'm talk with @zanmato1984 and learned from that. I'm sorry for be confusing yesterday My original problem is: Why `AddMinMaxAvx512AggKernels add AVX2`. And I've learned that: * The SIMD Level is the "lowest level a kernel requires", not the "exact level a kernel requires". A easy to vectorized kernel would be fast even if it mark as `SimdLevel::None` * More Simd Kernel would be useless in some case. When user enable `-march`, we might generate a good code in `None`. Here I don't know why it's separate to multiple flags. Maybe the libraray wants to compile it separately? However, the problem here is that: * `compute/kernels/aggregate_basic_avx512.cc` is compiled with forcing avx512 flags, so, "the minimal level requires" for it might be AVX512 rather than AVX2? I'm sorry for being irritable lastday night -- 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]
