vertexclique commented on pull request #8665: URL: https://github.com/apache/arrow/pull/8665#issuecomment-770385902
For some operations, it will use wider registers and their transfers with avx512, but not all algorithms are expressable using other simd sets. Main idea was instead of creating ordinary instructions of both feature sets, creating fast operations on a specific collection of data. packed_simd or stdsimd doesn't generate the optimal ordering as a code—neither some intrinsics in the language's core. Useful intrinsics are bound to llvm procedures. Not compiler optimized ones. AVX512 set is one of those sets. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected]
