mengxr commented on issue #27546: [SPARK-30773][ML]Support NativeBlas for level-1 routines URL: https://github.com/apache/spark/pull/27546#issuecomment-587132383 @yma11 Thanks for re-running the benchmark! Could you confirm whether AVX is used by MKL/OpenBLAS? I think that is why it is faster on level-1 on large vectors. It seems reasonable to switch between f2jBLAS and nativeBLAS based on vector size (256?). There is still one scenario we need to check. I don't think users will remember limiting native BLAS threads to 1. They might have 8 concurrent tasks running on a 8-core worker while each using multi-thread BLAS. I saw cases before that it significantly degrade the performance. It might be worth running a benchmark for this scenario.
---------------------------------------------------------------- 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] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
