yma11 commented on issue #27546: [SPARK-30773][ML]Support NativeBlas for level-1 routines URL: https://github.com/apache/spark/pull/27546#issuecomment-597114153 > I think the two remaining issues are: > > * Should this threshold be configurable, only within `mllib` at least? > * Should the default / hard-coded value be larger, to be 'safer'? > > Possibly still partly configurable, yes; I don't feel strongly about it. > @mengxr 's last question was, how does this compare on older hardware, like Ivy Bridge or similar, maybe? I think we're aiming for a value that should not cause a slowdown on hardware people might be realistically running now, especially if they haven't tuned BLAS threads. That number might be higher than 256 and I expect it is, but just guessing. I am trying to get perf data of an older IA platform to double check whether 256 should be a safe threshold, even it's not all the case but it can make us more confident. Not finished yet and will update you once get the result.
---------------------------------------------------------------- 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]
