Github user VinceShieh commented on the issue:
https://github.com/apache/spark/pull/18936
Hi Sean, sorry for late reply. Yeah, actually we do have some performance
data on F2J vs. OpenBLAS. It seems there is no performance gain from openblas,
not even on the unit test level. We are thinking maybe we should setup and test
on another cluster in case the result is due to the environment-related issues.
we will sort out the data and post here later.
it's possible that openblas may not give us any benefit in this case, but
still here are my two cents:
1. like openblas, MKL blas is also free, easy to get and much well
documented.
2. it's much easier for users to use MKL blas than openblas, just download
it and put it where the system can find it, no installation required. On the
other hand, install openblas on each machine is still troublesome, cost us a
while to get it work.
3. MKL blas is highly optimized thus has much better performance
4. this PR just opens up an opportunity for users to choose different BLAS
impl, they can still go for F2J, so I think it'd be more user-friendly than
binding all the level1 blas to F2J.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
