Github user mengxr commented on the pull request:
https://github.com/apache/spark/pull/4453#issuecomment-74425432
From JBLAS (http://mikiobraun.github.io/jblas/javadoc/):
> jblas uses double and float arrays to store the matrix. Whenever you call
a native function, the array is first copied. This means that it doesnât make
much sense to call a native routine if its computation is linear in the size of
the data, but this includes most of BLAS Level 1 and Level 2. jblas therefore
uses Java implementation for things like vector addition, or even matrix-vector
multiplication and is therefore not as fast as native BLAS. Currently, Iâm
contemplating some caching schemes to improve performance here.
So using Level 1 and Level 2 BLAS in JBLAS won't trigger native calls, we
used to call JBLAS' Cholesky solver in ALS, which is a Level 3 LAPACK routine.
But with the new implementation, we switched to netlib-java. So we don't
actually use any JBLAS native calls in Spark.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
