[GitHub] [spark] luhenry edited a comment on pull request #30810: [SPARK-33882][ML] Add a vectorized BLAS implementation

[GitBox <gi...@apache.org>]

luhenry edited a comment on pull request #30810:
URL: https://github.com/apache/spark/pull/30810#issuecomment-823660970


   > > Late LGTM.
   > > BTW, I am not familiar with the new vector api, is it possible to apply it in BLAS for sparse vec/mat?
   > 
   > This vector API provides access to hardware acceleration. So as long as you can express the sparse vec/matrix operations with hardware vectors, you should be able to use the Vector API. However, from my cursory glance at the implementation I’m BLAS.Scala, using hardware acceleration doesn’t seem trivial.
   
   @zhengruifeng I looked further into that today and what might be looking interesting is Intel MKL support for level-2 and level-3 operations on sparse vectors/matrices (https://software.intel.com/content/www/us/en/develop/documentation/onemkl-developer-reference-c/top/blas-and-sparse-blas-routines/inspector-executor-sparse-blas-routines.html). I'll research what's applicable to Spark, and how we could surface it. Feel free to reach out if you want to discuss it offline.


-- 
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:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org