You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@systemml.apache.org by "Nakul Jindal (JIRA)" <ji...@apache.org> on 2017/03/15 23:02:41 UTC

[jira] [Created] (SYSTEMML-1405) Explore the use of GPU instructions in a Parfor body (for Spark)

Nakul Jindal created SYSTEMML-1405:
--------------------------------------

             Summary: Explore the use of GPU instructions in a Parfor body (for Spark)
                 Key: SYSTEMML-1405
                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1405
             Project: SystemML
          Issue Type: Sub-task
          Components: Runtime
            Reporter: Nakul Jindal
             Fix For: SystemML 1.0


To use GPUs in distributed mode, we need to make it so that 
1. GPU instructions are called as part of the higher order functions invoked on the RDDs
2. They are shipped as part of the body of a parfor

This JIRA explores, implements & solidifies option 2.
Option 2 has an intuitive advantage over option 1. 
If operations in a parfor body are implemented on the GPU, the hope is that the data will stay on the GPU longer instead of ping-ponging between the host and the device memories.

We can later explore if and option 1 needs to be implemented.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)