You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@systemml.apache.org by "Niketan Pansare (JIRA)" <ji...@apache.org> on 2016/12/19 19:34:58 UTC

[jira] [Closed] (SYSTEMML-1152) Explore the use of jcuda-vec to improve cellwise binary ops

     [ https://issues.apache.org/jira/browse/SYSTEMML-1152?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Niketan Pansare closed SYSTEMML-1152.
-------------------------------------
       Resolution: Won't Fix
    Fix Version/s: SystemML 1.0

Hi Nakul,

Since we anyways need to add custom kernels for other operations (such as rowsums, ...) and since corresponding kernels are already implemented, I would recommend not adding additional jcuda-vec dependency unless the current kernels are bottleneck.

Thanks,

Niketan.

> Explore the use of jcuda-vec to improve cellwise binary ops
> -----------------------------------------------------------
>
>                 Key: SYSTEMML-1152
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1152
>             Project: SystemML
>          Issue Type: Improvement
>            Reporter: Nakul Jindal
>             Fix For: SystemML 1.0
>
>
> We should explore the use of [jcuda-vec|https://github.com/jcuda/jcuda-vec] to replace the set of custom kernels we write. The custom kernel already does a branch/switch-case on each invocation. This is probably not the most efficient thing we can do.
> Pinging [~niketanpansare].



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)