You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@systemml.apache.org by "Matthias Boehm (JIRA)" <ji...@apache.org> on 2017/12/02 23:24:00 UTC

[jira] [Closed] (SYSTEMML-1723) Parallel nnz recompute for all multi-threade convolution ops

     [ https://issues.apache.org/jira/browse/SYSTEMML-1723?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Matthias Boehm closed SYSTEMML-1723.
------------------------------------
       Resolution: Fixed
         Assignee: Matthias Boehm
    Fix Version/s: SystemML 1.0

> Parallel nnz recompute for all multi-threade convolution ops
> ------------------------------------------------------------
>
>                 Key: SYSTEMML-1723
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1723
>             Project: SystemML
>          Issue Type: Task
>            Reporter: Matthias Boehm
>            Assignee: Matthias Boehm
>             Fix For: SystemML 1.0
>
>
> All multi-threaded convolution operations currently make an additional pass over the output in order to {{recomputeNonZeros}}, which is a sequential operation. Since all convolution operations parallelize over rows (with guaranteed disjoint results), we should recompute the non zeros locally for each row partition and simply aggregate the final meta data.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)