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 2016/12/06 20:26:58 UTC

[jira] [Commented] (SYSTEMML-1142) ParFor execution crashes bc of singlethreaded warning in LibMatrixDNN

    [ https://issues.apache.org/jira/browse/SYSTEMML-1142?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15726607#comment-15726607 ] 

Niketan Pansare commented on SYSTEMML-1142:
-------------------------------------------

Two points:
1. The commit https://github.com/apache/incubator-systemml/commit/8c069ed8e54c6941b25ccf8ede41389875e25d0b deleted warnSingleThreaded() as it was only added for initial performance testing.
2. To answer your question "why this executes with one thread only": Almost all our CP instructions receive a parameter #number of threads which is provided by parfor optimizer. Please see: https://github.com/apache/incubator-systemml/blob/master/src/main/java/org/apache/sysml/hops/Hop.java#L69

> ParFor execution crashes bc of singlethreaded warning in LibMatrixDNN
> ---------------------------------------------------------------------
>
>                 Key: SYSTEMML-1142
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1142
>             Project: SystemML
>          Issue Type: Bug
>          Components: ParFor
>         Environment: ParFor Spark on Cluster with DeepLearning usecase
>            Reporter: Felix Schüler
>
> When using functions from the LibMatrixDNN in a Spark ParFor block an exception is thrown by the function `warnSingleThreaded()` due to single threaded execution of functions such as conv2d. 
> Disabling this warning resolves the crash but I am not sure this is the right way to go there. The question is also why this executes with one thread only and how to improve here.
> cc [~niketanpansare]



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