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

Posted to issues@systemml.apache.org by "Glenn Weidner (JIRA)" <ji...@apache.org> on 2017/09/09 02:09:02 UTC

[jira] [Updated] (SYSTEMML-1750) Optional dynamic recompilation for JMLC training

     [ https://issues.apache.org/jira/browse/SYSTEMML-1750?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Glenn Weidner updated SYSTEMML-1750:
------------------------------------
    Fix Version/s:     (was: SystemML 1.0)
                   SystemML 0.15

> Optional dynamic recompilation for JMLC training
> ------------------------------------------------
>
>                 Key: SYSTEMML-1750
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1750
>             Project: SystemML
>          Issue Type: Task
>            Reporter: Matthias Boehm
>            Assignee: Matthias Boehm
>             Fix For: SystemML 0.15
>
>
> There are scenarios where JMLC is used for training on moderately sized input. Due to the use of prepared scripts (which are compiled without size information) and forced singlenode execution type, this can lead to performance problems caused by poor plan choices. This task aims to (1) expose compiler configurations such as dynamic recompilation and multi-threading at JMLC API and (2) rework the recompilation framework for singlenode execution type.



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