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 2017/07/14 19:56:00 UTC
[jira] [Closed] (SYSTEMML-1767) Performance issues codegen rowwise
(column aggregation) w/ wide matrices
[ https://issues.apache.org/jira/browse/SYSTEMML-1767?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Matthias Boehm closed SYSTEMML-1767.
------------------------------------
> Performance issues codegen rowwise (column aggregation) w/ wide matrices
> ------------------------------------------------------------------------
>
> Key: SYSTEMML-1767
> URL: https://issues.apache.org/jira/browse/SYSTEMML-1767
> Project: SystemML
> Issue Type: Bug
> Reporter: Matthias Boehm
> Assignee: Matthias Boehm
> Fix For: SystemML 1.0
>
>
> On scenarios with wide matrices of millions of features, the codegen rowwise template shows performance issues due to unnecessary multi-threading which requires additional memory per thread for partial aggregation which leads to cache thrashing. We should similarly to the mmchain operator establish a threshold for maximum temporary results and fall back to sequential operations if this threshold is exceeded.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)