[
https://issues.apache.org/jira/browse/SYSTEMML-2034?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Matthias Boehm updated SYSTEMML-2034:
-------------------------------------
Description: The current implementation of sparse-dense maxpooling_backward
uses for every p-q combination a full scan over the sparse row just to find the
applicable range. We should instead flip this dense implementation to a "real"
sparse implementation that scans the sparse input just once.
> Performance issue sparse-dense maxpooling backward
> --------------------------------------------------
>
> Key: SYSTEMML-2034
> URL: https://issues.apache.org/jira/browse/SYSTEMML-2034
> Project: SystemML
> Issue Type: Bug
> Reporter: Matthias Boehm
>
> The current implementation of sparse-dense maxpooling_backward uses for every
> p-q combination a full scan over the sparse row just to find the applicable
> range. We should instead flip this dense implementation to a "real" sparse
> implementation that scans the sparse input just once.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
