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 2017/12/16 09:46:00 UTC

[jira] [Updated] (SYSTEMML-2046) Operations over large dense blocks

     [ https://issues.apache.org/jira/browse/SYSTEMML-2046?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Matthias Boehm updated SYSTEMML-2046:
-------------------------------------
    Description: 
This task aims to add support for large dense blocks to all amenable matrix block operations. In order to allow for incremental progress, we handle the following categories of operations separately:

* Matrix multiplications
* Unary and grouped aggregates
* Unary and binary element-wise operations
* Reorg operations (e.g., transpose, sort, removeEmpty)
* Left- and right indexing operations, incl update-in-place
* Deep learning operations (e.g., conv2d, maxpool)
* Codegen cell, row, outer operations
* Compressed linear algebra
* Remaining operations (e.g., parfor result merge)

After all categories of operations are re-implemented accordingly, the testsuite should be run with forced large dense blocks and constant block size of 10 in order to emulate the large dense block which otherwise would require input data >16GB.


> Operations over large dense blocks
> ----------------------------------
>
>                 Key: SYSTEMML-2046
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-2046
>             Project: SystemML
>          Issue Type: Sub-task
>            Reporter: Matthias Boehm
>             Fix For: SystemML 1.1
>
>
> This task aims to add support for large dense blocks to all amenable matrix block operations. In order to allow for incremental progress, we handle the following categories of operations separately:
> * Matrix multiplications
> * Unary and grouped aggregates
> * Unary and binary element-wise operations
> * Reorg operations (e.g., transpose, sort, removeEmpty)
> * Left- and right indexing operations, incl update-in-place
> * Deep learning operations (e.g., conv2d, maxpool)
> * Codegen cell, row, outer operations
> * Compressed linear algebra
> * Remaining operations (e.g., parfor result merge)
> After all categories of operations are re-implemented accordingly, the testsuite should be run with forced large dense blocks and constant block size of 10 in order to emulate the large dense block which otherwise would require input data >16GB.



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