[
https://issues.apache.org/jira/browse/SINGA-136?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15228526#comment-15228526
]
ASF subversion and git services commented on SINGA-136:
-------------------------------------------------------
Commit 9c2c29742fd27022b377b524028aa78fb64c888f in incubator-singa's branch
refs/heads/master from [~flytosky]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=9c2c297 ]
SINGA-136 Support cuDNN v4
Fix warnings from using data()/grad() for mutable values.
They are replaced with mutable_data/grad() functions.
Add batchnorm related files into Makefile.am.
The CPU version of batchnorm is empty now.
Note. If the cudnn.h is in cuda/include folder, then the --with-cudnn
is invalid as cuda/include is looked up first. TODO change the order of
these two paths during compilation.
> Support cuDNN v4
> ----------------
>
> Key: SINGA-136
> URL: https://issues.apache.org/jira/browse/SINGA-136
> Project: Singa
> Issue Type: Improvement
> Reporter: wangwei
> Assignee: Haibo Chen
> Labels: cuDNN
>
> Nvidia has release a new cuDNN release candidate for version 4, which
> improves the performance of convolutional layer. Batch normalization is also
> included.
> There should be minor updates for SINGA to use cuDNN v4.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
