[
https://issues.apache.org/jira/browse/SINGA-175?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15347865#comment-15347865
]
ASF subversion and git services commented on SINGA-175:
-------------------------------------------------------
Commit 077d13e8052aa92679909b619966481a383a651f in incubator-singa's branch
refs/heads/dev from [[email protected]]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=077d13e ]
SINGA-175 Add memory management APIs and implement a subclass using CNMeM
Add base memory pool class.
Implement two subclasses, CnMemPool and CudaMemPool.
Add test for the memory pools.
TODO replace Device* to std::shared_ptr<Device> to avoid memory error because
the order of destructing device and tensor are dynamic (device may be freed
before tensors)
> Add memory management APIs and implement a subclass using CNMeM
> ---------------------------------------------------------------
>
> Key: SINGA-175
> URL: https://issues.apache.org/jira/browse/SINGA-175
> Project: Singa
> Issue Type: New Feature
> Reporter: wangwei
>
> We have associated each Tensor associated with a reference counter for
> garbage collection. Tensor is then can be used without worrying about memory
> leak.
> To avoid frequently malloc/free gpu or cpu memory due to Tensor creation and
> deletion, we need a memory pool. This ticket is going to create a memory
> manager that manages a memory pool for each device.
> Particularly, we will implement a specific memory manager using the
> https://github.com/NVIDIA/cnmem for CudaGPU devices.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
