[
https://issues.apache.org/jira/browse/SINGA-171?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
wangwei resolved SINGA-171.
---------------------------
Resolution: Implemented
> Create CppDevice and CudaDevice
> -------------------------------
>
> Key: SINGA-171
> URL: https://issues.apache.org/jira/browse/SINGA-171
> Project: Singa
> Issue Type: New Feature
> Reporter: wangwei
> Assignee: wangwei
>
> Device is a core component in SINGA v1, which malloc/free memory and executes
> Tensor functions.
> This ticket is going to create two specific devices:
> 1. CppDevice which execute functions written in c++11 on host CPUs.
> 2. CudaDevice which execute functions written in cuda on nvidia GPUs.
> In our design, each device could have multiple executors (e.g., cuda streams)
> to run functions in parallel.
> We simplify the implementation for this ticket to create one executor for
> CppDevice and CudaDevice. We also skip the scheduler and virtual memory
> manager in this ticket. Instead the device controls the memory and scheduling
> by itself.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
