Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
Novice
50 Views

InferencePlugin::LoadNetwork fails with HETERO plugin

Things seem to work fine, as long as I call getPluginByDevice("CPU").

When getPluginByDevice("HETERO:GPU,CPU") is used, getPluginByDevice() works fine, but the call to LoadNetwork fails with exception in loadNetwork:

Cannot find plugin to use :Tried load plugin : clDNNPlugin,  error: Plugin clDNNPlugin cannot be loaded: cannot load plugin: clDNNPlugin from : Cannot load library 'libclDNNPlugin.so': [path redacted]

Looking at libclDNNPlugin.so, it seems to depend on libOpenCL.so.1, which isn't part of the SDK. I suspect that is the reason for the failure, but even so, questions remain:

- which OpenCL should be used with libclDNNPlugin.so?
- My assumption was, Hetero plugin handles exactly this sort of failover (for example, no GPU is present). If it doesn't, what is exactly its point?

 

Thanks for your help!

0 Kudos
2 Replies
Highlighted
New Contributor I
50 Views

I assume you are using the latest Intel Openvino toolkit  (RC1 2019).

The OpenCL comes along with toolkit but we need to manually install using below scripts. 

<OPENVION INSTALL DIR>/openvino_2019.1.094/install_dependencies/install_NEO_OCL_driver.sh intel-opencl_19.04.12237_amd64.deb

or

Refer link: http://docs.openvinotoolkit.org/2019_R1/_docs_install_guides_installing_openvino_linux.html#set-the-...

Thanks,

-a

 

 

0 Kudos
Highlighted
Novice
50 Views

I guess my question is more regarding redistribution -- 

- can Hetero plugin be safely used on a system without an Intel GPU?

- can Hetero plugin be safely used on a system without an OpenCL driver deployed?

- is there an API to determine, at runtime, whether Hetero/clDNN plugins can be initialized and used?

0 Kudos