Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6525 Discussions

[OpenVINO 2022.1 CPU Plugin] Blocked layout support for custom layers

youkis
Novice
1,369 Views

How can we use blocked layout such as nChw8c for custom layers? I'd like to use nChw8C on my custom layer since costly Reorder operations frequently occurred when I used NCHW.

I found the way on old 2018 version, but I couldn't find it on latest 2022.1 version. Could you tell me how to use the nChw8c on latest OpenVINO?

Related links:

- Performance Considerations for Custom Kernels and Custom Layers (old version)

https://www.intel.com/content/www/us/en/developer/articles/technical/openvino-custom-layers-support-in-inference-engine.html#inpage-nav-3

- Model Optimizer Extensibility (current version)

https://docs.openvino.ai/latest/openvino_docs_MO_DG_prepare_model_customize_model_optimizer_Customize_Model_Optimizer.html#

 

0 Kudos
1 Solution
5 Replies
Wan_Intel
Moderator
1,287 Views

Hi Youkis,

Thanks for reaching out to us. Referring to Pull request #6686 in OpenVINO™ GitHub, it indicated the feature to support Blocked layout was implemented in OpenVINO™ toolkit 2022.1. As part of this feature, you may try to load and run your model without code changes. Please let us know if you encounter any issues.

 

 

Regards,

Wan


0 Kudos
youkis
Novice
1,281 Views

Thank you for your quick reply, Wan. I know that the blocked layout is supported on 2022.1. But, I couldn't find the simple template like following identity op to enable nChw8c on a custom layer. I found Node class has getSupportedDescriptors and we can declare to use nChw8c layout. However, we have to use ov::op::Op inherited class for the custom layer and there is not getSupportedDescriptors function.

https://github.com/openvinotoolkit/openvino/blob/2022.1.0/docs/template_extension/new/identity.cpp

https://github.com/openvinotoolkit/openvino/blob/2022.1.0/src/plugins/intel_cpu/src/nodes/fullyconnected.cpp

0 Kudos
Wan_Intel
Moderator
1,269 Views

Hi Youkis,

Let us check with our engineering team, and we will update you once we've obtained feedback from them.

 

 

Regards,

Wan


youkis
Novice
1,202 Views

continued in github #12838

Wan_Intel
Moderator
1,164 Views

Hi Youkis,

Thanks for informing us that you would like to continue your conversation with our developer in this GitHub thread.

If you need any additional information from Intel, please submit a new question as this thread will no longer be monitored.

 

 

Regards,

Wan


0 Kudos
Reply