Using OpenVX with GPU, Channelextract nodes cannot process at the same time

[Hayashi__Noriyoshi]

I’m tring to accelerate image processing using OpenVX with GPU. But it is slower than CPU.
I analyzed profile with Vtune, and I noticed channelextract nodes didn’t process 3 channels at the same time. Why is this processing divided? Is it better to process channelextract nodes on CPU than GPU?

part of code

vxCreateImageFromHandle(context, VX_DF_IMAGE_RGB, &frameFormat, &frameData, VX_IMPORT_TYPE_HOST),

vxCreateVirtualImage(graph, inputWidth, inputHeight, VX_DF_IMAGE_U8),

vxCreateVirtualImage(graph, inputWidth, inputHeight, VX_DF_IMAGE_U8),

vxCreateVirtualImage(graph, inputWidth, inputHeight, VX_DF_IMAGE_U8),

vxChannelExtractNode(graph, images[0], VX_CHANNEL_R, images[1]),

vxChannelExtractNode(graph, images[0], VX_CHANNEL_G, images[2]),

vxChannelExtractNode(graph, images[0], VX_CHANNEL_B, images[3]),

[Shubha_R_Intel]

Dear Hayashi, Noriyoshi,

Unfortunately this is the wrong forum for OpenVX questions. Please use the below forum for your OpenVX questions:

https://community.khronos.org/c/openvx

This forum is targeted to Model Optimizer and Inference Engine (OpenVino) questions.

Thanks,

Shubha