OpenCL

[jsollie1]

Hello everyone,

I'm playing with a (currently very abstract) idea of pattern recognition software for images.

I'd like it to be implemented in OpenCL.

Is there any way to ask the VCN decoder to decode a JPEG image and put it in a opencl Image object?

I know you can use VAAPI / Vulkan to decode the jpeg in hardware, but transferring it back to main memory, embedding in a openCL Image object, and uploading to GPU again is probably way slower than decoding on the CPU.

Any paths I could follow?

[dipak]

Hi @jsollie1 ,

You may check the Advanced Media Framework(AMF) SDK available here:

https://github.com/GPUOpen-LibrariesAndSDKs/AMF 

https://gpuopen.com/advanced-media-framework/

For AMF related support, please use it's Github page: https://github.com/GPUOpen-LibrariesAndSDKs/AMF/issues

Thanks.

[jsollie1]

This looks like what I need indeed, but I see nowhere an option to upload a jpeg binary, only video streams seem to be supported.

Are you sure it would be in there?

[dipak]

Yes, it supports video streams. I am not sure if a single jpeg file can be uploaded for decoding. However, AMF supports "MJPEG" decoding, so I wonder whether it can be used in this case.

Anyway, for AMF related query, I would suggest that you post here:  page: https://github.com/GPUOpen-LibrariesAndSDKs/AMF/issues  

Thanks.