[Request]: Add HIP interop support for AMF Compute

[TheNewSound]

Is your feature request related to a problem? Please describe. I would like to have better performing AMF Compute kernels with greater flexibility provided by the HIP runtime. To achieve efficient realtime machine learning on video feeds, eliminating CPU-GPU roundtrip.

Describe the solution you'd like Adding HIP interop with AMF, so HIP kernels can be scheduled before/after encoding/decoding. Especially a way to maintain shared memory contents across the processing of frames would be useful for temporal analysis. An intermediate solution to this could be to decode & process 5 frames per HIP kernel launch.

Describe alternatives you've considered none

Additional context Is adding HIP interop support something the AMF team can consider?

[LuisB79]

the amf team is as flexible as glass, i doubt they'll even read your post.

[MikhailAMD]

AMF team does read every post but not comment on plans, upcoming releases or possible features.

[LuisB79]

As glass.