Using NvOFFRUC with cuda runtime / cudaArray

I’m having a hard time getting FRUC running in a plugin where I need to pass in a cudaArray/surface and write to a cudaArray/surface. The FrameGenerator/BufferManager in the example file uses the cuda driver API, and interop between it and the runtime is proving to be quite difficult.

I have been able to compile a plugin using cudaArrays as resources, and it appears that NvOF is running, but when I pass a cudaArray pointer to stOutParams.stFrameDataOutput.pFrame I just get a nullptr after running NvOFFRUCProcess.

I’ve also tried using the driver API and the provided FrameGenerator/BufferManager, but I haven’t been able to copy the interpolated texture of type CUarray to a cudaArray which is what the output of my plugin requires.

Any guidance would be greatly appreciated.