Are NvOFBuffers special?

Hi everyone,

I’m working on processing some data with Nvidia OF and wondered about the GPU buffers needed to work with Optical Flow. As far as I can tell they are created using “nvOFCreateGPUBufferCuda” and are stored in a different (much lower) address range than regular cuda memory.

I’d like to know whether those memory allocations are special in some regard and what limitations they have opposed to regular memory allocated with cuMemAlloc.

Can they be used as arguments of CUDA kernel calls or are NvOF executions always necessarily preceded and followed by memCpy to and from these buffers?
Are they physical memory or are they memory mapped IO to interface the special hardware?

Best regards,
Michael F.