I have a single thread that produces a GpuMat. I was looking at a design where I would pass that to N threads as a const GpuMat, and then let them operate on that matrix arbitrarily.