Thread-safety with GpuMat
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.
If none of the threads are modifying the GpuMat, is this considered thread safe generally? with cavaets? never?
The reference counting is atomic, so that part looks fine, but I wasn't sure if anything in particular with the cuda context might cause problems.