memory management and multiple GPU in CUDA

Question

I am working on a class to manage the host and the device memory in one single place, including RAII and synchronisation.

When I allocate the memory on the device via