General Add a configuration option for passing ccbin to nvcc Differences with NVIDIA implementation cuLaunchKernel and similar should throw INVALID_VALUE when any dimension is 0