I believe it should invoke nvcc to interpret the cuda commands properly instead of g++
I believe it should invoke nvcc to interpret the cuda commands properly instead of g++