OpenCUDA+MPI: A Framework for Heterogeneous GP-GPU Distributed Computing