Asynchronous Communication Schemes for Finite Difference Methods on Multiple GPUs