Coarse grain computation-communication overlap for efficient application-level checkpointing for GPUs