Auto-Generation of Parallel Finite-Differencing Code for MPI, TBB and CUDA