An algorithmic incremental and iterative development method to parallelize dusty-deck FORTRAN HPC codes in GPGPUs using CUDA