Matrix-Matrix Multiplications on GPUs for Accelerating a Parallel Fluid Dynamics Code