Quasi-maximum Accuracy Floating-point Computations with GPGPU for Applications in Digital Signal Processing