Efficient Complex High-Precision Computations on GPUs without Precision Loss