Batched Matrix Computations on Hardware Accelerators Based on GPUs