By using new developed GPU-based min-reduction data parallel primitive in the key step of the algorithm, higher efficiency is achieved.