Reducing thread divergence in a GPU-accelerated branch-and-bound algorithm