Capacity-achieving polar codes have grown attention in recent years, and the belief propagation (BP) algorithm is one way to decode them. With the help of min-sum approximation and G-matrix early termination schemes, it has been proven that complexity and computation can be further reduced without significant performance loss. Although G-matrix early termination is a strong stopping criterion, there are still some redundant cycles that need to be reduced, since G-matrix early termination only reduces the computation caused by redundant iterations but not the redundant cycles inside of one iteration. In this paper, a stage-stopping BP algorithm with round-trip scheduling is proposed to reduce the redundant computation. Different thresholds for different signal-to-noise ratio (SNR) regions are simulated to seek the proper combination.