Efficient Algorithm for the Iteration Period Computation of Unfolded Synchronous Dataflow Graphs
Xue-Yang Zhu · 2018
Synchronous dataflow graphs (SDFGs) are widely used to model streaming applications. Unfolding is one of the most important techniques for performance optimization of SDFGs. It may reduce the iteration period (IP) without affecting functionality. We present a novel method to compute the IPs of SDFGs by state-space exploration, without converting them to their equivalent homogeneous SDFGs (HSDFGs), and without further unfolding the HSDFGs. The conversion procedures are time and space-consuming. We also consider the cases when there are resource constraints, which cannot be dealt with by existing methods. Combining with retiming technique, we further present a method to compute the reduced IP of unfolded SDFGs. Our experimental results show that the proposed method outperforms the existing methods significantly.