Bound-oriented parallel pruning approaches for efficient resource constrained scheduling of high-level synthesis

As a key step of high-level synthesis (HLS), resource constrained scheduling (RCS) tries to find an optimal schedule which can dispatch all the operations with minimum latency under specific resource constraints. Branch-and-bound heuristics are promising to achieve such an optimal schedule quickly, since they can prune away large parts of infeasible solution space during the exploration. However, few of them are based on the prevalent multi-core platforms. Based on the bound information, this paper exploits the parallel pruning potentials from different perspectives and proposes various efficient techniques that can substantially reduce the overall RCS search efforts. The experimental results demonstrate that our approach can reduce the RCS time drastically.