Generating several solutions for the scheduling problem in high-level synthesis

This paper presents a new approach for the scheduling problem in High-Level Synthesis. Starting with a maximal parallel data flow graph, the method adds successively edges to the graph so that all hardware constraints are observed. Every existing schedule can be managed in such a way. The result is a bundle of solutions with different scheduling variants. All solutions observe the same time- and hardware constraints as the original schedule. Succeeding binding algorithms can use these degrees of freedom and are able to generate better solutions