The Scheduling Problem in High-Level Synthesis: Extended 0/1 LP Model and Solution Methods

Abstract

As demonstrated by the activity in this area high-level synthesis — often referred to as behavioral synthesis — is becoming an increasingly popular research topic. A main part of high-level synthesis is scheduling where each operation from a data flow graph is assigned to a control step besides several constraints. We have formulated the scheduling problem as a 0/1 LP. For the first time we have extended the model to different execution times and multifunctional function units. To decrease the computation time we have developed a method for constructing new lower bounds with the help of LP (!). The results show that computation time is superior to comparable approaches though our model is able to describe a larger set of problems.