Optimal Schedules for High-Level Programming Environments on FPGAs with Constraint Programming

Abstract

Scheduling tasks on reconfigurable hardware is a well-known problem. Yet, the adoption of advanced scheduling strategies for reconfigurable systems is still low. We argue that a pragmatic solution not relying on low-level features like partial reconfiguration is feasible. Our theoretical framework describes reconfigurable hardware in a simple and abstract way. The constraints of a schedule are used to derive a constraint programming formulation. We present two heuristic algorithms based on list scheduling and on clustering, respectively. The model is evaluated and compared to partial reconfiguration using parameters from a previously observed LU decomposition on an FPGA. The losses are compared to a conventional, optimal approach. It can be integrated into existing technologies to aide the adoption of high-level FPGA programming environments.