Efficient heuristic algorithms for path-based hardware/software partitioning

Abstract

Hardware/software (HW/SW) partitioning is one of the crucial steps of co-design systems. It determines which components of the systems are implemented in hardware and which ones are in software. In this paper the computing model is extended to cater for the path-based HW/SW partitioning with the fine granularity in which communication penalties between system components must be considered. On the new computing model an efficient heuristic algorithm is developed, in which both speedup in hardware and communication penalty are taken into account. In addition, an efficient tabu search algorithm is also customized in this paper to refine the approximate solutions produced by the heuristic algorithm. Simulation results show that the heuristic algorithm runs fast and is able to produce high-quality approximate solutions. Moreover, the tabu search algorithm can further refine them to nearly optimal solutions within an acceptable runtime. The difference between the approximate solutions and the optimal ones is bounded by 0.5%, and it hardly increases with the increase of the problem size.