NodeRank:An Efficient Algorithm for Hardware/Software Partitioning

Abstract

Hardware/software(HW/SW)co-design is a key technique for the development of modern embedded systems.HW/SW partitioning is a crucial step in HW/SW co-design that determines which components of the computer system are implemented on hardware and which ones on software.In this paper,we present an efficient algorithm for hardware/software partitioning:NodeRank.Formulating the HW/SW partitioning problem as a variant of 0-1knapsack problem with dynamic communication costs,NodeRank iteratively calculates the rank of each node,updates the expectation of communication costs,and thus generates the corresponding heuristic solutions to the problem.Experimental results show that,when the computation cost and communication cost are roughly of equal weight and the real-time constraint is loose,NodeRank is inferior to the state-of-the-art Tabu Search method at most by 1.2%for task graphs with edge to node ratio equal or greater than 2,but saves more than 95% running time on average.For communication-intensive cases,NodeRank outperforms Tabu Search by up to 3.5% and saves runtime over 75%.