A hovering swarm particle swarm optimization algorithm based on node resource attributes for hardware/software partitioning

Hardware/software (HW/SW) partitioning is a vital aspect of HW/SW co-design. With the development of the design complexity in heterogeneous computing systems, existing partitioning algorithms have demonstrated inadequate performance in addressing problems relating to large-scale task nodes. This paper presents a novel HW/SW partitioning algorithm based on node resource attributes hovering swarm particle swarm optimization (HSPSO). First, the system task graph is initialized via the node resource urgency partitioning algorithm; then, the iterative solution produced by HSPSO algorithm yields the partitioning result. We present new initialization by combining node resource attribute information and introduce two improvements to the learning strategy of HSPSO algorithm. For the main swarm, a directed sample set and the addition of perturbation particles are designed to direct the main swarm's particle search process. For the secondary swarm, a dynamic particle update equation is formulated. Iterative updates are performed based on previous rounds' prior information using adaptive inertia weight. The experimental results illustrate that, in large-scale systems task graph partitioning with more than 400 nodes, when compared with mainstream partitioning algorithms, the proposed algorithm improves partitioning performance by no less than 10% for compute-intensive task graphs and no <5% for communication-intensive task graphs, with higher solution stability.

Automated summary

This paper presents a novel hardware/software partitioning algorithm based on node resource attributes. The algorithm initializes the system task graph and improves the partitioning performance through learning strategy enhancements. Experimental results show that the proposed algorithm achieves high partitioning performance and solution stability in large-scale systems task graph partitioning.