Research On Hardware/Software Partitioning Technology Of Embedded System

Hardware/Software partitioning is one of the critical step in Hardware/Software Codesign flow, and has very important influence on the final design. The three decisions that must be made during partitioning are PE allocation,task assignment and scheduling. According to the target architecture the HW/SW partitioning can be classfied as binary partitioning, multi-way partitioning and dynamic reconfigurable system partitioning. In terms of the number of optimizing objects, it can be classfied as single object partitioning and multi objects partitioning. The dissertation studied the HW/SW partitioning of embedded computer system and including four aspects.Firstly a constructive partitioning algorithm based on list scheduling for single objective binary partitioning is proposed. The algorithm uses system resource criticality and node's local property as evaluating measurement, determines the system task's mapping and beginning time during node travelling, and dynamically adjusts the mapping objective to avoid being trapped into local minima.Then the single objective multi-way partitioning is studied, according to the features of the multi-way partitioning and existing research, an approach based on global searching is presented, which use genetic algorithm to solve PE allocation problem and the task assignment and scheduling are left to the heterogeneous task scheduling algorithm. All individual's partitioning are accomplished in one loop.Thirdly a multi objective multi-way partitioning algorithm based on Pareto Ranking is proposed, which combines genetic algorithm with Pareto methodology, uses individual's Pareto rank and sharing function to determine its fitness, generates new population with son population and father population together, searches the solution space by genetic algorithm.Finally, the partitioning of dynamic reconfigurable system is investigated and a mapping algorithm based on J. Noguera's work is proposed. In order to increase the resource utilization, reduce reconfigure latency and accelerate the system running speed, the algorithm takes into account of the reconfigure time parameters and adopts context prefetching technique during task's mapping.