硬件多路映射的软硬件划分算法研究

发布时间：2018-10-07 20:54

【摘要】：随着嵌入式系统的不断发展，软件和硬件结合的越来越紧密，传统设计方法已经无法满足越来越复杂的设计需要。为了克服传统设计方法的不足，嵌入式设计者们提出并逐渐完善软硬件协同设计方法。软硬件划分技术是软硬件协同设计中的关键技术之一，它是指在系统设计时，将系统各项任务划分到软件或者硬件上实现，划分结果直接决定系统设计的优劣。因此研究嵌入式系统的软硬件划分技术，，具有十分重要的应用价值。目前对软硬件划分的研究主要集中在二元(binary partitioning)划分上，这种划分方式默认了系统中的每项任务有一种软件实现方式和一种硬件方式，而忽略了一项任务可能存在多种硬件实现方式的可能性，即硬件多路映射问题。本文对硬件多路映射问题进行研究，并比较局部搜索算法BUB与遗传算法，通过实验发现遗传算法能够得到更好的划分结果。在进一步研究过程中，针对遗传算法局部搜索能力较弱的情况以及硬件多路映射问题的特点，本文采用强化学习方法设计了变异算子以代替标准遗传算法中随机变异的方式，使不同的染色体能够自适应的选择的动作进化，提高了遗传算法的局部搜索能力和收敛速度。经实例验证，改进的遗传算法的收敛速度和最优解都优于标准遗传算法。多次运行表明该算法具有较强的稳定性，具有很好的效果。
[Abstract]:With the development of embedded system and the combination of software and hardware, the traditional design method can not meet the needs of more and more complex design. In order to overcome the shortcomings of traditional design methods, embedded designers put forward and gradually improve the hardware and software co-design method. Software / hardware partitioning technology is one of the key technologies in hardware and software co-design. It means that the tasks of the system are divided into software or hardware when the system is designed, and the partition results directly determine the merits and demerits of the system design. Therefore, it is very important to study the software and hardware partitioning technology of embedded system. At present, the research on software and hardware partitioning is mainly focused on binary (binary partitioning) partitioning, which acquires a software implementation mode and a hardware mode for each task in the system. It ignores the possibility that a task may have a variety of hardware implementations, that is, the problem of hardware multiplexing. In this paper, the hardware multiplex mapping problem is studied, and the local search algorithm BUB is compared with the genetic algorithm. The experimental results show that the genetic algorithm can get better partition results. In the process of further study, in view of the weak local search ability of genetic algorithm and the characteristics of hardware multipath mapping, this paper designs a mutation operator to replace the random mutation in the standard genetic algorithm by reinforcement learning. The local search ability and convergence speed of genetic algorithm are improved by making different chromosomes self-adaptively selective motion evolution. An example shows that the convergence speed and optimal solution of the improved genetic algorithm are better than that of the standard genetic algorithm. Multiple runs show that the algorithm has strong stability and good effect.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.1;TP18

【参考文献】