可重构嵌入式系统软硬件划分方法的研究

发布时间：2019-01-22 18:06

【摘要】：随着大规模集成电路技术的发展和现代嵌入式系统发展的需要，以FPGA为代表的可重构硬件被越来越多的嵌入式系统采用。由于可重构硬件兼具ASIC的计算速度和微处理器的灵活性，所以可重构系统被认为是一种高效的计算平台。实践证明，将传统的嵌入式系统的软硬件划分方法应用到可重构嵌入式系统的软硬件划分上显然是不可行的。在含有可重构硬件的嵌入式系统结构中，考虑到可重构器件在硬件结构上与微处理器核的差异，先要合理有效的划分输入的应用程序，这种划分包括空域划分以及时域划分，为了使可重构器件能够发挥重构的特性，不仅要把决定系统任务是在硬件域上还是软件域上实现，对于可能指派到可重构器件上的任务要考虑将其划分成在时间域上不能重叠，，同时划分过程中还必须采取措施以减小由重构带来的延迟。在研究了可重构硬件具有的新特性的基础上，论文对实际应用中常用的软硬件划分算法进行了研究，从而提出了一种适用于可重构嵌入式系统的软硬件划分算法，并在算法中采用调度算法以评价划分结果。论文根据课题需求设计了合适的可重构嵌入式系统的作为系统模型，然后采用有向无环图(DAG)来描述任务图表示任务之间的约束关系。接着创新性提出一种遗传/蚂蚁（GAMMAS）算法融合策略，将由DAG描述的特定应用映射到设计好的结构模型的软硬件平台上；关于可重构嵌入式系统内每个处理单元上任务和通信的开始执行时间和顺序的设定，是利用配置预取策略以及任务调度方法去确定，保证划分结果可以满足系统任务之间的预先设计的控制和数据依赖关系，最终得到系统性能整体优化的结果。论文最后对提出的融合策略以及采用的调度算法编写了验证程序。从实验数据可以分析，GAMMAS融合算法在求解精度上优于GA和MMAS算法，配置准备时间也因为采用配置预取和的调度方法而被有效降低，从而在整体上使任务图的整体完成时间被减少，达到了应用任务图到可重构系统的时空映射的目的。
[Abstract]:With the development of LSI technology and the development of modern embedded systems, reconfigurable hardware represented by FPGA has been adopted by more and more embedded systems. Reconfigurable hardware is considered to be an efficient computing platform because of both the computing speed of ASIC and the flexibility of microprocessor. It is proved that it is not feasible to apply the traditional software and hardware partitioning method of embedded system to the partition of software and hardware of reconfigurable embedded system. In the embedded system structure with reconfigurable hardware, considering the difference between reconfigurable device and microprocessor core, the input application program should be partitioned reasonably and effectively, which includes spatial partition and time domain partition. In order to make the reconfigurable device play the reconfiguration characteristic, we should not only decide whether the system task should be implemented in the hardware domain or the software domain, but also consider dividing the task that may be assigned to the reconfigurable device into the time domain without overlapping. At the same time, measures must be taken to reduce the delay caused by refactoring. Based on the study of the new characteristics of reconfigurable hardware, this paper studies the hardware and software partitioning algorithms commonly used in practical applications, and then proposes a software and hardware partition algorithm suitable for reconfigurable embedded systems. The scheduling algorithm is used to evaluate the partition results. According to the requirements of the project, this paper designs a suitable reconfigurable embedded system as a system model, and then uses directed acyclic graph (DAG) to describe the constraint relationship between tasks expressed by task graph. Then an innovative fusion strategy of genetic / ant (GAMMAS) algorithm is proposed, which maps the specific application described by DAG to the hardware and software platform of the designed structural model. The setting of the start time and sequence of tasks and communication on each processing unit in a reconfigurable embedded system is determined by using the configuration prefetching strategy and the task scheduling method. The partition result can satisfy the pre-designed control and data dependence between the tasks of the system, and the result of the overall optimization of the system performance is obtained. At the end of the paper, a verification program is developed for the proposed fusion strategy and the scheduling algorithm. From the experimental data, the GAMMAS fusion algorithm is superior to GA and MMAS algorithm in solving accuracy, and the configuration preparation time is reduced because of the scheduling method of configuration prefetch sum. Therefore, the overall completion time of the task graph is reduced, and the spatio-temporal mapping of the task map to the reconfigurable system is achieved.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP368.1

【参考文献】