摘要
在编译过程中,为了对运行于可重构计算系统上的应用程序自动地进行软硬件划分,提出一种基于模拟退火和链式调度的划分算法,该算法将程序的划分和调度结合起来,采用软硬件协同工作的方式,利用系统中微处理器核与可重构硬件之间的并行性加速程序的执行,同时在调度过程中,通过配置预取减少系统的重构开销.实验结果表明,相对于现有软硬件自动划分方法,该算法可以平均减少25%左右的程序执行时间.
In order to automatically complete the hardware-software partitioning of the application programs running on reconfigurable computing systems at compile time, a simulated annealing and list scheduling based partitioning algorithm (SALSPA) is proposed, which combines the program partitioning and scheduling. Hardware-software cooperation and the characteristic that the micro-processor core and reconfigurable hardware in the system can work in parallel were utilized to reduce the application's execution time. Also, configuration prefetch was used during scheduling to reduce the system's reconfiguration overhead. Experiments show that compared with the current automatic hardware-software partitioning methods, the proposed algorithm can reduce 25% of the programs' execution time on the average.
基金
高校博士点基金(20050358040)
安徽省自然科学基金(070412030)资助
关键词
可重构计算
软硬件划分
模拟退火
链式调度
配置预取
reconfigurable computing hardware-software partition simulated annealing list scheduling configuration prefetch