基于任务分组的动态可重构结构编译方法

Compilation approach for dynamically reconfigurable architecture based on grouped tasks

针对动态可重构结构的优化编译问题,提出了一种基于任务分组的编译方法.在应用算法分割成多个任务的前提下,建立应用算法的任务流图.根据任务之间的关联数据量和局部数据存储器的容量,确定任务的执行顺序.在使装入任务的配置数据和执行任务在时域重叠的前提下,将任务分成不同的组合.对每个任务组合进行循环变换,减小配置数据的读取次数,提高了应用算法的性能.实验结果表明,该编译方法能有效地提高应用算法在动态可重构结构上的执行速度.

A new compilation approach based on grouped tasks was proposed for optimal compilation of dynamically reconfigurable architecture. By partitioning an application algorithm into tasks, a task flow graph was established. Based on communication data between tasks and the size of local data memory, the executing order of tasks was determined. Tasks were grouped with loading configuration data and executing tasks overlapping in time domain. For each grouped tasks, loop transformation was applied to diminish the frequence of reading configuration data, which resulted in improved application performance. Experiment result showed that the compilation approach can efficiently improve the execution speed of the applied algorithm on dynamically reconfigurable architecture.