结合片上数据布局和数据传输效率:一种针对multi-bank CGRA的优化算法
Joint Data Placement and Transformation: An Optimizing Algorithm for Multi-bank CGRA System
-
摘要: 多bank的粗粒度可重构处理器结合了高性能和高灵活性的优点, 但是现有的编译技术难以充分利用其多bank存储资源, 本文在同时考虑数据放置和数据传输效率的基础上, 提出了一种可以从优化整体运算性能的一种算法, 并建立模型检验优化效果, 结果显示, 优化后的可重构处理器效率是现有优化算法的2.74倍.Abstract: Multi-bank Coarse-grained reconfigured processor combines the advantages of high performance and high flexibility, but the existing compiler technology is difficult to make full use of the Multi-bank memory resources, this paper proposes an efficient optimizing algorithm on optimizing the data placement and efficiency of data transformation simultaneously, then we build a model to evaluate the performance, the result shows our approach is capable of optimizing the performance to 2.74x of the state-of-the-art method.