基于模板的粗粒度可重构处理器编译环境设计

A Template-Based Compiling System Design for Coarse -Grained Reconfigurable Processors

摘要: 粗粒度可重构处理器以优良的结构和出色的性能日益受到广泛的应用，其丰富的运算资源为应用程序高效并发执行提供了可能。本文面向REmus II粗粒度可重构处理器硬件架构，首次提出一种基于模板的程序编译方法。从应用程序中的计算密集型代码出发，定义了程序模板的格式和规范，建立模板开发的设计流程，设计了面向视频图像处理的程序模板库。在此基础上，面向程序员设计了基于模板的程序语言标注技术，并最终完成了相应编译环境的构建。实验结果表明，该编译环境有效提高了编译后代码在可重构处理器上执行效率和编译速度。

Abstract: Coarse -grained reconfigurable processors are increasingly popular by their optimized architecture and outstanding performance,which make it possible for applications to run with high parallel efficiency.In this paper we introduce for the first time a template -based compiling system on REmus II coarse -grained reconfigurable processor.Starting from the compute -bound code parts,we create a template -lib system for typical multimedia decoder algorithms, which involves the template structure and developing flow design. Then we complete the compiling system with an annotation technique on programming codes.The results show that the compiling system effectively increases the compiled codes'operating efficiency as well as the compiling speed.