异构粗粒度可重构处理器的自动任务编译器框架设计  

A Task Compilation Framework Design for a Heterogeneous Course-Grained Reconfigurable Processor

作　　者：楼杰超 绳伟光[1] 何卫锋[1] 毛志刚[1] 

机构地区：[1]上海交通大学电子信息学院微纳电子学系,上海200240

出　　处：《微电子学与计算机》2015年第8期110-114,共5页Microelectronics & Computer

基　　金：国家"八六三"计划(2012AA012702)

摘　　要：面向异构粗粒度可重构处理器——GReP,提出了一种对程序员友好的ANSI C扩展语言——GR-C,并设计了全自动的任务编译器框架.任务编译器前端以GR-C语言作为输入,经过解析器解析后生成具有静态单赋值特性的中间表示.后端以中间表示作为输入,经过软硬件划分、代码优化、任务调度和映射等步骤生成可执行文件与配置信息.针对IDCT、FIR等几种典型计算密集型算法的实验表明,任务编译器较好地发掘了异构粗粒度可重构处理器的并行可重构计算能力,取得了较好的加速效果,从而证明了GR-C语言及自动任务编译器框架的合理性与可行性.For the heterogeneous course-grained reconfigurable processor-GReP,we proposes an extension language called GR-C,which is based on ANSI C and friendly to programmer,and a full automatic task compilation framework is designed.The task compiler frontend takes the GR-C as input and the parser transforms the GR-C into an intermediate representation（IR）with static single assignment form.The task compiler backend takes the IR as input and generates executable files and configuration information after the steps of software-hardware partition,code optimization,task scheduling and mapping and so on.The experimental results of typical computing-intensive algorithms like IDCT and FIR show that our task compiler efficiently exploits the parallel-computing ability of the heterogeneous course-grained reconfigurable processor,which proves the rationality and feasibility of our GR-C language and automatic compilation framework.

关 键 词：粗粒度 可重构处理器 并行性 编程语言 编译器 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]