粗粒度可重构阵列分支指令的优化设计与实现  

Design and Implementation of the Optimization of Branch Instructions for CGRA

作　　者：汪翔 WANG Xiang(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《现代计算机》2021年第11期17-23,共7页Modern Computer

摘　　要：为了优化粗粒度可重构阵列条件分支执行的性能与功耗,基于自定义的发散汇聚操作设计一种优化的分支指令执行方式,利用与数据位绑定的分支位决定是否执行有效运算和访存,降低粗粒度可重构阵列分支指令执行的功耗,提升分支指令执行的性能。实验结果表明,与传统的针对粗粒度可重构阵列的分支实现方法部分断言技术相比,该方法能够获得1.31倍的性能优化、21%的功耗降低以及12%的资源降低。In order to optimize the performance and power consumption of the conditional branch execution of the Coarse-grained Reconfigurable Ar⁃ray,an optimized branch instruction execution method is designed based on the custom divergence and convergence operation,and the branch bit bound to the data bit is used to determine whether to perform effective operations and Memory access,this method decreases power consumption of branch instruction execution in coarse-grained reconfigurable arrays,and improves the performance of branch in⁃struction execution.Experimental results show that our method achieves up to 1.31x performance gain compared to conventional partial predication techniques of branch implementation methods for CGRA while consuming 21%less energy and 12%less resources.

关 键 词：粗粒度可重构阵列 分支语句 嵌套分支 部分断言 

分 类 号：TP332[自动化与计算机技术—计算机系统结构]