基于流体系结构的VLIW二维压缩及并行解压  

2-D Compression and Parallel Decoding of VLIW Based on Stream Architecture

作　　者：李功丽[1,2] 戴紫彬[1] 徐进辉[1] 王寿成[1] 朱玉飞 李丹[3] 

机构地区：[1]解放军信息工程大学 [2]河南师范大学计算机与信息工程学院 [3]中国人民解放军61840部队

出　　处：《电子学报》2017年第9期2256-2262,共7页Acta Electronica Sinica

基　　金：国家自然科学基金项目(No.61404175)

摘　　要：VLIW(Very Long Instruction Word)指令因为含有较多的空操作导致严重的代码体积膨胀问题,代码压缩是解决这一问题的有效措施.VLIW代码压缩需要解决三个关键问题,一是提高压缩率;二是降低解压操作对性能的影响;三是分支目标重定位.针对流体系结构上的VLIW指令特点,提出了二维压缩,对VLIW进行垂直与水平两个方向上的压缩,且水平解压可以与代码执行并行,并通过设置堆栈寄存器缓存循环入口地址.实验结果表明二维压缩有效解决了VLIW代码体积膨胀问题,可以使指令存储器的面积减少36.48%,并使得整个CISP系统面积减少了7.85%.Due to having many NOPs,VLIW（ Very Long Instruction Word） exists serious code size expansion problem. As an efficient way to solve this problem,the code compression needs to deal with three key points： improving the compression ratio（ CR）,simplifying decomposition operations,and relocating the branch target. According to the characteristics of VLIW on stream architecture,a two-dimension（ 2-D） compression scheme is put forward,where VLIW code is compressed in both vertical and horizontal directions,the horizontal decompression and code execution can be implemented in parallel,and loop entrance addresses are buffered by stack registers. The experiment results illustrate that 2-D compression scheme can resolve code expansion issue effectively. Specifically,it has achieved a 36. 48% area reduction of the on-chip instruction memory and a 7. 85% area reduction of the CISP system.

关 键 词：流处理器 VLIW 二维压缩 并行解压 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]