首页 | 本学科首页   官方微博 | 高级检索  
     

VLIW上的软件旁路与细粒度并行调度
引用本文:朱凯佳,王雷,尹宝林. VLIW上的软件旁路与细粒度并行调度[J]. 北京航空航天大学学报, 2003, 29(10): 914-918
作者姓名:朱凯佳  王雷  尹宝林
作者单位:北京航空航天大学 计算机学院, 北京 100083
基金项目:北京市自然科学基金;4023012;
摘    要:描述并分析了一种新型的VLIW(Very Long Instruction Word)机器模型,为充分利用该模型的并行特性,提出了基于静态单一赋值(SSA)表示的软件旁路方法,以及集成的基于操作的指令调度与寄存器分配算法.与常规的软件旁路方法和基于周期的调度方法相比,该方法可以灵活地进行结果寄存器的分配,并有效减少由重用结果寄存器导致的"危险值"溢出次数,生成高质量的目标代码. 

关 键 词:指令调度   寄存器分配   超长指令字   软件旁路
文章编号:1001-5965(2003)10-0914-05
收稿时间:2003-06-05
修稿时间:2003-06-05

Software bypass and fine-grained parallel scheduling on VLIW
Zhu Kaijia,Wang Lei,Yin Baolin. Software bypass and fine-grained parallel scheduling on VLIW[J]. Journal of Beijing University of Aeronautics and Astronautics, 2003, 29(10): 914-918
Authors:Zhu Kaijia  Wang Lei  Yin Baolin
Affiliation:School of Computer Science and Technology, Beijing University of Aeronautics and Astronautics, Beijing 100083, China
Abstract:A new very long instruction word(VLIW) machine model was described and analyzed.In order to exploit the parallelism in this model, a software bypass method based on static single assignment and an integrated operation-based instruction scheduling and register allocation algorithm were presented. Compared with traditional software bypass and cycle based scheduling algorithm, the new methods can efficiently allocate result registers, schedule micro operations, reduce "hazard value" spills caused by result register reuse, thus yield high quality target code.
Keywords:instruction schedule  register allocation  very long instruction word  software bypass
本文献已被 CNKI 维普 万方数据 等数据库收录!
点击此处可从《北京航空航天大学学报》浏览原始摘要信息
点击此处可从《北京航空航天大学学报》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号