全局指令调度综述  

A Survey on Global Instruction Scheduling

作　　者：杨书鑫[1] 张兆庆[1] 

机构地区：[1]中科院计算技术研究所先进编译技术研究组,北京100080

出　　处：《计算机工程与应用》2004年第21期44-48,89,共6页Computer Engineering and Applications

基　　金：国家自然科学基金项目(编号:69933020);国家863高技术研究发展计划项目(编号:2001AA11061)资助

摘　　要：指令调度通过调整指令之间的顺序来提高指令级并行度(ILP)。然而基本块通常很小,因而潜在的ILP也很小。随着芯片设计技术的发展,现代的处理机所包含的资源却越来越丰富。指令调度只有跨越基本块的边界(即全局指令调度)才能够充分发挥处理机潜在的和程序中固有的ILP。全局指令调度可划分为有环和无环两种。该文介绍了无环全局指令调度的几种影响力较大的算法。同时还简单介绍了有关全局指令调度的新的热点。Instruction scheduling is used to exploit the instruction level parallelism(ILP)inherent in program through reordering its instructions.Since a basic block is usually very small,the ILP in a block is hence very small.However,modern processors feature rich resources.For the sake of better utilization of those ample resources,scheduler should move instructions across block boundaries.Global instruction scheduling techniques can be roughly divided into two categories:cyclic scheduling and acyclic scheduling.In this paper,we survey some typical acyclic global scheduling algorithms.Some related techniques are also summarized and some new trends in instruction scheduling are reported.

关 键 词：全局指令调度 指令级并行 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]