基于寄存器压力差异化的VLIWDSP编译器超块调度算法  

A Novel Hyperblock Scheduling Algorithm for VLIW DSP Based on Register-pressure

作　　者：鲍丽丹[1,2] 张铁军[1] 王东辉[1] 

机构地区：[1]中国科学院声学研究所,北京100190 [2]中国科学院大学,北京100039

出　　处：《微电子学与计算机》2015年第9期18-22,共5页Microelectronics & Computer

摘　　要：为了有效开发VLIW DSP处理器的指令级并行性,提出一种基于寄存器压力差异化的超块调度算法.该算法在传统列表调度的基础上扩展调度区间,同时以启发式的优先级函数综合考虑不同代码段的执行频率特性,设置差异化的寄存器压力敏感度.实验显示该调度算法在有限的寄存器资源下合理开发指令并行度,达到提高程序性能的目的.Instruction scheduling and reordering are always used to boost instruction parallelism,especially for VLIW DSP processors whose performance is largely exploited by static compilation because of its simple hardware structure.This paper proposes a hyperblock instruction scheduling algorithm considering with register pressure differentiation of intensive and non-intensive code.This algorithm builds the topological order to schedule each node of the data-dependence graph based on priority models.The experimental result shows that the proposed algorithm is effective in exploiting instruction parallelism and improving performance.

关 键 词：数字信号处理器 超长指令字 编译器 指令调度 寄存器 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]