一种面向分布式寄存器文件的VLIW调度新策略

A Novel VLIW Schedule Strategy for Distributed Register Files

新一代面向密集计算的高性能处理器普遍采用分布式寄存器文件来支撑ALU阵列,并通过VLIW开发指令级并行。面向分布式寄存器文件的编译成为新兴的研究热点,在斯坦福大学的kernelC编译器ISCD中最早提出了面向分布式寄存器的VLIW调度问题,在该领域处于领先水平,但是没有解决重负载下的分布式寄存器分配问题,使应用编程受到极大限制。在其基础上提出了一种新的VLIW调度策略,实验结果表明能够很好地解决重负载下的分布式寄存器分配问题。

Newly-emerging high performance processors for intensive computing generally use distributed register files to support ALU array and to explore instruction level parallelism （ILP） by VLIW. Research of the VLIW compiler for distributed register file has been a hotspot. KernelC compiler ISCD developed by Stanford University was the first of its kind to deal with VLIW schedule for distributed register file and the only compiler actually applied as well. However, it limited the program for its weakness of scheduling the code with heavy workload. Based on ISCD, this paper presents a new VLIW schedule strategy. Experimental results show that it can deal with distributed register file allocation with heavy workload very well.