摘要
随着问题规模的增大和对实时性要求的提高,SIMD向量处理器尤其是带有向量运算单元的处理器在业界得到广泛应用。处理器上程序的运行状态一般由编译器通过堆栈进行管理。已有编译器堆栈设计机制在SIMD体系结构中严重影响了整个应用程序的运行性能。根据SIMD体系结构特点,提出了一种高效分布式堆栈设计方法——HEDSSA。实验结果表明,HEDSSA堆栈使得应用程序在进行局部数据访问、函数调用、发生中断以及动态分配数据时能够以更高的效率访问堆栈数据。
As the complexity and real-time requirement of applications become increasingly deman-ding, SIMD vector processors with vector process element are widely used in various domains. The status of the programs running on processors is usually managed by compiler stack. The existing continuous compiler stack technique is not suitable for SIMD-vector processors. We propose a high efficient dis-tributed stack for SIMD architecture called HEDSSA, which is more suitable for SIMD-vector architecture. Experimental results show that the HEDSSA can achieve higher performance in accessing data,calling functions, handling interrupt and dynamically allocating data.
出处
《计算机工程与科学》
CSCD
北大核心
2017年第11期1986-1990,共5页
Computer Engineering & Science
