摘要
SIMD计算机不能直接执行MIMD语言,但通过SIMD解释程序器能够执行,这个解释程序器克服了只允许单指令流的SIMD限制,将MIMD的进程当做解释程序器的数据(SIMD中的MD),而解释程序器本身是一个单指令流.应用于MIMD程序的SIMD解释程序器运行,速度取决于解释程序器中Loop的组织,并且组织的优化又取决于MIMD程序的执行.本文通过引入Markov链模型,给出一个自适应算法,对在解释程序器Loop中的原始指令进行动态优比排序,即重新组织,使代码执行效率提高.
SIMD Computer can't directly execute an MIMD Language,but can by an SIMD interpreter. The interpreter circumvents the SIMD restrication of only allowing a single instuction stream, and takes the MIMD processes as its data (the MD of SiMD),while the interpreter itself is a single instruction stream. The speed of an SIMD interpreter for MIMD program depends on the organization of interpreter loop,and the optimal organization depends on the executed MIMD program. In this paper, we introduce a Markov Chain model and present an adaptive algorithm for dynamically optimizing the ordering of primitive operations of interpreter loop,that is.organization again. This will make code execute very efficiently.
出处
《江西师范大学学报（自然科学版）》
CAS
1993年第2期133-136,165,共5页
Journal of Jiangxi Normal University(Natural Science Edition)
关键词
单指令多数据
计算机
多指令多数据
single instruction multiple data
multiple instruction multiple data
intepreter
Markov chain model
