数据并行高层描述语言中的数据抽象

数据并行高层描述语言DPHL是我们开发的面向科学计算的一个描述语言，变量定义和基于域上的计算是该语言的重要组成部分，本文通过对数据并行方法、科学计算特点以及现有并行语言的研究和分析，论述了变量定义中数据抽象的两个主要方面；一是引入了变量的定义域、子域和依赖域概念，给出了它们的形式化描述，二是引入了变量的数学属性，并和域分离，便于数据并行自动识别，重构高效的并行程序，使DPHL描述符合或接近于专业上数学表示。

DATA ABSTRACTIONS FOR A DATA PARALLEL HIGH-LEVEL MODELLING LA NGUAGE

Data Parallel High level Modeling Language(DPHL), which started in 1997 at JiangNan Institute of Computing Technology, is developed for scientific computing applications. Variable definition and computation over domain are main contents of DPHL designing. Through the study and analysis for data parallel methods?scientific computation features and existing parallel programming languages, this paper deals with two important aspects in the dataAbstraction: one is the introduction of concept definition Domain?SubDomain and Dependence Domain and their formal modeling, the other is the introduction of mathematics attributes and the separation between mathematics attributes and domain. Such data abstraction can be used for the auto detection of data parallelism and the reconfiguration of parallel programs. It may help the users in concentrating applications' DPHL modeling consistent with concept modeling within research projects.