Cluster环境下p-HPF编译器支持的并行计算范式  

MULTI-PARADIGM PARALLEL COMPUTING OF p HPF COMPILER ON CLUSTER ENVIRONMENT

作　　者：胡长军[1] 余华山[2] 丁文魁[2] 许卓群[2] 

机构地区：[1]石油大学计算机科学系,东营257062 [2]北京大学计算机科学与技术系,北京100871

出　　处：《计算机研究与发展》2001年第8期954-959,共6页Journal of Computer Research and Development

基　　金：国家"八六三"高技术研究发展计划基金资助 (86 3 -3 0 6 -ZT0 1-0 2 -3 )

摘　　要：p- HPF是研制的一个符合 HPF(high performance Fortran)规范的并行编译系统 ,以 HPF为核心实现多范式并行计算是开发大型并行应用系统的基础 .首先论述了 Cluster环境下的并行运行范式 ,包括 farm parallel范式、流水线并行、流循环并行、基本数据并行和组合数据并行等 ,抽象分析了它们的性能 ,接着给出了利用 p- HPF的外部过程机制、任务并行机制以及 FORAL L,INDEPENDENT DO等典型并行语句实现几种典型并行范式的方法 ,给出了实例程序 。p HPF is a parallel compiler, which accords with HPF (high performance Fortran) specification. Implementing multi paradigm parallelism based on HPF is the foundation for developing large scale parallel application systems. First detailed in this paper are the parallel programming and executing paradigms in cluster environment, and their performance is analyzed abstractly, including farm parallelism, pipeline parallelism, stream iterative parallelism, basic data parallelism and composite data parallelism. Then their implementations are discussed using task parallel, extrinsic procedure calling, and typical parallel clauses such as FORALL, INDEPENDENT DO in p HPF. Some program examples are also provided and their execution results are analyzed.

关 键 词：并行计算模型 HPF语言 程序设计语言 编译器 编译程序 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]