“SEQ　OF　PAR”结构化并行程序设计

“SEQ OF PAR” STRUCTURED PARALLEL PROGRAMMING

本文基于有关因果关系复合的ＣｏｍｍｕｎｉｃａｔｉｏｎＣｌｏｓｅｄＬａｙｅｒ（ＣＣＬ［１］）理论提出了一种结构化并行程序设计方法以支持不同并行平台的“通用”并行程序设计．该模型采用“ＳＥＱＯＦＰＡＲ”的层次化结构和基于伪同态的并行抽象数据类型来分析和刻画程序的并行行为．首先，本文通过因果关系复合的范式定理的证明在理论上论证了“ＳＥＱＯＦＰＡＲ”形式进行结构化并行程序设计的可行性，然后详细介绍了该模型中并行程序的设计和性能评价模型．良好的代数性质使得在这一模型中可采用代数方法的程序变换来进行并行程序设计、优化和验证．与其它并行程序设计模型和方法相比，该方法为自然地综合描述数据并行和控制并行提供了一种途径；特别是基于代数变换的程序变换技术使得并行程序设计既独立于具体的体系结构又便于高效实现．

This paper proposes a structured parallel programming paradigm, based on the principles of communication closed layer(CCL), to support general parallel programming across different platforms. The proof of normal form of causal order provides theoretical foundation for parallel programming in “SEQ OF PAR” style. The nice properties of CCL make it possible to design, optimize and verify parallel programs through algebraic transformation approach. Thus, the framework provides a natural integration of task and data parallelism and maintains portability and efficiency.