High Performance Implementation of a Parallel Object-Oriented Language using Type Inference

大规模并行计算机的出现需要新的编程范例。作为一种并行面向对象语言，它允许用户灵活地描述各种风格的并行操作，被公认为是描述高度并行和分布式程序的最有前途的解决方案之一。本文针对并行面向对象语言A-NETL提出了包括类型推理在内的编译时和运行时优化技术，并在工作站/个人计算机集群、商用并行机、富士通AP1000以及面向A-NETL的并行机A-NETL多机上进行了实现。实验结果表明，利用所提出的优化技术，高级语言工具得到了高效的实现。

The advent of massively parallel computers requires new programming paradigms. As a parallel object-oriented language allows the user to describe various styles of parallel operations flexibly, it is recognized as one of the most promising solutions for describing highly parallel and distributed programs. However, the high-level language structure requires new techniques for the high-performance implementation.This research proposes compile-time and run-time optimization techniques, including type inference, for a parallel object-oriented language, called A-NETL.They have been implemented on various platforms, such as clusters of workstations/personal computers, a commercially available parallel machine, Fujitsu AP1000, and the A-NETL oriented parallel machine, called the A-NET multicomputer.Experimental results show that the high-level language facilities have been implemented efficiently, utilizing the proposed optimizattion techniques.