Efficient Implementation of Concurrent Object-Oriented Languages for General Purpose MIMD Parallel Computers.

我们的研究目标是在通用的MIMD并行机上为并发面向对象语言开发一个高效的语言处理器(即编译器和运行时系统)，以证明并发面向对象范例在实际环境中的可行性。经过三年的研究，我们的目标已经基本实现，并且在设计一种新的高效的多线程并行程序调试方案方面取得了很好的结果。具体研究成果如下：(1)设计了一种新的并发面向对象语言ABCL/f。在这种语言的正常编程风格中，一个可变的数据结构被表示为一个并发对象，其中只允许通过它的相关方法访问它[参考文献7]。(2)通过简化ABCL/f，我们新设计了一种称为Schematic的语言。这种语言可以看作是一种并发的、面向对象的方案语言扩展，该方案语言是一种非常流行的Lisp方言。[参考文献14](3)我们设计了一种名为StackThread的抽象机，并开发了高效的StackThree实现技术。[参考文献1，2，8](4)基于StackThread的实现，我们为ABCL/f实现了一个语言处理器，获得了良好的性能指标。(5)设计并实现了一种新的垃圾收集方案，并对其性能进行了测试。该方案被并入我们在(4)中提到的语言处理器中。[10，11](6)我们设计了一种新的调试方案，支持重放和跟踪，即使在程序中有大量线程活动的情况下，也需要少量的日志信息。这一方案已经实施。[参考文献15](7)在ABCL/f中，我们开发了(A)预测RNA二级结构和(B)有限元方法应用和N体问题应用的程序[9，12，13]。

The goal of our research is to develop a highly efficient languages processor (i.e., compiler and runtime system) for concurrent object-oriented languages on general purpose MIMD parallel machines, demonstrating the viability of the concurrent object-oriented paradigm in the practical setting. After three-year research effort, our goal has been basically achieved and furthermore, much good results have been obtained in designing a new efficient debugging scheme for multi-thread parallel programs. The concrete research achievements are in order.(1) A new concurrent object-oriented language called ABCL/f was designed. In the normal style of programming in this language, a mutable data structure is represented as a concurrent object where its access is only allowed through its associated methods which are invoked mutually exclusively [Ref 7].(2) By simplifying ABCL/f, we newly designed a language called Schematic. This language can be viewed as an concurrent object-oriented extension of the Scheme languages which is a very popular dialect of Lisp. [Ref14](3) We designed an abstract machine called StackThreads amd highly efficient techniques for implementing StackThreads were developed. [Ref 1,2,8](4) Based on the implementation of StackThread, we implemented a language processor for ABCL/f where good performance numbers were obtained.(5) A new garbage collection scheme was disigned and implemented and its performance was measured. This scheme was incorporated in our language processor mentioned in (4). [10,11](6) We designed a new debugging scheme which supports replay and tracing, requiring a small amount of logging information even where a large number of threads are active in a program. This scheme has been implemented. [Ref 15](7) In ABCL/f, we developed programs for (a) predicting RNA secondary structures and (b) finite element method application and N-body problem applications [9,12,13].