Parallelising Compiler for a Declarative Language

声明性语言的并行编译器

• 批准号: 07458062
• 负责人: AMAMIYA Makoto
• 金额: $ 4.86万
• 依托单位: KYUSYU UNIVERSITY
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (B)
• 财政年份: 1995
• 资助国家: 日本
• 起止时间: 1995 至 1997
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-07458062/
• 关键词: Declarative Parallel Language; Prallel Processing; Fine-grain Multi-threading; Parallelizing Compiler; Dataflow; 超並列処理; 並列オプジェクト指向; 超並列マシン; メッセージフロー; 並列オブジェクト指向

We have implemented a dataflow declarative language on stock (non-dataflow) machines.・Practical CompilerThe compiler has two phases : machine-independent phase and machine independent phase. The former generates a fine-grain virtual machine code, called DVMC (Datarol Virtual Machine Code). The latter optimizes the DVMC and translates it to the suitable code for a particular target machine. Since our source language supports non-strictness, frequent dynamic scheduling, which may incur heavy overhead on stock machines, will occur during the program execution. However, our compiler can find an appropriate execution order of a program, so that the frequency of useless dynamic scheduling may be reduced. In addition, our runtime system provides an efficient execution environment even if a program has potential points of dynamic scheduling.・Experiment Environment for Fine-grain parallelismIn our implementation, we have built a virtual machine which supports fine-grain parallel execution. We have also developed an interpreter of this fine-grain virtual machine. The interpreter provides an interactive execution environment of fine-grain multithread programs. Since the interpreter is augmented with GUI,we can visually check the fine-grain execution.

我们已经在库存（非数据流）机器上实现了数据流声明性语言。实用编译器编译器有两个阶段：机器独立阶段和机器独立阶段。前者生成细粒度的虚拟机代码，称为DVMC （Datarol virtual machine code）。后者优化DVMC并将其转换为适合特定目标机器的代码。由于我们的源语言支持非严格性，频繁的动态调度将在程序执行期间发生，这可能会给库存机器带来沉重的开销。然而，我们的编译器可以找到一个合适的程序执行顺序，这样可以减少无用的动态调度的频率。此外，即使程序有潜在的动态调度点，我们的运行时系统也提供了一个高效的执行环境。·细粒度并行的实验环境在我们的实现中，我们已经构建了一个支持细粒度并行执行的虚拟机。我们还开发了这个细粒度虚拟机的解释器。解释器提供了细粒度多线程程序的交互式执行环境。由于解释器增强了GUI，我们可以直观地检查细粒度执行。

项目成果

Hiroshi Tomiyasu, et al.: "Co-Processor System Design for Fine-grain Message Handling in KUMP/D" Proc.of 3rd European Conference on Parallel Processing (Euro-Par'97)(LNCS 1300). 779-788 (1997)

Hiroshi Tomiyasu 等人：“KUMP/D 中细粒度消息处理的协处理器系统设计”Proc.of 第三届欧洲并行处理会议 (Euro-Par97)(LNCS 1300)。

Shigeru Kusakabe, et al.: "Hybrid Support for Lenient Implementation of Array-Comprehension" In Proc.of Asian Computing Science Conference (LNCS 1345). 383-384 (1997)

Shigeru Kusakabe 等人：“对数组理解的宽松实现的混合支持”，亚洲计算科学会议记录 (LNCS 1345)。

Peter Kacsuk, et al.: "A Multihtreaded Implementation Concept of Prolog on Datarol-II Machie" Proc.of International Symposium on High Performance Computing (ISHPC) (LNCS 1336). 91-106 (1997)

Peter Kacsuk 等人：“Datarol-II Machie 上 Prolog 的多线程实现概念”Proc.of 国际高性能计算研讨会 (ISHPC) (LNCS 1336)。

Hiroshi Tomiyasu, et al.: "Fine-grain Message Handling Mechanism in Multi-media Oriented Parallel Process KUMP/D (in Japanese)" Transaction of Ionformation Processing Society of Japan. 38・9. 1694-1705 (1997)

Hiroshi Tomiyasu 等人：“面向多媒体的并行处理中的细粒度消息处理机制 KUMP/D（日语）”日本信息处理学会会刊 38・9（1997 年）。

Makoto Amamiya: "Datarol:A Parallel Machine Architecture for Fine Grain Multithreading" Proceedongs of lst Autrian-Hungarian Workshop on Distributed and Parallel Systems,Invited Paper. 3-18 (1996)

Makoto Amamiya：“Datarol：细粒度多线程的并行机器架构”第一届奥匈帝国分布式和并行系统研讨会论文集，特邀论文。