Integrated Parallel Processing Environment Combined Parallelizing Compiler with Runtime System for Data Distribution Feature

集成并行处理环境将并行编译器与数据分发功能的运行时系统相结合

• 批准号: 13480085
• 负责人: KUNIEDA Yoshitoshi
• 金额: $ 7.87万
• 依托单位: Wakayama University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (B)
• 财政年份: 2001
• 资助国家: 日本
• 起止时间: 2001 至 2003
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-13480085/
• 关键词: parallelizing compiler; distributed shared memory; Laputa test; dependence analysis; data distribution; Loop parallelizing; MPI; automatic parallelizing; 並列処理; PCクラスタ; COMA; 分散処理

The main purpose of this research project is to develop a total system for parallel processing program developing environment ; especially an automatic parallelizing compiler and a software distributed shared memory (SDSM) system as its runtime support system. This compiler receives Fortran77 source programs and generates corresponding parallelized C programs, as the object codes, which are executed on commodities PC/WS clusters equipped with our SDSM system as a representative of distributed memory parallel architecture. In each object code, our compiler embeds appropriate primitive functions to control the SDSM system, for example, to send/receive data and synchronize several (all) PCs, in adequate places of object C programs for their parallel execution..For the dependence analysis module of our compiler, we have newly designed and proposed a practical data dependence analysis method which is combining the established ones, i.e. GCD Test, Banerjee Test, and Separability Test, with our original method which is also combined basically with both linear programming method (Simplex method) and exhaustive integer solution search. As the results of several evaluation experiments, it is clarified that our new method can analyze exactly with almost the same ability of the famous exact test, Omega Test, in view of the analyzable patterns of complexity of the given subscript expressions. Moreover in general cases, our methods can analyze more than or equally rapidly comparing with Omega Test.

本课题的主要目的是开发一个整体系统的并行处理程序开发环境；特别是自动并行编译器和软件分布式共享内存（SDSM）系统作为其运行支持系统。该编译器接收Fortran77源程序并生成相应的并行化C程序作为目标代码，在配备我们的SDSM系统的商品PC/WS集群上执行，作为分布式内存并行架构的代表。在每个目标代码中，我们的编译器在对象C程序的适当位置嵌入适当的原语函数来控制SDSM系统，例如发送/接收数据和同步多台（所有）pc机，以便并行执行。对于我们编译器的依赖性分析模块，我们新设计并提出了一种实用的数据依赖性分析方法，该方法结合了现有的GCD测试、Banerjee测试和可分离性测试。该方法基本上结合了线性规划方法（单纯形法）和穷举整数解搜索。几个评价实验的结果表明，鉴于给定下标表达式的复杂性的可分析模式，我们的新方法可以精确地分析出与著名的精确检验——Omega检验几乎相同的能力。此外，在一般情况下，我们的方法可以比Omega Test分析更快或同样快。

项目成果

Hironobu Nobuhara, Masaaki Mineo, Tetsutaro Uehara, Shoichi Saito, Yoshitoshi Kunieda: "Design and Implementation of the Loop Restructuring Feature for the Parallelizing Compiler, MIRAI"Proc. of Int. Conf. on Parallel and Distributed Processing Techniques

Hironobu Nobuhara、Masaaki Mineo、Tetsutaro Uehara、Shoichi Saito、Yoshitoshi Kunieda：“并行编译器 MIRAI 的循环重构功能的设计与实现”Proc。

Masaaki MINEO, Satoshi YOKOTE, Tetsutaro UEHARA, Shoichi SAITO, Yoshitoshi KUNIEDA: "An Automatic Parallelizing Compiler MIRAI with Data Distribution Function and its Runtime Support System Fagus for Distributed Memory Architecture"Proc.of Int.Conf.on Par

Masaaki MINEO、Satoshi YOKOTE、Tetsutaro UEHARA、Shoichi SAITO、Yoshitoshi KUNIEDA：“具有数据分布功能的自动并行编译器 MIRAI 及其用于分布式内存架构的运行时支持系统 Fagus”Proc.of Int.Conf.on Par

Hironobu Nobuhara, Masaaki Mineo, Tetsutaro Uehara, Shoichi Saito, Yoshitoshi Kunieda: "Design and Implementation of the Loop Restructuring Feature for the Parallelizing Compile, MIRAI"Proc.of Int.Conf.on Parallel and Distributed Processing Techniques and

Hironobu Nobuhara、Masaaki Mineo、Tetsutaro Uehara、Shoichi Saito、Yoshitoshi Kunieda：“并行化编译循环重构功能的设计和实现，MIRAI”Proc.of Int.Conf.on 并行和分布式处理技术和

Masaaki MINEO, Tetsutaro UEHARA, Shoichi SAITO, Yoshitoshi KUNIEDA: "A New Practical Array Data Dependence Analysis for Parallelizing Compilers"Innovative Architecture for Future Generation High-Perfomnance Processors and Systems '2003. (印刷中).

Masaaki MINEO、Tetsutaro UEHARA、Shoichi SAITO、Yoshitoshi KUNIEDA：“用于并行化编译器的新型实用阵列数据依赖性分析”，下一代高性能处理器和系统的创新架构（正在出版）。

Asami Iwasaka, Tomomi Yamaguchi, Mariko Sasakura, Hayaru Shouno, Kazuki Joe: "Classification of Visualized Data Dependence"Proc. of Int. Conf. on Parallel and Distributed Processing Techniques and Applications. Vol.III. 1444-1450 (2002)

Asami Iwasaka、Tomomi Yamaguchi、Mariko Sasakura、Hayaru Shouno、Kazuki Joe：“可视化数据依赖性分类”Proc。