ViC*: A Compiler and Runtime System for Out-of-Core C*

ViC*：核外 C* 的编译器和运行时系统

• 批准号: 9625894
• 负责人: Thomas Cormen
• 金额: $ 10.92万
• 依托单位: Dartmouth College
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1996
• 资助国家: 美国
• 起止时间: 1996-08-01 至 1999-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9625894&HistoricalAwards=false
• 关键词: ViC; Compiler; Runtime; System; Out

The proposed project completes the implementation, already underway, of the ViC* compiler and run-time system. The compiler translates the data-parallel language C* with some data explicitly declared to be "out of core." Out-of-core data reside on a parallel disk system. The compiler produces C for a distributed-memory multiprocessor. It adds I/O and library calls to access data on disk, and it also produces virtual processing loops to allow each physical processor to be shared by multiple parallel-variable elements. A ViC* program will run significantly faster than one run with traditional sequential virtual memory on the individual nodes of a MIMD machine. Unlike demand paging, ViC* will take advantage of aggregate data-parallel operations by optimizing away many I/O calls. Moreover, the ViC* run-time library will execute optimal parallel-disk algorithms that have appeared in the literature recently. There are two major components to ViC*: a compiler and a run time library. Both are designed to minimize the number of parallel disk accesses made by the executing program. The runtime library will implement the entire C* library for both incore and out-of-core parallel variables. It will run on any platform that supports PVM and any of the following filesystem interfaces: Scotch, Galley, Scalable I/O Initiative API, and UFS for multiple disks. ***

拟议的项目完成了已经在进行中的ViC*编译器和运行时系统的实施。编译器在翻译数据并行语言C*时，将一些数据显式声明为“外核”。核外数据驻留在并行磁盘系统上。编译器为分布式内存多处理器生成C语言。它添加I/O和库调用来访问磁盘上的数据，它还生成虚拟处理循环，以允许多个并行变量元素共享每个物理处理器。ViC*程序的运行速度明显快于在MIMD机器的单个节点上使用传统顺序虚拟内存运行的程序。与需求分页不同，ViC*将通过优化许多I/O调用来利用聚合数据并行操作。此外，ViC*运行时库将执行最近在文献中出现的最优并行磁盘算法。ViC*有两个主要组件：编译器和运行时库。两者都是为了最小化执行程序所进行的并行磁盘访问的数量。运行时库将为内核和内核外并行变量实现整个C*库。它可以在任何支持PVM和以下文件系统接口的平台上运行：Scotch、Galley、Scalable I/O Initiative API和用于多个磁盘的UFS。＊＊＊