Studies in High-Performance, Fine Grain Parallel Architectures

高性能、细粒度并行架构的研究

• 批准号: 8919635
• 负责人: Gurindar Sohi
• 金额: $ 25.33万
• 依托单位: University of Wisconsin-Madison
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1990
• 资助国家: 美国
• 起止时间: 1990-04-01 至 1993-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8919635&HistoricalAwards=false
• 关键词: Studies; Performance; Fine; Grain; Parallel

To design a high-performance, fine-grain parallel architecture, three main problems must be addressed. They are: (i) the provision of an appropriate operation issuing mechanism, (ii) the provision of an adequate resource architecture and (iii) the provision of an adequate memory architecture. The solutions to each of these problems is heavily influenced by the nature of parallelism that exists in application programs. In this research, these problems are investigated, and solutions are developed and evaluated. The major difference between the effort and previous efforts will be the focus on exposing, rather than hiding, data dependencies.

为了设计高性能、细粒度并行架构，必须解决三个主要问题。 它们是：（i）提供适当的操作发布机制，（ii）提供足够的资源架构，以及（iii）提供足够的内存架构。 这些问题的解决方案很大程度上受到应用程序中存在的并行性本质的影响。 在本研究中，将对这些问题进行调查，并制定和评估解决方案。 这项工作与之前的工作之间的主要区别在于重点在于公开而不是隐藏数据依赖性。