Next Generation Software: A Staged Compilation Architecture for Program Optimization

下一代软件：用于程序优化的分阶段编译架构

• 批准号: 9975057
• 负责人: Susan Eggers
• 金额: $ 50万
• 依托单位: University of Washington
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-10-01 至 2002-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9975057&HistoricalAwards=false
• 关键词: Next; Generation; Software; Staged; Compilation

EIA-9975057Susan J. EggersUniversity of WashingtonCISE Next Generation Software: A Staged Compilation Architecture for Program OptimizationThe proposed research is developing a more flexible model, called a staged compilation model, that allows interesting analysis and optimizations to be performed at any of these three stages (and also at one or more library assembly stages between static compilation time and link time.) Each stage performs the analyses and optimizations appropriate for its (partial) knowledge about the program, and produces as output partially-compiled code and analysis summary information for use by later stages. Assignments of optimizations to later stages incur more compilation-time cost, but they present optimization opportunities not available at earlier stages.The staged compilation model also includes the notion of staging individual optimizations. A staged optimization is divided into two or more optimization stages, with the earlier optimization stages performing preplanning and set-up work, and the final optimization stage completing the optimization. Compared to a single-stage optimization, a mulit-stage optimization yields reduced compilation cost and/or increased optimization effectiveness: by adding earlier preplanning stages to a single-stage optimization, some compilation work is moved to an earlier stage where it is cheaper. Conversely, by adding later stages to a single-stage optimization, additional opportunities for optimization are exploited without repeating the entire analysis at the later stage.

下一代软件：用于程序优化的分阶段编译体系结构建议的研究正在开发一种更灵活的模型，称为分阶段编译模型，它允许在这三个阶段中的任何一个阶段（以及在静态编译时间和链接时间之间的一个或多个库组装阶段）执行有趣的分析和优化。每个阶段执行适合其关于程序的（部分）知识的分析和优化，并产生输出部分编译的代码和分析摘要信息，以供后期阶段使用。将优化分配到后期阶段会产生更多的编译时间成本，但它们提供了在早期阶段无法获得的优化机会。分阶段编译模型还包括分阶段进行单个优化的概念。阶段优化分为两个或多个优化阶段，其中较早的优化阶段进行预规划和设置工作，最后的优化阶段完成优化。与单阶段优化相比，多阶段优化降低了编译成本和/或提高了优化效率：通过将早期的预规划阶段添加到单阶段优化中，一些编译工作被转移到更便宜的早期阶段。相反，通过将后期阶段添加到单阶段优化中，可以利用额外的优化机会，而无需在后期重复整个分析。