SPAR: Synergistic Program Analysis and Restructuring

本项目研究将雄心勃勃的程序分析和程序重组技术引入自动程序优化系统，目标是显著提高现代计算机体系结构编译器生成的代码质量。尽管存在各种各样的编译器优化技术，但它们并没有在生产编译器中充分使用，因为分析和转换是计算密集型的。此外，通过分析获得的信息必须在不同的编译器传递和程序转换中进行维护和更新。该项目依赖于先前关于“稀疏”数据流框架的工作，并探索了三种方法来克服目前在生产编译器中使用大规模程序分析的限制：(1)将密集数据流框架分解为一组相互关联的稀疏数据流框架，(2)选择性地计算和使用代表特定程序优化的昂贵信息。(3)将程序分析与转换相结合。

This project investigates the introduction of ambitious program analysis and program restructuring techniques into automatic program optimization systems, with the goal of significantly improving the quality of code produced by compilers for modern computer architectures. Although a variety of compiler optimization techniques exist, they are not used to their full extent in production compilers because, the analysis and transformations are computationally intensive. In addition, the information obtained by analyses must be maintained and updated across the various compiler passes and program transformations. The project relies on the previous work on ``sparse'' data flow frameworks and explores three methods for overcoming the current limitations to using ambitious program analyses in production compilers: (1) decomposing a dense data flow framework into an interconnected set of sparse data flow frameworks, (2) selectively computing and using costly information on behalf of a particular program optimization. and (3) integrating program analysis and transformation.