Advanced Techniques in Interprocedural Analysis

这项工作是过程间数据流工作的继续 在编程环境中进行分析和优化。 新的工作 将围绕三个主要方向展开。 1. 用于过程间分析的算法， 信息比标准技术。 这包括 副作用问题的公式，决定了 最大的一个涉及数组的子部分，改进， 不断传播以保留丢弃的有价值信息 用目前的方法，和近似方法的流动- 敏感集 2. 将程序返回到一致的 一个或多个组成部分发生变化后状态 子部分。 在使用过程间 信息，三个不同的问题出现：纠正 过程间信息，以反映 程序，确定哪些模块必须重新编译，因为 过程间信息的变更，以及 重新生成这些模块的编译代码。 3. 使用过程间优化的全局方法 信息来指导程序的优化， 整体 在这样的方案中，编译器检查整个 计划应用程序的优化，如链接 裁剪或活动记录合并，并确定 优化的水平是值得为个人进入 点 本研究的背景将是Rn的编译系统 FORTRAN编程环境。 环境设计为 一个研究全程序分析和优化的试验平台。 目前，它正在扩展到执行增量依赖 支持并行编程研究的分析。

This work is a continuation of work on interprocedural data-flow analysis and optimization in a programming environment. The new work will emphaize three main directions. 1. Algorithms for interprocedural analysis that produce sharper information than standard techniques. This includes formulations of the side effect problems that determine the largest subsection of an involved array, improvements to constant propagation to retain valuable information discarded by current methods, and approximation methods for flow- sensitive sets. 2. Incremental methods for returning a program to a consistent state after a change to one or more of its constituent subparts. In a compiler that uses interprocedural information, three distinct problems arise: correcting the interprocedural information to reflect the new state of the program, determining which modules must be recompiled because of changes to the interprocedural information, and regenerating compiled code for those modules. 3. A global approach to optimization that uses interprocedural information to direct the optimization of the program as a whole. In such a scheme, the compiler examines the entire program to plan application of optimizations like linkage tailoring or active record merging, and determines what levels of optimization are merited for individual entry points. The setting for this research will be the compilation system of the Rn FORTRAN Programming Environment. The environment has been designed as a test bed for research into whole program analysis and optimization. Currently, it is being extended to perform incremental dependence analysis in support of parallel programming research.