Compile-Time Analysis Tools for Parallel Software Development Environments

点击翻译按钮获取中文摘要

Several compile-time analyses aimed at aiding compiling and debugging of programs in parallel software development environments will be developed. The emphasis is on using our parallel data flow analysis and approximate pointer-induced aliasing algorithms to facilitate better transformation of programs for parallel execution, and further developing the work in compile-time anomaly detection for explicitly parallel programs. The former goal includes both parallelization of sequential codes and optimization of already parallel codes. The latter is aimed at the problems of deadlock and race detection. Specific plans are: > to further profile and optimize the parallel hybrid data flow analysis algorithm, especially on interprocedural problems; > to extend the approach in the approximate pointer alias algorithm to accommodate conflict analysis for parallelization of C-like languages; > to explore directed tracing for race detection and implement the approach in a prototype; and > to profile and extend the static deadlock detection algorithm.