The Construction of an Universal Linearized Control Flow Graph for Static Code Analysis of Algorithms

Abstract

This paper presents the description of a possible way to build the universal linearized control flow graph which is supposed to be architecture-independent and applicable to the description of any high level language programs. The practical usefulness of the graph considered is the existence of the fast and optimal search of the unique execution paths that is valuable in the methods of static code analysis of algorithms for race condition search. Optimizing compiler CLANG&LLVM is used as a technical tool for building a linearized control flow graph. The analysis of LLVM compiler procedural optimizations is carried out in the article. Transformations of intermediate representation of those optimizations result in reduction of the number of instructions responsible for conditional or unconditional branches in the code as well as the elimination or simplification of the whole loops and conditional constructions. The results of the analysis performed in the paper allowed revealing the most effective optimizations line of the LLVM compiler, which leads to a significant linearization of the control flow graph. That fact was demonstrated by the example code of the Peterson mutual execution algorithm for 2 threads.

Keywords

• состояние гонки
• статический анализ
• многопоточные алгоритмы
• SSA
• оптимизирующий компилятор