With The Adaptability Of Program Analysis Techniques

Compiler is very critical to exploit the parallelism for all kinds of parallel processing systems. Program analysis is the most important way to find the parallelism among all phases of a compiler. But the increasing requirement on the aspects of accuracy and functionality causes that the program analysis phase become more and more complicated to be developed, ported and maintained. So, the research on the flexibility of the program analysis has become a hotspot on the compilation field. After the deeply study of the algorithm and the way to support different parallel architectures of the program analysis, this paper presents an integral approach to improve the flexibility of the program analysis phase. By using many useful techniques, such as multi-view programming paradigm, automatic generation of the interprocedural analyzer and optimizer and incremental analysis, the approach can acquire fairly good effect.The main contributions of this paper are listed below. Present an interprocedural model, which can describe all data flow problem. In the model, each interprocedural problem is the extension of its corresponding intraprocedural problem, which can avoid solving the interprocedural problems and the intraprocedural problems completely separately. By using this model, we can solve an interprocedural problem by applying an algorithm consisting of two cooperative parts— the interprocedural part and the intraprocedural part and can get context-sensitive analysis result. Present and implement an automatic generator of analyzers and optimizers — IGEN. In the design, we provide a proper description method to users and select the compatible frameworks for the interprocedural part and the intraprocedural part. IGEN can generate interprocedural analyzers and optimizers and the features of its products are: they are independent of intermediate representation; they can maintain the analysis information incrementally; they provide several selections to users to select algorithms. Generally, IGEN can not only save a lot of developing work of programmers, but also improve the flexibility of the program analysis phase greatly. Present an approach of using multi-view programming paradigm to construct the program analysis phase, which mainly solve such problems as the decision of the size of views, the division of views and the abstractive level of algorithms.