I will present two techniques in static analysis that help improve the analysis cost-accuracy balance: sparse analysis and modular analysis. Sparse analysis exploits two types of sparsities in program semantics: spatial sparsity and temporal sparsity. Spatial sparsity is that a program fragment accesses only a sparse part among the whole memory. Temporal sparsity is that data flow occurs only along sparse data-dependent paths among the whole execution paths. I will discuss a sparse analysis framework that guides how to transform analysis designs into their sparse versions without degrading the accuracy of the original analyses. Second part is on modular analysis. Modular static analysis analyzes open program fragments in advance and completes the whole analysis later when the fragments are closed. Unlike global whole-program analysis, modular analysis allows for the separate and parallel analysis of modules. It supports incremental analysis and improves the analysis precision by making a module’s analysis results sensitive to its various use contexts. I will present our recent proposal for a framework to facilitate the design of modular static analyses.

Professor