A New Strategy for Shape Analysis Based on Coexistent Link Sets

The analysis of dynamic heap-based data structures is difficult due to the alias problem. Shape analysis tries to gather information conservatively about these structures at compile time. In the context of parallelizing compilers, information about how memory locations are arranged in the heap at runtime is essential for data dependence analysis. With proper shape information we can reveal parallelism for heap-based structures, which are typically ignored by compilers. Existing shape analysis techniques face a dilemma: either they are too costly to be useful for real compilers or they are too imprecise to be useful for real programs. In this work, we present a new strategy for shape analysis based on a compact representation for the shape of data structures. This is done by using Coexistent Links Sets for nodes in a graph. The technique is simple to implement and very precise at the core level. Further precision-vs-cost balance can be tuned with the use of extensible properties.