The specification of efficient data distribution schemes is one of the major tasks in programming DMMPs with state of the art parallel languages. Although there are no optimal strategies for generating such data distributions, several heuristics have been developed to provide some support to the user. Alignment analysis, for instance, is able to help find a good distribution scheme and is furthermore a useful prerequisite for automatic data distribution tools. We present ed an overview of an automatic alignment analysis tool within the framework of VFCS elsewhere, which is able to automatically generate alignment proposals for the arrays accessed in a procedure and thus simplifies the data distribution problem. In this paper we extend our previous work to interprocedural analysis taking into account dynamic realignment. This feature is essential for applying alignment analysis to real programs. We present a heuristic algorithm for interprocedural dynamic alignment analysis as well as our cost models based upon sequential profiling data. Furthermore we show that the locality concept introduced elsewhere for intradimensional alignment preferences proved to be applicable and helpful also in this context.

Top