Compiling dataflow analysis of logic programs

摘要

Abstract interpretation is a technique extensively used for global dataflow analyses of logic programs. Existing implementations of abstract interpretation are slow due to interpretive or transforming overhead and the inefficiency in manipulation of global information. Since abstract interpretation mimics standard interpretation, it is a promising alternative to compile abstract interpretation into the framework of the WAM (Warren Abstract Machine) for better performance. In this paper, we show how this approach can be effectively implemented in a low-cost manner. To evaluate the possible benefits of this approach, a prototype dataflow analyzer has been implemented for inference of mode, type and variable aliasing information of logic programs. For a subset of benchmark programs in [15], it significantly improves the performance by a factor of over 150 on the average.