Towards Dynamic Shaping

摘要

Shape theory supports techniques for static analysis of the shapes of data structures. This information is being used both to detect errors and to optimise sequential computation. The same techniques will improve the compilation of data parallel programs when all shapes can be statically determined. Even when shapes are not known until execution, research into nested data parallel languages shows that shape information may also be beneficial. This paper considers how shape analysis should be approached in the dynamic setting.