The Automatic Transformation of Linked List Data Structures

Irregular access patterns are a major problem for today's optimizing compilers. It prevents the application of many optimizations as data dependence analysis cannot be performed on code containing such patterns. We present a novel approach that transforms linked list pointer access patterns in regularized array access patterns. Our method separates the computations from the linked list traversal which results in an intermediate code which is suitable for recompilation at runtime and as such can take both computation code and the data the code operates on into account enabling generation of code and restructured data structures. This is radically different from other approaches, which focus mainly on improving cache performance by placing list elements, which are logically close to each other physically close to each other.