A novel program representation for interprocedural analysis

Interprocedural optimizations and transformations are increasingly being used in industrial and academic compilers, especially the compilers for high performance systems. Interprocedural analysis involves a natural trade-off between the accuracy and efficiency of the analysis, which is typically manifested in the choice of program representation. In this paper, we present a novel full program representation, which we believe is concise enough to keep the compilation costs reasonable, but is detailed enough for useful flow-sensitive optimizations. We report on our experience in developing two novel optimization frameworks using this representation. We have implemented this framework on top of the SUIF infrastructure. We present the comparison of the size of the call graph, Myer's supergraph, and our representation on a set of benchmark programs.