Complexity Estimation in the PIPS Parallel Programming Environment

In order to choose the best optimized version of a real Fortran program, we have to compare the execution times of different optimized versions of the same Fortran program. Since real scientific Fortran programs can run for hours on expensive machines, it is useful to perform a static analysis that predicts the running-time. In this paper, we introduce our complexity model, which is a part of the PIPS project. This model is composed of a library of polynomial models of program performance and dynamically-derived program statistics to estimate the running time of Fortran programs. Then, complexity results of a real example from the PERFECT Club are presented.