Re-engineering Fortran programs for parallel execution

The author describes features of a parallel scientific software re-engineering and development environment. The ultimate goal is the creation of an integrated scientific programming environment that can assist both in reverse engineering of existing applications for parallel execution and in forward engineering of new applications. The restructured form allows much easier portability across a wide range of parallel processor architectures. The author has chosen Fortran as the first target language, both because of the large number of existing programs that could benefit from restructuring, and because the relatively simple, straightforward language model makes it possible to focus more attention on basic restructuring issues. The approach puts the application programmer (scientist) in the loop to help make high-level decisions regarding code restructuring and parallelization strategies. The resulting code at any stage is better ´structured´ and can thus serve as a good basis for further (automatic or manual) optimizations. The proposed approach has been used to generate a large-grain parallel (LGDF2) version of a 5300 line vectorized Monte Carlo code (EGS4V).<>