A Transformation--Based Approach for the Design of Parallel/Distributed Scientific Software: the FFT

摘要: We describe a methodology for designing efficient parallel and distributed scientific software. This utilizes sequences of mechanizable algebra--based optimizing transformations. In this study, we apply our to the FFT, starting from high--level algebraic algorithm description. Abstract multiprocessor plans are developed refined specify which computations be done by each processor. Templates then created that locations data on processors, as well flow among processors. in both MPI OpenMP programming styles. Preliminary experiments comparing code constructed using with several standard libraries show is often competitive sometimes performs better. Interestingly, handled larger range problem sizes one target architecture.