A Tool for the Development of Structured Parallel Applications

Skeleton/template based programming models allow performance prediction of parallel applications to be performed by exploiting the analytical performance models associated to the implementation templates. In this work, we present a software tool (P3L-prof) that supports programmers using skeleton/template programming based languages in the performance debugging/tuning of parallel applications. P3L-prof allows different “structures” (skeleton compositions) of a parallel application to be evaluated with respect to their expected performance onto a given target architecture before the application is actually coded and run. It also allow programmers to compare the expected performances on different target architectures in such a way that the “best performing” one can be individuated