Estimating and Optimizing Performance for Parallel Programs

P 3 T is an interactive performance estimator that assists users in performance tuning of scientiic Fortran programs. It detects performance bottlenecks in the program, identiies the causes of performance problems, and guides the user in selecting eeective program transformations in order to gain performance. Four of the most critical performance aspects of parallel programs are estimated: load balance, cache locality, communication and computation overhead. P 3 T is an integrated tool of the Vienna Fortran Compilation System, which enables the estimator to aggressively exploit considerable knowledge about the compiler's analysis information and code restructuring strategies. We evaluate existing features and describe substantial enhancements in three key areas: graphical user interface, performance parameters and input programs. P 3 T 0 s graphical user interface directs the user to bottlenecks in a computation that prevent the program from performing well. In addition, it allows ltering and visualizing of performance data at various levels of detail. A new performance parameter, which models computation time is discussed. Furthermore, P 3 T has been extended to handle a subset of HPF programs. Experiments show the ability of P 3 T to successfully guide the user in performance tuning of parallel programs. This includes the selection of performance eecient data distribution strategies and prootable program transformations. The results are very encouraging, and demonstrate the feasibility of automatic performance prediction for parallel programs.