Compile-time performance prediction of HPF/Fortran 90D

A novel interpretative approach for accurate and cost-effective performance prediction in high-performance computing (HPC) systems is described. An extensive characterization scheme composes the system and application components of the HPC area into well-defined parameters. An interpreter engine describes the performance of the composed application in terms of the parameters exported by the system. A source-driven High-Performance Fortran (HPF)/Fortran 90D performance prediction framework is implemented based on the interpretative approach. The numerical results via the benchmark kernels and applications from the Northeast Parallel Architectures Center HPF/Fortran 90D Benchmark Suite verify the accuracy, utility, and usability of the performance prediction framework.