Refining High Performance FORTRAN Code from Programming Model Dependencies

For next generation applications, programmers will be required to adapt to a new style of programming to utilize the parallelism in the processors available to them. Abstrac-tions in parallel programming languages and directives or annotations in sequential code have shown initial promise in reducing some of the burden of parallel programming. How-ever, even with all of these advances, parallel programming still requires skill beyond that possessed by an average pro-grammer. This is primarily due to the architecture-specific details in the domain. In this paper, we introduce a new approach to separate architecture dependencies from the program logic, enabling the programmer to execute the same computation in different platforms without actually changing the programming logic, but with capabilities to fine-tune the performance in the target platform. In this paper, we focus on refactoring existing FORTRAN code to support the latest HPC libraries. 1