| Details: |
The radial solver which calculates the wavefunctions of a Lippman-Schwinger (L-S) equation, consumes more than 75% of compute time in the simulations arising in KKR (Korringa-Kohn-Rostoker) family of methods. However, the global problem of solving the L-S equation can be decomposed into smaller, independent units, where the so-called local L-S equation is solved. This decomposition makes radial solver an ideal candidate for shared memory parallelization.
The current talk will initially introduce KKR method and then briefly touch upon the two paradigms in parallel computing, viz., the Message Passing Interface (MPI) and Open Multi-Processing (OpenMP). With this background, we would discuss the shared memory implementation of the radial solver and present the results on its performance and scalability studied in the KKR code based on density functional theory (DFT). |