Abstract
High-end computers such as T2K or the K computer requires hybrid parallel programming model because of their hierarchical hardware architecture. Conjugate gradient solver in a MPI parallel structural analysis application FrontISTR that invokes HEC-MW is parallelized in a MPI-OpenMP hybrid way. In the current implementation, 1.5 - 3.0 x speedup in SpMV operations and 1.5 x speedup in actual preconditioned CG solver have been achieved.
