Hybrid parallel reduction algorithms for the multi-level CMFD acceleration in the neutron transport code PANDAS-MOC

The coarse mesh finite difference (CMFD) technique is considered efficiently in accelerating the convergence of the iterative solutions in the computational intensive 3D whole-core pin-resolved neutron transport simulations. However, its parallel performance in the hybrid MPI/OpenMP parallelism is inadequate, especially when running with larger number of threads. In the original Whole-code OpenMP threading hybrid model (WCP) model of the PANDAS-MOC neutron transport code, the hybrid MPI/OpenMP reduction has been determined as the principal issue that restraining the parallel speedup of the multi-level coarse mesh finite difference solver. In this paper, two advanced reduction algorithms are proposed: Count-Update-Wait reduction and Flag-Save-Update reduction, and their parallel performances are examined by the C5G7 3D core. Regarding the parallel speedup, the Flag-Save-Update reduction has attained better results than the conventional hybrid reduction and Count-Update-Wait reduction.