The present invention proposes a three - level coarse - mesh finite - difference acceleration method for domestic DCU architecture, belonging to the cross - technical field of
neutron physics and high -
performance computing. The method includes: in the three - dimensional characteristic - line transport scanning stage, according to the domain -
decomposition method, the computing tasks are sent to the device side (domestic DCU). Inside the DCU, according to the coarse - mesh vertices, edges and faces to which the trajectory segments belong, the
neutron flux on the coarse - mesh surfaces to which the trajectory segments belong is stored, and the
neutron flux that needs data transfer at the communication - domain boundary is stored; in the three - level coarse - mesh finite - difference method, the calculation results of the characteristic - line three - dimensional characteristic - line transport solution are transferred to the three - level coarse - mesh finite - difference solution module. Among them, the three - level coarse - mesh finite - difference performs multiple coarsening acceleration calculations in terms of energy groups and geometric structures, that is, a coarser mesh with a higher coarsening degree is used to accelerate the iterative calculation of the previous - level
coarse mesh. The coarsening degrees of the three - level meshes from small to large are represented as the multi - energy - group
cell - level mesh, the single - energy - group
cell - level mesh and the single - energy - group
assembly - level mesh. Then, the finite - difference equations of each level of the
coarse mesh are constructed and solved using the successive over - relaxation iteration method. Finally, the fluxes in each flat - source region are updated according to the calculation results, and the effective
neutron multiplication factor in the current iteration is calculated. The present invention relates to the cross - technical field of neutron
physics and high -
performance computing.