PARALLELIZATION AND OPTIMIZATION OF THE CONSTRUCTION OF BLOCK COMPUTATIONAL GRIDS IN PREPROCESSOR OF "LOGOS" PROGRAM PACKAGE
V.V. Lazarev VANT. Ser.: Mat. Mod. Fiz. Proc 2018. Вып.1. С. 54-63.
The methods optimizing the block computational grid construction using the previously made decomposition of a geometric model into blocks are considered. Data structures representing the decomposition and block grids are described. The parallelized algorithms of constructing a block grid are described on the base of these representations. The plots of the parallelization efficiency and speedup versus the number of threads are presented. An algorithm that allows reconstructing a grid on the varied blocks only with respect to the modified decomposition into blocks has been also developed. For grids of sizes beyond the memory size, the fragment-by-fragment construction method with recording to separate files is proposed. A grid of 1 billion cells has been constructed. The developed algorithms have been implemented in the LOGOS preprocessor beginning from version 5.1. Keywords: parallelization, OpenMP, a block computational grid, geometry decomposition into blocks, preprocessor of the LOGOS program package.
|