FINE-GRAIN PARALLELING METHODS IN THE TIM-2D CODE
A. A. Voropinov, I. G. Novikov, S. S. Sokolov VANT. Ser.: Mat. Mod. Fiz. Proc 2012. Вып.3. С. 24-33.
TIM-2D is a continuum mechanics simulation code, which uses arbitrary-shape unstructured polygonal Lagrangian grids. Paralleling in TIM-2D is provided at three levels: 1) paralleling in domains using an MPI-based distributed-memory model; 2) paralleling in para-domains, again using the MPI; 3) paralleling of computing cycle iterations using an OpenMP-based shared-memory model. The article addresses fine-grain paralleling algorithms (second level). These algorithms supplement the paralleling in domains and OpenMP-based paralleling implemented earlier. The fine-grain paralleling can be done both with para-domain overlapping in one row of cells and without overlapping. The methods are compared in their paralleling efficiency using one of test simulations.Keywords: TIM-2D code, shared-memory model paralleling, MPI, unstructured grids.
|