Fine-grain parallel smoothing by asynchronous iterations and incomplete sparse approximate inverses for computational fluid dynamics

We demonstrate the use of fine-grain parallel iterations in multigrid smoothers for com-putational fluid dynamics. Asynchronous iterations, incomplete sparse approximate inverses and their combinations are proposed and their effectiveness studied for a few benchmark cases of external aerodynamics. Our implementations of these solvers have focused on exploiting the wide vector units and the large number of cores on the Xeon Phi Knights Landing proces-sors. The new smoothers have been tested on some benchmark aerodynamics cases. Two of the proposed solvers, block symmetric Gauss-Seidel and asynchronous block incomplete LU factorization, both with incomplete sparse approximate inverse application, have been found to be promising.