Mark F. Adams

Reducing Data Movement in the Galerkin Product of Block Algebraic Multigrid on GPUs

A new prolongator filtering, a new PETSc GAMG algorithm that drops small blocks from the coarse space under a Frobenius criterion with a kernel-preserving projection, and a new PETSc GAMG algorithm that reduces P^TAP traffic, coarse-operator fill, and memory, and cuts the hot time on the fine grid with iteration counts...