1: \begin{abstract}

2:   Multigrid methods are well suited to large massively parallel computer architectures

3:   because they are mathematically optimal and display excellent parallelization

4:   properties. Since current architecture trends are favoring regular compute patterns to

5:   achieve high performance, the ability to express structure has become much more

6:   important. The \hypre{} software library provides high-performance multigrid

7:   preconditioners and solvers through conceptual interfaces, including a semi-structured

8:   interface that describes matrices primarily in terms of stencils and logically

9:   structured grids. This paper presents a new semi-structured algebraic multigrid (SSAMG)

10:   method built on this interface. The numerical convergence and performance of a CPU

11:   implementation of this method are evaluated for a set of semi-structured problems. SSAMG

12:   achieves significantly better setup times than \hypre{}'s unstructured AMG solvers and

13:   comparable convergence. In addition, the new method is capable of solving more complex

14:   problems than \hypre{}'s structured solvers.

15: \end{abstract}

16: