Memory efficient adaptive mesh generation and implementation of multigrid algorithms using Sierpinski curves

Abstract

Abstract:We will present an approach to numerical simulation on recursively structured adaptive discretisation grids. The respective grid generation process is based on recursive bisection of triangles along marked edges. The resulting refinement tree is sequentialised according to a Sierpinski space-filling curve, which leads to both minimal memory requirements and inherently cache-efficient processing schemes. The locality properties induced by the space-filling curve are even retained throughout adaptive refinement of the grid. We demonstrate the efficiency of the approach by implementing a multilevel-preconditioned conjugate gradient solver for a simple, yet adaptive, test problem: solving Poisson's equation on a re-entrant corner problem.Keywords: adaptive grid generation; space-filling curves; cache efficiency; simulation.Reference to this paper should be made as follows: Bader, M., Schraufstetter, S., Vigh, C.A. and Behrens, J. (2008) 'Memory efficient adaptive mesh generation and implementation of multigrid algorithms using Sierpinski curves', Int.

Quasi-matrix-free Hybrid Multigrid on Dynamically Adaptive Cartesian Grids

“…As such, the present studies have academic character, and it is important in the future to weaken the single-sweep paradigm if it renders it possible to realize stronger smoothers. Candidates for suitable smoothers are 2-sweep Krylov schemes [5] or red-black Gauss-Seidel with pipelining which combines multiple sweeps [26]. While giving up on single touch harms implementational elegance, it might even turn out to be favourable from a parallelization point of view to run over the grid multiple times as long as the rank-local work increases faster than the exchanged data cardinality.…”

“…Better iterative or direct solvers are more reasonable in many cases. See [5] for a (preconditioned) CG that fits to the present matrix-free paradigma.…”

Data structures and algorithms for high-dimensional structured adaptive mesh refinement

“…Klöfkorn and Nolte described the implementation of the SPGrid interface of the DUNE framework [14,15], which allows anisotropic structured grids of higher dimensionality but only handles static meshes. An alternative approach to tackling higher dimensions is to discretize the domain into structured simplex meshes [16][17][18][19]. In [11], a two-level approach of combining hypercubes and simplices is presented.…”

Operational tsunami modelling with TsunAWI – recent developments and applications

“…In particular the representation of variables in an unstructured mesh, if not constructed carefully, can result in many cache misses during calculation and considerably slow down the code. For TsunAWI simulations we therefore order the variables along a space filling curve (SFC, for triangular meshes with regular structure see Bader et al, 2008), which guarantees good data locality on all levels of the memory hierarchy by construction. The ordering is visualised in Fig.…”