Cache-oblivious parallel multigrid solvers on adaptively refined grids
Miriam Mehl, Christoph Zenger · mediaTUM (Technical University of Munich) · 2005
In many implementations of modern solvers for partial differential equations, the use of multigrid methods in particular in combination with dynamically adaptive grids causes a non-negligible loss of efficiency in data access and storage usage due to an increasing complexity of data structures.We propose a concept based on space-partitioning grids, spacefilling cuves, and stacks with the task to establish an efficiently parallelizable, storage saving, and, in particular, cache-oblivious multigrid solver.The side condition for this work is that we do not restrict the possibilities for arbitrarily local grid refinements.