Matrix-free framework, step-37, step-48
A new interface for matrix-free implementation of finite element operators has
been merged to the library. Functionality for CPU and memory efficient
implementations is provided, especially beneficial for high-order elements
(sparse-matrix computations are outperformed for orders two and higher in
common applications). The implementation is extensively parallel, including MPI
for distributed meshes, thread-parallelized for shared-memory computations, and
usage of processors' vector units. As part of the project, the evaluation of
(high-order) Gauss-Lobatto elements has been improved which are especially
efficient in the matrix-free framework due to the diagonality of the function
evaluation (and thus mass matrices). The new example program step-48
demonstrates how to use the framework to solve a nonlinear problem and how the
special structure of the Gauss-Lobatto elements can be exploited. Another
example, step-37, explains how to use the matrix-free framework in a multigrid
setting.
Martin Kronbichler and Katharina Kormann
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
