Answers are grounded in this paper and cite their sources.
A complete suite of algorithms to implement parallel assembly of sparse finite element matrices on multicore computers is presented. The approach is broken down into six algorithms. The goal is to cater both to nonlinear computations, where it is possible to amortize sequential algorithms over multiple assembly passes, and linear computations, where sequential algorithms cannot be amortized. Therefore, we consider the entire process, starting with the mesh, and finishing with the sparse matrix in the compressed column format, in order to compare meaningfully parallel and sequential implementations. Importantly, the entire construction of the sparse matrix is parallelized. Numerical experiments are carried out on five distinct multicore computers (including A64FX and Nvidia Grace systems) with up to 144 cores, and the results indicate that decent scaling can be obtained up to 48 computing threads. The algorithms are described in the programming language Julia, and their complete source code is available on Github.
No takes yet. Share an insight, caveat, or question.
Petr Krysl (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: