GPU Parallelization of the Finite Element Toolkit Gascoigne 3d

These scripts accompany the publication An adaptive finite element multigrad solver using GPU acceleration by Manuel Liebchen, Utku Kaya, Christian Lessig and Thomas Richter. See the preprint on arXiv.

This is the GPU parallelization of the Gascoigne 3d multigrid method. Gascoigne 3d is a multi-purpose finite element toolkit, see its website for a brief description and the release on Zenodo. The most up-to-date version of Gascoigne 3d is hostet on a GitLab Server.

Installation 

Gascoigne 3d is based on cmake. External dependencies are SuiteSparse and Eigen. A brief description for installing Gascoigne is found on its website. To activate CUDA for GPU acceleration you have to set the flag USE_CUDA in cmake. Multithreading is controlled by the flag WITH_THREADS.

Test problems

Besides the library, a couple of test problems is compiled. Sources are found in GASCOIGNE/examples/ and the binaries are within the build directory in BUILD/bin/. The test-problems must be started from the directories GASCOIGNE/examples/.../ as parameter data and a mesh is read from the disc. The test-cases reproduce the data used in the numerical section of the manuscript An adaptive finite element multigrad solver using GPU acceleration.