A PERFORMANCE COMPARISON OF CONTINUOUS AND DISCONTINUOUS GALERKIN METHODS WITH FAST MULTIGRID SOLVERS

被引:63
作者
Kronbichler, Martin [1 ]
Wall, Wolfgang A. [1 ]
机构
[1] Tech Univ Munich, Inst Computat Mech, Boltzmannstr 15, D-85748 Garching, Germany
关键词
high-order finite elements; discontinuous Galerkin method; hybridizable discontinuous Galerkin; multigrid method; matrix-free method; high-performance computing; MULTILEVEL METHODS; HDG; ALGORITHMS; IMPLICIT; CG;
D O I
10.1137/16M110455X
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
This study presents a fair performance comparison of the continuous finite element method, the symmetric interior penalty discontinuous Galerkin method, and the hybridized discontinuous Galerkin (HDG) method. Modern implementations of high-order methods with state-of-the-art multigrid solvers for the Poisson equation are considered, including fast matrix-free implementations with sum factorization on quadrilateral and hexahedral elements. For the HDG method, a multigrid approach that combines a grid transfer from the trace space to the space of linear finite elements with algebraic multigrid on further levels is developed. It is found that high-order continuous finite elements give best time to solution for smooth solutions, closely followed by the matrix-free solvers for the other two discretizations. Their performance is up to an order of magnitude higher than that of the best matrix-based methods, even after including the superconvergence effects in the matrix-based HDG method. This difference is because of the vastly better performance of matrix-free Operator evaluation as compared to sparse matrix-vector products. A roofline performance model confirms the superiority of the matrix-free implementation.
引用
收藏
页码:A3423 / A3448
页数:26
相关论文
共 61 条
[1]   Parallel multigrid smoothing: polynomial versus Gauss-Seidel [J].
Adams, M ;
Brezina, M ;
Hu, J ;
Tuminaro, R .
JOURNAL OF COMPUTATIONAL PHYSICS, 2003, 188 (02) :593-610
[2]  
ADAMS M., 2016, HIGH PERFORMANCE GEO
[3]   The deal.II library, Version 9.0 [J].
Alzetta, Giovanni ;
Arndt, Daniel ;
Bangerth, Wolfgang ;
Boddu, Vishal ;
Brands, Benjamin ;
Davydov, Denis ;
Gassmoller, Rene ;
Heister, Timo ;
Heltai, Luca ;
Kormann, Katharina ;
Kronbichler, Martin ;
Maier, Matthias ;
Pelteret, Jean-Paul ;
Turcksin, Bruno ;
Wells, David .
JOURNAL OF NUMERICAL MATHEMATICS, 2018, 26 (04) :173-183
[4]  
[Anonymous], 2013, THESIS
[5]  
[Anonymous], 2005, FINITE ELEMENTS FAST
[6]   Unified analysis of discontinuous Galerkin methods for elliptic problems [J].
Arnold, DN ;
Brezzi, F ;
Cockburn, B ;
Marini, LD .
SIAM JOURNAL ON NUMERICAL ANALYSIS, 2002, 39 (05) :1749-1779
[7]   Algorithms and Data Structures for Massively Parallel Generic Adaptive Finite Element Codes [J].
Bangerth, Wolfgang ;
Burstedde, Carsten ;
Heister, Timo ;
Kronbichler, Martin .
ACM TRANSACTIONS ON MATHEMATICAL SOFTWARE, 2011, 38 (02)
[8]  
Bastian P, 2014, LECT NOTES COMPUT SC, V8806, P530, DOI 10.1007/978-3-319-14313-2_45
[9]   A massively parallel multigrid method for finite elements [J].
Bergen, Benjamin ;
Gradl, Tobias ;
Ruede, Ulrich ;
Huelsemann, Frank .
COMPUTING IN SCIENCE & ENGINEERING, 2006, 8 (06) :56-62
[10]  
Brenner S.C., 2008, MATH THEORY FINITE E, V3rd, DOI [10.1007/978-0-387-75934-0, DOI 10.1007/978-0-387-75934-0]