Multilevel Interior Penalty Methods on GPUs

Abstract

We present a matrix-free multigrid method for high-order discontinuous Galerkin (DG) finite element methods with GPU acceleration. A performance analysis is conducted, comparing various data and compute layouts. Smoother implementations are optimized through localization and fast diagonalization techniques. Leveraging conflict-free access patterns in shared memory, arithmetic throughput of up to 39% of the peak performance on Nvidia A100 GPUs are achieved. Experimental results affirm the effectiveness of mixed-precision approaches and MPI parallelization in accelerating algorithms. Furthermore, an assessment of solver efficiency and robustness is provided across both two and three dimensions, with applications to Poisson problems.

References in corpus (16)

• MFEM: a modular finite element methods library
• Nodal Discontinuous Galerkin Methods on Graphics Processors
• Fast matrix-free evaluation of discontinuous Galerkin finite element operators
• Efficient Exascale Discretizations: High-Order Finite Element Methods
• Approximate tensor-product preconditioners for very high order discontinuous Galerkin methods
• Dissecting Tensor Cores via Microbenchmarks: Latency, Throughput and Numeric Behaviors
• GPU performance analysis of a nodal discontinuous Galerkin method for acoustic and elastic models
• GPU accelerated spectral finite elements on all-hex meshes
• Multigrid methods for Hdiv-conforming discontinuous Galerkin methods for the Stokes equations
• Geometric Multigrid for Darcy and Brinkman models of flows in highly heterogeneous porous media: A numerical study
• High-order matrix-free incompressible flow solvers with GPU acceleration and low-order refined preconditioners
• High-performance finite elements with MFEM
• Scaling to the stars -- a linearly scaling elliptic solver for $p$-multigrid
• Fast Tensor Product Schwarz Smoothers for High-Order Discontinuous Galerkin Methods
• An implementation of tensor product patch smoothers on GPU
• Smoothers with localized residual computations for geometric multigrid methods