<p>The Material Point Method (MPM) is a numerical method commonly used for solid mechanics simulations involving large deformations, history-dependent materials, or complex interactions such as contact and fracture. MPM can be treated as a finite element method where the integration points are allowed to move independently of the mesh. Matrix-free implementations of finite element operators provide higher performance on modern supercomputing hardware due to the higher arithmetic intensity, as measured by FLOPs per byte of memory transferred from memory, and the lower memory bandwidth requirements per degree of freedom when compared to assembled sparse matrices. In this paper, we derive a mathematical formulation for evaluation of finite element bases at arbitrary quadrature points per element and demonstrate the performance of matrix-free implementation of these MPM operators on GPU hardware. The measured performance compares favorably with matrix-free finite element operators on elements without a tensor product structure.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Matrix-free finite element methods with arbitrary quadrature point locations

  • Zachary R. Atkins,
  • Jed Brown,
  • Rezgar Shakeri,
  • Jeremy L. Thompson

摘要

The Material Point Method (MPM) is a numerical method commonly used for solid mechanics simulations involving large deformations, history-dependent materials, or complex interactions such as contact and fracture. MPM can be treated as a finite element method where the integration points are allowed to move independently of the mesh. Matrix-free implementations of finite element operators provide higher performance on modern supercomputing hardware due to the higher arithmetic intensity, as measured by FLOPs per byte of memory transferred from memory, and the lower memory bandwidth requirements per degree of freedom when compared to assembled sparse matrices. In this paper, we derive a mathematical formulation for evaluation of finite element bases at arbitrary quadrature points per element and demonstrate the performance of matrix-free implementation of these MPM operators on GPU hardware. The measured performance compares favorably with matrix-free finite element operators on elements without a tensor product structure.