题目：Solving two-phase shallow granular flow equations with a well-balanced NOC scheme on multiple GPUs
报告人：Prof. Yuan Li （袁礼），中国科学院/计算数学与科学工程计算研究所
摘要：In this talk, I will present a modified non-oscillatory central (NOC) difference scheme for numerical solution of two-phase shallow granular flow equations and its implementation on multiple GPUs. The governing equations are composed of mass conservation and momentum equations for solid and fluid phases with non-conservative terms in the momentum equations, which make the underlying quasi-linear equations have complex eigenvalues. Conventional Godunov methods based on Riemann problem solvers are hard to be applied to solve the equations. NOC scheme is a kind of finite volume methods that does not need to solve Riemann problems, so is worthy of being used to solve the two-phase shallow granular flow system. We designed non-oscillatory discretization scheme for the non-conservative terms and well balanced correction technique for bed slope source terms. To carry out fast simulations, we implement the modified NOC scheme on several K40 GPUs and compare the efficiency of openMP-CUDA and multi-stream-CUDA strategies.
报告人简介：Yuan Li received his BS, MS, and PhD degrees in 1986, 1989, and 1992 respectively from the department of flight vehicle design and applied mechanics of Beihang Universisty. He did postdoc during 1993-95 at Institute of Mechanics, CAS. He had been CN Yang visiting scholar at the department of physics of the Chinese University of Hong Kong in 1999. He was a recipient of “hundreds of talents” program of CAS in 1998-2001, and became a professor since 1999 at Institute of Computational mathematics, CAS. He had been members of editorial board of《Chinese Journal of Computational Physics》 and 《Physics of Gases》. His research areas include high resolution high order numerical methods for modeling multiphase flows, chemically reacting flows, and debris flows.