地下多相流动数值模拟的GPU并行优化

摘要

In order to decrease the simulating time and getting more detailed simulating results, this article implements a parallel bi-conjugate gradient stabilized method on GPU. The solver was integrated to TOUGHREACT and used in the numerical simulation of underground multiphase flow on GPU. CUDA implementing methods for sparse matrix-vector multiplication and vector inner-product are given in this paper. The experimental results show that GPU can speed up the process of solving the underground multi-phase flow simulation. On different scales of simulation grids, the speed was increased by 1. 7—3. 4 times.%针对减少模拟计算时间及提高问题求解规模问题,基于设备编程架构(CUDA)将使用预处理的稳定双共轭梯度法在图形处理器(GPU)上实现,并将其整合到TOUGHREACT软件中,在GPU平台实现了对地下多相流动数值模拟问题的并行求解,并给出了稳定共轭梯度算法中最耗时的两个操作稀疏矩阵向量乘积和向量内积计算的GPU平台实现及优化方法.实验结果表明,GPU的使用对求解过程有良好的加速效果,针对不同的网格规模进行多相流模拟实验,达到了1.7～3.4倍的加速比.