基于CUDA的GPU并行优化重力三维反演

摘要

笔者介绍了一种在PGI Fortran平台上开发的重力三维GPU并行反演算法. 该方法采用重加权正则化共轭梯度算法( Re-Weight Regularized Conjugate Gradient) ,可以在具有NVIDIA显卡的个人计算机上使用CUDA进行并行计算,无需借助工作站即可实现几十至上百倍的计算加速,提供稳定可信的反演结果. 并对可视化操作系统进行了优化,实现了在高端计算机系统上亿网格点的反演计算,同时在中、低端计算机也可以实现加速. 模型计算结果表明,该算法是一种高效且可靠的重力三维反演并行方法.%We introduce a new 3D parallelgravity inversion method based on CUDA GPU in PGI Fortran,which could be used in PCs and Laptops with NVDIA graphic cards to accelerate iteration speed for Re-Weight Regularized Conjugate Gradient method up to hundreds times.Storage and threads optimization was made for visual OS, leading a more than 0.1 billion cell's number for inversion in PCs.The result of model study shows that this method is an efficient and believable parallel computing algorithm.