一种基于GPU的逆时偏移并行算法

A GPU-BASED PARALLEL ALGORITHM OF REVERSE TIME MIGRATION

逆时偏移可以对来自大角度甚至超过90°反射界面的反射波精确成像,是解决地下复杂构造成像的有力工具,但是由于传统PC集群和基于该架构的逆时偏移应用软件无法满足其对计算能力的需求,因而阻碍了该方法的应用和推广。提出一种基于GPU的逆时偏移并行加速算法,从而极大提高逆时偏移的并行度和运算效率,使逆时偏移的运行效率相对于PC集群达到数十倍的提高。将逆时偏移和单程波动方程成像结果进行比较,结果表明:基于GPU的逆时偏移的并行算法不仅能满足巨大的运算量需求,同时其成像效果更精确。

Reverse time migration can precisely image the reflected wave from a reflector with high-angle or even exceeds 90. It is a very efficient means to solve the imaging problem of underground complex structure, but the traditional PC cluster and the application software of reverse time migration based on PC cluster hinder the application and promotion of it because they can hardly meet its demand on computing power. In this paper, a GPU-based parallel acceleration algorithm for reverse time migration is presented; it greatly improves the parallel efficiency and the computational efficiency of reverse time migration, and makes the operation efficiency of reverse time migration method achieves dozens of times improvement with respect to PC cluster. At the same time, the imaging result of reverse time migration is compared with that of the one-way wave equation, the result demonstrates that the GPU-based parallel algorithm of reverse time migration can satisfy the demand of huge amount of computation, and also has more accurate imaging results as well.