Research Of Multi-component Seismic Data Anisotropy Pre-stacked Time Migration Parallel Algorithm

With the advent of the era of big data,the scale of the problems faced by researchers in various fields has become very huge.Traditional algorithms have become very limited in the face of such a large scale of problems.And the traditional algorithms take a long time to deal with large-scale data.At the same time,parallel computing has developed rapidly along with the advancement of computer hardware and software,and has been applied to some scientific research fields,improving the computational efficiency of the traditional algorithms.In seismic exploration,the pre-stack time migration algorithm based on Kirchhoff integrals is a common algorithm reflecting the underground structure.Based on this,using multiple components of seismic data and anisotropy of the actual underground structure can demonstrate the features of underground structures from multiple angles.However,with the high precision and high efficiency requirements of modern exploration,multi-component seismic data anisotropic pre-stack time migration algorithm takes a long time in practical applications.In order to solve the above problems,this paper implements a MPI-based CPU homogeneous parallel algorithm,a CUDA-based CPU and single GPU heterogeneous parallel algorithm,and a OpenMP-CUDA-based CPU and multiple GPUs heterogeneous parallel algorithmfor multi-component seismic data anisotropic pre-stack time migration algorithm.The CPU homogeneous parallel algorithm using coarse-grained peer-to-peer mode MPIaccelerating algorithm and distributing computational loads evenly,but the speedup ratio is limited to the number of processors.In GPU heterogeneous parallel algorithms,the reading and storage methods of seismic data and speed data are optimized based on the structural characteristics of the CPU and GPU.Using CUDA to achieve heterogeneous parallel computing of a single GPU,and then using OpenMP to achieve heterogeneousparallel computing between the CPU and multiple GPUs.The multi-GPU heterogeneous algorithm dividing the entire job into multiple computing cycles to reduce memory consumption andthe number of inlines computed by each calculation cycle is the number of GPUs.The algorithm use lots of GPU threads and each GPU thread is used for calculating the migration of one trace input seismic data to fully utilize the GPU computing power.The speedup that parallel computing to a 29 GB seismic data by CPU with different numbers of GPUs is compared.Between P-P wave and P-S wave seismic data,when parallel computing by 6 GPUs to cooperate with CPU,speedups are 444 and 449.