Research On Parallel Acceleration Of Seismic Data Processing Based On

In the part of seismic data processing, As the processing object and the complexity of the algorithmt is becoming higher and higher,as an important part of high precision, revers-time migration and many other seismic data processing, Seismic wave forward modeling of simulation on the demand of computing speed and efficiency is also becoming higher and higher, The traditional CPU-based serial algorithm has been unable to meet the increasingly complex needs of seismic computational efficiency, At present, the computer cluster parallel computing system of the actual work facing the problem about system energy consumption and maintenance costs.The GPU has its own characteristics of parallel architecture, And low power consumption, small cost, those make GPU parallel computing can solve these problems now.When using the finite difference method solving seismic wave equations, the numerical dispersion problem has a great influence on the simulation results. Through the experiment contrasts, we have several methods deal with numerical dispersion. Then, studied the PML (perfectly matched layer) absorbing boundary conditions and treatment of boundary reflection. Through the experiment contrasts, we know that boundary conditions are important to the forward modelling, and deduce the corresponding discrete finite difference form of acoustic wave equation, finite difference method and absorbing boundary treatment are very suitable for doing parallel computing on GPU platform.NVIDIA company has developed a unified device architecture platform based on GPU, CUDA is on the based of hardware support, it is a software platform when GPU used multi-thread parallel computing, When dealing with finite difference calculation’s dispersion of seismic wave field, because of the increasing difference order and the grid size result in the low calculation efficiency, and, on subdivision mesh finite difference of seismic wave field computation, grid on each node of numerical calculation is quite parallelism, the finite difference forward modeling is very suitable for doing multithreaded parallel computing on GPU platform.In this paper, based on the software structure of CUDA, the specific method and process of solving the finite-difference numerical value of two dimension wave equation is achieved. At the same time, by comparing and analyzing the operational efficiency in different hardware environments, the CUDA software platform is fully proved to be able to obviously improve the computational efficiency of seismic wave forward modeling, which offers the possibility for seismic data processing technically.