Research And Application Of GPU Based Kirchhoff Pre-stack Depth Migration

The Kirchhoff pre-stack depth migration method is widely used in the industry because of its characteristics of precise imaging and rapid calculation.With the increasing requirements of oil and gas exploration for the imaging of complex structures in the mid-deep and strong velocity regions,the application of high-density seismic acquisition technology has a strong demand to the ability of Kirchhoff pre-stack depth migration program to process massive seismic data.With further breakthroughs in chip technology,NVIDIA GPUs,a multi-core processor solution specifically designed for computing,have begun to help the "high-performance computing" field to thrive,and combine with the oil industry driven by massive data processing to promote seismic data processing algorithms to update on the implementation platform.This study takes the Kirchhoff pre-stack depth migration as the core,designs and implements a parallel processing method based on the GPU implementation platform and achieves a better processing speed.The research mainly contains the following aspects:(1).Did the research on the hardware architecture,software implementation,execution configuration,memory management and data transformation of GPU acceleration platform.(2).Did the research on the ray-tracing method based on wavefront reconstruction and the theoretical derivation of Kirchhoff integral solution of scalar wave equation.(3).Analyzed and compared the computational efficiency of each module mainly based on the imaging part,combining the theoretical method with GPU hardware architecture and computer implementation,and designed a corresponding improvement plan.Firstly,the GPU parallel implementation of the computationally intensive offset imaging part is completed.Based on this,a new CUDA stream processing method is designed,which realizes the CPU part based on travel-time calculation overlapped with the GPU part based on imaging.The overlap calculation makes the overall theoretical processing time of the program completely dependent on the consumption of the travel-time calculation part.A series of corresponding optimization and improvement strategies are proposed.The register,arithmetic algorithm and condition judgment are optimized first,and the GPU occupancy rate is increased.Through the improvement of the texture memory and data transmission scheme,the overall processing speed is improved.