Accelerating Kirchhoff Pre-stack depth migration on a GPU by overlapping ray tracing and imaging

Abstract

Kirchhoff-based depth migration is in great demand for imaging at large scales and high-density seismic data exploration. A Graphic Processing Unit, or GPU, can compose a heterogeneous hardware that contains a traditional Central Processing Unit (CPU), and it can greatly accelerate migration calculations. An improved acceleration method is proposed here that uses overlapped ray tracing, which provides two-way travel times, and an imaging kernel. In our method, ray tracing is calculated by the CPU due to its complicated divergences. The imaging kernel is passed to the GPU by unlooping the imaging volume. Streams and asynchronous calculation functions are used to overlap the imaging kernel of a previous trace to the ray tracing of the next trace. The nvprof profiling tool, along with the elapsed time, are used to measure the efficiency of the accelerated work, and it is found that the GPU code results in a speedup factor of 60 relative to single thread CPU code, and when using the overlapped computation, the GPU efficiency is improved by 30%. This method shows great potential for further application in large scale seismic data imaging with GPU.