Summary
GPU has become a booming technology in RTM to perform the expensive computation. Compared with saving the forward modeled wavefield on the disk, wavefield reconstruction method is a better way because computation is much faster than CPU-GPU data transfer. In this paper, we introduce the concept of effective boundary saving in reconstruction-based back propagation for GPU-based RTM. Starting from the computational strategies by Dussaud et al. (2008), we discuss the minimum requred saving amount in backward wavefield reconstruction for regular and staggered grid finite difference. Particularly, the minimum storage for staggered finite difference scheme using PML boundary condition is first explored. We demonstrate the proposed approach with numerical examples.
