Multi-stream Parallel Optimizing for GPU Implementation of Reverse Time Migration

Reverse time migration (RTM) based on the full wave equation discretization are tools of major importance because they give an accurate representation of complex wave propagation areas. However, these applications impose high computational requirements, thus efficient parallelization of the RTM algorithm is crucial. In this paper, we present a multi-stream optimizing approach for the RTM algorithm on the latest Kepler architecture GPU. Our approach take full advantage of the Hyper-Q new feature of the Kepler architecture GPU so that multiple streams generated from a quantity of OpenMP threads operating on the CPU multicore processor can be rapidly running on massive fine-grain GPU processing cores. Experimental results show that our approach 17% performance improvement to the single stream implementation on Nvidia Tesla K40 GPU.