The Parallel Forward Modeling of the Wave Equation Based on AVX Instruction Set

This paper describes the parallel simulation of the memory/computing-intensive and large-scale three-dimensional acoustic wave equation with CPU stencil optimization. Taking the 8-core shared storage platform as an example, we obtain a one-time speed-up ratio of 6.7× compared with the serial program by using a coarse-grained OpenMP parallel scheme. Our method vectorizes the data on the template buffer with Single Instruction-Multiple Data techniques to further exploit the computing potential of the CPUs. We apply an 8-channel parallel vector to simulate seismic wave fields with the 256-bit AVX instruction set. This increases the computing bandwidth, thereby eliminating a significant volume of computing instructions and ultimately obtaining a secondary speed-up ratio of 3-7×. Finally, we analyze the factors affecting the secondary speed-up effect of AVX through complicated three-dimensional forward modeling experiments using the Salt model. The results indicate that the memory, cache, and register can better cooperate with each other when vectorization is conducted along the shortest direction of the model data cube, and that the speed-up effect can be enhanced by optimizing the AVX algorithm under such a principle.