The Use of Parallel Programming Technologies for Modeling Seismic Waves Using Grid-Characteristic Method

Abstract

INTRODUCTIONThe problem of wave propagation in linearelastic medium is of interest in seismology and geophysics. With the increasing performance of modern computer systems and the possibility of parallel execution of the program code, it is possible to carry out computer modeling of seismic waves with great precision. The question is how to efficiently use the available computing resources.The first solution to this problem is to use of technologies of parallel programming for shared memory systems. Abstract. The paper deals with the problem of modeling seismic wave propagation in solid deformable bodies. The state of these bodies can be represented as a solution of hyperbolic system of equations. Grid-characteristic method is used to find the solution. This formulation of physical problem is used to demonstrate the use of technologies of parallel programming for shared memory systems using OpenMP and POSIX Threads technology. To achieve high performance, vector SSE and AVX instructions of CPU is used. In addition, GPU implementation is written using CUDA and OpenCL technologies. Results of multi-GPU parallelisation is also obtained. Use of NVIDIA GPUDirect technology that allows GPUs to make data exchanges directly through the PCI Express bus is considered. The paper demonstrates achieved acceleration and percentage of the theoretically possible performance. The paper demonstrates achieved acceleration and percentage of the theoretically possible performance. The paper describes the optimizations, which allowed us to achieve the obtained acceleration results. The study was financially supported by RFBR as part of a research project number 15-07-01931 A.