GPU based Explicit Time Migration
The Explicit Time Evolution (ETE) method is an innovative Finite-Difference (FD) like method to simulate the wave propagation in acoustic media with higher spatial and temporal accuracy. However, different from FD, it is difficult to achieve an efficient GPU design because of the poor memory access patterns caused by the off-axis points and spatially variant coefficients.we present a set of new optimization strategies for ETE stencils according to the memory hierarchy of NVIDIA GPU. To handle the problem caused by the complexity of the stencil shapes, we design a one-to-multi updating scheme for shared memory usage. To alleviate the performance damageresulted from poor memory access pattern of variant coefficients, we propose a stencil decomposition method to reduce uncoalesced global memory access. Based on the state-of-the-art GPU architecture, combining with existing spatial and temporal stencil blocking schemes, we manage to achieve 9.6x and 9.9x speedups compared with a well-tuned 12-core CPUs version for 37-point and 73-point ETE stencils, respectively. Compared with a well-tuned MIC version, the best speedups for the 2 type stencils are 3.7x and 4.7x. Our designs leads to an ETE method that is 31.2x faster than conventional CPU-FD meth
1.This is the first work that manages to accelerate the ETE method kernels through the state-of-the-art GPU platforms.
2.We propose a set of novel optimization methods on GPU for stencils with spatially variant coefficients and complex shapes involving off-axis points, which have not been comprehensively investigated before. Based on GPU, combined with existing 2.5D spatial blocking and 1D temporal blocking strategies, we propose a 1-tomulti updating scheme to use shared memory for stencils involving off-axis points. In addition, we design stencil decomposition schemes for stencils with spatially variant coefficients. We gain approximately 10x speedup compared with a CPU version with 12 cores.
3.We also propose some insights on GPU and MIC architectures for optimizations of spatially variant coefficients stencils, which can provide a guidance for a board of FD-like forward modeling methods.
Jiarui Fang, Haohuan Fu, He Zhang,Wei Wu,Nanxun Dai,Lin Gan ,Guangwen Yang, Optimizing Complex Spatially-Variant Coefficient Stencils For Seismic Modeling on GPU, ICPADS 2015, CCF C
He Zhang，Haohuan Fu，Jiarui Fang，Wei Wu，Nanxun Dai， Lin Gan，Guangwen Yang，Parallel optimization algorithm for ETE seismic forward modeling in isotropic and TTI media.HPC China 2015，CCF O
Jiarui Fang,2th year PhD.student,Tsinghua University
He Zhang,1st year M.Sc.student,Tsinghua University