Modelador de propagação de onda acústica 2D por diferenças finitas no domínio do tempo em CUDA
Seismic, modelling, propagation, parallel computing, CUDA
The full waveform inversion (FWI) is a quantitative seismic imaging method that can provide a high degree of reliance on subsurface structures. However, modeling wave propagation is computationally very exhaustive, thus limiting FWI. Nowadays, with the evolution of hardware, it is already possible to use parallel computing to accelerate FWI calculations. However, to make use of these tools, a different programming approach is required. In this work, the CUDA platform is used to develop a code that makes use of the parallel processors available in graphic cards (GPUs) to model the propagation of the acoustic wave 2D using data of speed and density, besides the use of staggered grid. In order to verify the performance of the code in different architectures, it was tested in two different ways: Intel CPUs, where the code was only in C language; Nvidia GPUs, where the code was modified for the CUDA platform. The results show that using GPUs with CUDA can greatly speed up computations when compared to CPU usage.