Accelerating a Stochastic Seismic Inversion Algorithm using OpenCL-based Heterogeneous Platforms

Seismic inversion algorithms have been playing a key role in the characterization of oil and gas reservoirs, where the retrieved subsurface elastic models need to be reliable to improve decision making, for example decisions about well locations, thus maximizing the extraction of such resources. Since these algorithms usually rely in computer simulations that generate, process and store huge amounts of data, their usage in the industry is often limited by their long execution times. As a result, the demand for an efficient and accelerated execution of these algorithms is not only required to decrease the time to decision, but also to allow the development of larger and higher resolution models of the subsurface. In accordance, this thesis proposes a novel parallelization approach of a state of the art Stochastic Seismic Amplitude versus Offset Inversion algorithm, exploiting heterogeneous computing platforms based on a unified OpenCL programming framework, thus enabling the application to be executed in devices with different architectures and from different vendors. To take full advantage of the computational power made available by systems composed of multiple (and possibly different) CPUs and GPUs, a spatial division of the simulation space is performed, enabling the parallel simulation of multiple regions of the geological model. With this approach, a performance speed-up as high as 27.65× using two distinct GPUs was achieved, without compromising the accuracy of the obtained models.