GPU acceleration of FSI simulations by the immersed boundary-lattice Boltzmann coupling scheme

This paper proposes an approach to accelerate the simulations of fluid structure interaction (FSI) by implementing the immersed boundary-lattice Boltzmann coupling scheme on a single GPU chip. For the lattice Boltzmann simulations, a standard procedure is adopted, in which the one-dimensional (1D) blocks and shared memory are used to solve the misalignment problems in global memory transactions. For the immersed boundary simulations, the sub-processes, including the velocity interpolation, boundary force computation and force spreading, are implemented by different kernels. Besides, the 1D arrays are used to store the information of all Lagrangian points, which can exploit the parallelism of Lagrangian points without violating the coalescing and aligned access pattern. The relaxation of a three-dimensional (3D) ellipsoidal membrane is simulated to assess the performance of the GPU implementation. All the GPU kernels are found to be bandwidth bound and the achieved memory efficiencies for LB, IB spreading and IB interpolation & forcing kernels are up to 66%, 65% and 61% for a testing case. Finally, a strong FSI process of a 3D inflating and breaching balloon is simulated to demonstrate the capability of the present strategy. (C) 2016 Elsevier Ltd. All rights reserved.