A High-Order WENO Finite Difference Water Wave Model for Interactive Ship-Wave Simulation

We present recent work directed towards merging new trends in parallel computing together with recent improvements of a water wave model. The goal of this work is the development of efficient and accurate interactive simulations of ship-wave interaction for use in modern ship simulators for inexpensive training of naval officers. To enable interactive and realistic shipwave simulations it is of fundamental importance to address both i) modeling issues such as discretization of discontinuous and moving transitions in the free surface water wave profile as a result of moving ship hulls and ii) requirements for enabling simulations at interactive speed using modern many-core architectures. At the conference, we propose and present preliminary results regarding the use of a highorder Weighted Essentially Non-Oscillatory (WENO) finite difference discretization for the discretization of linearized free surface boundary equations of a unified potential flow model [?, ?] (referred to as OceanWave3D). The model is formulated in a moving frame of reference fixed to move with the ship hull. The goal of using a WENO method for the discretization is to increase robustness and improve accuracy without sacrificing overall performance in a massively parallel implementation. To be able to do interactive simulations, the wave simulation software will be based on a recently developed massively parallel and scalable iterative algorithm [?], namely, an efficient multigrid preconditioned defect correction method. In combination with modern manycore commodity hardware such as Graphics Processing Units (GPUs) our preliminary analysis suggest that performance close to required interactive rates can be achieved through stringent algorithmic requirements and tuning the implementation for efficient utilization of computational resources. The simulation software is to be based on a recently developed generic library, which targets fast proto-typing of Partial Differential Equation (PDE) solvers on modern many-core architectures, e.g. such as many-core GPUs with high on-chip bandwidths. The work is carried out in the Graphics Processing Unit Laboratory (GPUlab) of Department of Informatics and Mathematical Modeling, Technical University of Denmark in collaboration with Department of Mechanical Engineering.