Font Size: 

This paper presents a GPU-accelerated Smoothed Particle Hydrodynamics (SPH) solver enhanced by dynamic parallelism, designed for efficient multi-phase and multi-resolution simulations within coupled updated Lagrangian and total Lagrangian frameworks. The solver is developed on the Compute Unified Device Architecture (CUDA) dynamic parallelism paradigm and incorporates a dual-criteria time stepping integration strategy to reduce frequent CPU–GPU scheduling and data exchanges, thereby alleviating communication and synchronization overhead. Within this framework, the frequency of neighboring search operations in SPH is significantly reduced at the algorithmic level, while particle interaction computations fully exploit GPU dynamic parallelism for efficient execution, leading to a substantial improvement in overall computational efficiency. Systematic performance evaluations demonstrate that this solver exhibits good convergence and numerical stability across a wide range of fluid-soil-structure coupling problems. Significant speedups are achieved, with performance improvements of up to an order of magnitude compared with traditional GPU-accelerated SPH solvers in certain cases. These results indicate that this solver provides an efficient and versatile computational platform for high-fidelity simulations of complex multi-physics problems.

Smoothed Particle Hydrodynamics (SPH); CUDA Dynamic Parallelism, Dual-criteria time stepping; Verlet lists; Multi-resolution