Parallel Approach to High-Precision Smoke Simulation Based on a Multi-GPU Framework

In this paper, a distributed parallel algorithm for smoke flow field simulation based on a multi-graphical processing unit (GPU) framework is designed, aiming at the requirements of large-scale, high-precision and real-time performance. This algorithm is based on a smoke Navier-Stokes equation with more comprehensive external force interference term to be fully described under the action of various forces. The algorithm solves the problems of data partitioning for the whole flow field, flow field grid-state synchronization between multiple GPUs and global calculation synchronization of the whole framework by designing a variable-transfer and interrupt mechanisms. A general computing framework for GPU-based smoke flow field simulation is also proposed to adaptively partition and assign tasks for the smoke flow field simulation to provide a real-time, accurate solution for simulating smoke flow fields with large-scale data. Simulation experiments show that the algorithm can ensure the correctness of the calculation and offers good efficiency in simulating the large-scale smoke flow field.