Parallel Acceleration Of The Lattice Boltzmann Simulation Based On GPU-OpenACC

At the moment when traditional computational fluid dynamics is well developed,the lattice Boltzmann method is regarded as an emerging simulation method for flow field due to its unique mesoscopic numerical technique,simple implementation of algorithms and intrinsic data parallelism.It has a promising application foreground considering its suitableness in complex flow field problems such as multi-phase flow,turbulent flow simulation,engineering heat transfer and microfluidics etc.In the past decades,the innovation of numerical algorithm never ends,meanwhile,the computer hardware technology is developing with an amazing speed.The emerging GPGPU demonstrates its overwhelming computing potential which far exceeds the existing CPUs,and plays a crucial role in high performance computing.Hence,in this paper we combines the Open ACC parallel programming interface with LBM to transplant the CPU-based LBM algorithm code into a GPU computing platform in the hope of better computational efficiency.In the LBM numerical simulation of the two-dimensional Lid-driven cavity flow based on the single relaxation time D2Q9 model,the parallel simulation results under different Reynolds number and various mesh size consist well with the CPU-based serial results.With the increase of mesh size,the speedup also grows.Comparing with the serial version,its speed-up ratio rises from 13 times to 47times;While in the IB-LBM based cylinder flow simulation,the steady and unsteady flow simulation results demonstrate that via Open ACC,it allows a speedup around 22 times,in the situation of an elastically mounted cylinder,the number becomes 11.In the application of multiphase flow,the phase-field-based LBM simulation of 2D droplet impact result proves the parallel computing efficiency can be promoted to 32 times.The number would rise to 57 if we further increase the mesh size.Furthermore,we simulated the process of 3D droplet impact,the implementation of Open ACC brings a speedup around 20 times.On such basis we succeed in simulating the dynamics of droplet splitting on a mixed-wettability surface.All the numerical results above demonstrate the overwhelming advantage of GPU parallel computation via Open ACC in the field of LBM application.