Research On OpenMP 4.0 Based Heterogeneous Parallel Computing Techniques For CFD Applications

Computational Fluid Dynamics(CFD)uses numerical methods to simulate the actual fluid flow.It is of practical engineering value and has been increasingly applied to engineering.But its numerical simulation requires fine meshes and a lot of time steps,resulting in huge computational overhead.In recent years,the " main processor +accelerator or co-processor" approach to the work of all-nuclear heterogeneous high-performance computers are increasingly becoming the mainstream.So how to use modern high-performance computer systems to achieve efficient simulation becomes very important.The LESAP(Large Eddy Simulation for Air-breathing Propulsion)is a typical CFD application software developed by the National University of Aeronautics and Astronautics.In this paper,how to realize the efficient combustion simulation software of the scramjet engine using modern high performance computer system is studied,and the efficient parallel computing technology based on OpenMP4.0 for CFD software on the CPU + MIC all-core heterogeneous platform is studied.Combining with the architecture features of the CPU + MIC all-core heterogeneous platform,the LESAP porting to the CPU + MIC hybrid parallel platform is realized by the new OpenMP4.0programming standard,to achieve an efficient and scalable scramjet engine combustion simulation software.The main works of this paper as follows.(1)Aimed at the general CPU platform,this paper uses OpenMP compiler instructions to achieve the MPI / OpenMP mixed parallel of LESAP software with OpenMP parallel efficiency of 80%.Then,according to the LESAP code characteristics,the power optimization algorithm,manual algorithm optimization and communication optimization are applied.When simulating a real-world combustion simulation configuration with 530,8866 grid cells on one “ Tianhe-2 ” supercomputer node consists of two 12-core Intel Xeon E5-2692 CPUs and three Intel Xeon Phi 31S1 P coprocessors,the speedup of optimized LESAP is 7.95.(2)For heterogeneous parallel computing of CFD applications,this paper presents using OpenMP 4.0 to port the LESAP on the heterogeneous CPU + MIC platform.This paper presents a load balancing optimization method based on grid partitioning,for heterogeneous multi-core platform composed of general-purpose CPU and MIC.And SIMD vectorization and data transmission optimization method are applied.When simulating a real-world combustion simulation configuration with 532,0896 grid cells on one “Tianhe-2” supercomputer node consists of two 12-core Intel Xeon E5-2692 CPUs and three Intel Xeon Phi 31S1 P coprocessors,the heterogeneous computing achieves a speedup of 3.07 times over the original CPU only computing.(3)In order to verify the scalability of the CPU + MIC version of LESAP,this paper does a large-scale scalability test on Tianhe-2.The code scales form 10 nodes to640 nodes with a parallel efficiency of 75%.The largest number of cores used is 14720,and the maximum grid size is over 3.3623 billion.