NUMERICAL SIMULATION OF PLANETARY FLUID DYNAMICS ON CPU-MIC HETEROGENEOUS MANY-CORE SYSTEMS
Changmao Wu,Chao Yang,Liang Yin,Fangfang Liu,Qiao Sun,Ligang Li
2017-01-01
Abstract:Massively parallel computing is becoming a primary tool for the numerical simulation of planetary fluid dynamics.In this paper,Numerical simulation of the planetary fluid dynamics for distributed memory Xeon Phi-accelerated systems is studied.Firstly,we start from a legacy parallel code [1-3] using PETSc software package,which employs a pure MPI approach for parallel computing,to date,is in lack of support for multi-threaded parallelism on manycore accelerated systems,and then we extend the legacy code to multi-threaded parallelism on Xeon Phi-accelerated systems.Furthermore,based on PETSc software package,a sparse linear solver for Xeon Phi-accelerated cluster,which utilizes restarted generalized residual method(GMRES(m)),is presented and optimized.Secondly,a novel sparse matrix-vector multiplication(SpMV) algorithm for Xeon Phi-accelerated cluster is proposed,it combines highly aggressive use of asynchrony with offload,compute,communication,all of which serve the overlap of computation and communication.What's more,based on our SpMV algorithm,a polynomial preconditioner is given,which mainly consists of SpMV operations,hide and reduce communication,whether to local memory,across the network,or over PCIe.Finally,some optimized measures are taken to the extended code.Experiments on Tianhe 2 Supercomputer show that as compared to the original code,our Xeon Phi-accelerated design is able to deliver 6.93x and 6.00x speedups for single MIC device and 64 MIC devices,respectively.