结构网格CFD应用程序在天河超级计算机上的高效并行与优化

摘要

How to improve the parallel performance of CFD applications with typical multi-block structured grid based on the CPU sub-platform of Tianhe-1A and CPU+MIC co-processor heterogeneous platform of Tianhe-2 supercomputer system ,is focused in this paper .Some strategies of performance optimization matched with both the characteristic of CFD application and the architec-tures of high-performance computing (HPC) platform are discussed in detail .Some numerical experiments are performed on Tianhe-2 supercomputer system with the maximum of grid cells achieving 6.8 × 1011 ,and the total amount of processors and/or co-proces-sors being 1.376 × 106 .It shows that the optimized code can get a speedup of 2.6 times faster on CPU and co-processor hybrid plat-form than that on the CPU platform only ,and good scalability is also observed from the test results .%对多区结构网格大规模CFD流场模拟的高效并行方法进行了研究，以天河超级计算机平台的CPU同构计算环境和CPU＋MIC异构计算环境为例，重点讨论了CFD应用特点与超级计算机运行环境相适应的性能优化与改进策略，发展了一系列多层次并行与性能优化方法．通过在天河2高性能计算平台上进行了多个算例的数值模拟，验证了这些优化方法的并行效果；在CPU＋MIC异构平台上模拟的最大CFD问题规模达到6800亿个网格单元，共使用137．6万CPU＋MIC处理器核，测试结果表明在CPU＋MIC异构平台上移植优化后的程序性能提高2．6倍左右，且具有良好的可扩展性．