OpenACC + Athread collaborative optimization of Silicon-Crystal application on Sunway TaihuLight

• Published in: Parallel computing 2022-07, Vol.111, p.102893, Article 102893
• Main Authors: Liang, Jianguo, Hua, Rong, Zhu, Wenqiang, Ye, Yuxi, Fu, You, Zhang, Hao
• Format: Article
• Language: English
• Subjects: Athread; Molecular dynamics simulation; OpenACC; Silicon-Crystal application; SW26010
• ISSN: 0167-8191; 1872-7336
• DOI: 10.1016/j.parco.2022.102893

The Silicon-Crystal application based on molecular dynamics (MD) is used to simulate the thermal conductivity of the crystal, which adopts the Tersoff potential to simulate the trajectory of the silicon crystal. Based on the OpenACC version, to better solve the problem of discrete memory access and write dependency, task pipeline optimization and the interval graph coloring scheduling method are proposed. Also, the part of codes on CPEs is vectorized by the SIMD command to further improve the computational performance. After the collaborative development of OpenACC+Athread, the performance has been improved by 16.68 times and achieves 2.34X speedup compared with the OpenACC version. Moreover, the application is expanded to 66,560 cores and can simulate reactions of 268,435,456 silicon atoms. •Combining with the unique architecture of the SW26010 processor, we analyzed the characteristics of its main memory access modes and tested the memory access bandwidth between the MPE and CPEs. Based on the analysis and testing, the rules of porting the application on the SW26010 are given.•Basing on the data distribution, operation, and computing characteristics of the Silicon-Crystal application on the SW26010 processor, we pointed out the challenging problems ported the Silicon-Crystal application to the SW26010 processor.•After implementing OpenACC optimization, to better solve the problem of discrete memory access and write dependency, we propose task step-by-step pipeline optimization and the interval graph coloring scheduling method. Also, the part of codes on CPEs is vectorized by the SIMD command to further improve the computational performance.•The performance was evaluated and analyzed on the Sunway TaihuLight after the parallel optimization of the Silicon-Crystal application. The test results under the single CG show that the optimization methods in (3) have achieved good results. Meanwhile, it has better scalability in different situations from 8 to 1024 CGs, and the overall optimization effect is better than the parallel version of OpenACC.