On November 12, the research group of the associate professor Huang Xiaomeng from DESS published a paper entitled OpenArray v1.0: A simple operator library for the botany of ocean modeling and parallel computing in Geoscientific Model Development (GMD), this paper proposes an efficient and automatically parallel pattern development framework OpenArray, based on which the 3D ocean model GOMO is constructed, and the decoupling between ocean model and parallel computing is realized. This work provides a new idea for the development of ocean model.

Ocean model research and development is a complex work, researchers need to possess solid domain knowledge, mathematical and computer parallel programming ability, and since mode of parallel program structure is complex, difficult to write, also computing platform is increasingly complex, how to in a variety of architecture platform (such as CPU, GPU, and Sunway) on efficient mode of development and application is the challenge for the community.

In this paper, the idea of computing middleware is adopted to develop OpenArray, an automatic parallel operator library, to provide 12 simple and automatically parallel basic operators to solve partial differential equations for pattern developers, and to encapsulate the tedious parallel model computation, so as to achieve the effect of "equation is code".

Based on OpenArray, develop regional ocean mode GOMO, 1860 lines of code for mode, support for common CPU and Sunway platform. After several optimization methods such as calculation chart, fusion core and communication hiding, the GOMO code can achieve the same parallel efficiency as the handwritten MPI, and reach the 68% expansion performance of 200,000 cores on Sunway platform.

Fig. Calculation efficiency and scalability of GOMO. Figure (a) and (b) respectively show the comparison of strong and weak scalability of GOMO and sbPOM, while figure (c) shows the scalability of GOMO on Sunway platform.

Huang Xiaomeng, associate professor of DESS is the first author and corresponding author of this paper. The collaborators include PhD students of DESS Huang Xing, Wang Dong, Wang Mingqing, Tang Qiang, Chen Yue, Fang Zheng, graduate student Wu Qi, Chen Yuwen, postdoctoral fellow Li Yi professor Yang Guangwen, and researcher of the institute of oceanography Song Zhenya.

The paper link：

https://www.geosci-model-dev.net/12/4729/2019/