OpenSBLI: A framework for the automated derivation and parallel execution of finite difference solvers on a range of computer architectures

详细信息    查看全文

• 作者：Christian T. Jacobs ; ^{C.T.Jacobs@soton.ac.uk}Author Vitae ; Satya P. JammyAuthor Vitae ; Neil D. SandhamAuthor Vitae
• 关键词：High-Performance Computing ; Code generation ; Computational Fluid Dynamics ; Finite difference methods ; Graphics processing units
• 刊名：Journal of Computational Science
• 出版年：2017
• 出版时间：January 2017
• 年：2017
• 卷：18
• 期：Complete
• 页码：12-23
• 全文大小：3687 K
• 卷排序：18

文摘

A new numerical modelling framework called OpenSBLI is introduced. Users write their model equations in high-level Einstein notation. The code that performs the finite difference discretisation is automatically derived. Source-to-source translation targets the code towards different hardware backends. OpenSBLI is verified and validated with a suite of test cases.