Parallel Acceleration Improvement of Physical Quantity Gradient Algorithm Based on CFD

Abstract

In order to solve the problem that the physical quantity gradient algorithm consumes a lot of time in the calculation process, the algorithm is improved by parallel acceleration. In computational fluid dynamics (CFD), the key to the discretization of viscous terms is to calculate the gradient of various physical quantities (such as velocity, temperature, etc.) at the interface. By gauss-green formula, the gradient of each physical quantity in the center of the unit can be obtained by integrating the value of the center of the plane. In this paper, the gradient algorithm is optimized based on the application software PHengLEI, which will solve the gradient of four physical quantities on the “unit interface” successively from three directions. In order to solve the interface gradient more efficiently and improve the program execution efficiency and software performance, this paper will make use of diversified methods to optimize the performance of software PHengLEI, including using C++ class template to expand and vectorize, using Openmp to create multi threads for independent ‘for’ loop, using static scheduling method to reduce scheduling overhead and exchange loop sequence and so on. After many tests and verification, under the condition that the results are correct, the optimization effect has been significantly improved, reaching an acceleration ratio of 5666.67%.