FLOW36: A spectral solver for phase-field based multiphase turbulence simulations on heterogeneous computing architectures

We present FLOW36, a GPU-ready solver for interface-resolved simulations of multiphase turbulence. The simulation framework relies on the coupling of direct numerical simulation of turbulence, used to describe the flow field, with a phase-field method, used to describe the shape and deformation of a deformable interface and the presence of surfactants. An additional transport equation for a passive scalar can be solved to describe heat transfer in multiphase turbulence. The governing equations are solved in a cuboid domain bounded by two walls along the wall-normal direction where no-slip, free-slip or fixed/moving wall boundary conditions can be applied, while periodicity is applied along the streamwise and spanwise directions. The numerical method relies on a pseudo-spectral approach where Fourier series (periodic directions) and Chebyshev polynomials (wall-normal direction) are used to discretize the governing equations in space. Equations are advanced in time using an implicit-explicit scheme. From a computational perspective, FLOW36 relies on a multilevel parallelism. The first level of parallelism relies on the message-passing interface (MPI). A second level of parallelism uses OpenACC directives and cuFFT libraries; this second level is used to accelerate the code execution when heterogeneous computing infrastructures are targeted. In this work, we present the numerical method and we discuss the main implementation strategies, with particular reference to the MPI and OpenACC directives and code portability, performance and maintenance strategies. FLOW36 is released open source under the GPLv3 license.本研究提出FLOW36，一款适配GPU的多相湍流界面解析模拟求解器。该模拟框架耦合了湍流直接数值模拟（direct numerical simulation），用于刻画流场，与相场方法（phase-field method），用于描述可变形界面的形态、变形特性及表面活性剂的分布。此外还可通过求解被动标量（passive scalar）输运方程，刻画多相湍流中的传热过程。控制方程求解于长方体计算域：沿壁面法向由两个壁面界定，可施加无滑移、自由滑移或固定/运动壁面边界条件；流向与展向则均采用周期性边界条件。该数值方法采用伪谱离散格式（pseudo-spectral approach），在周期性方向使用傅里叶级数（Fourier series）、壁面法向使用切比雪夫多项式（Chebyshev polynomials）对控制方程进行空间离散。控制方程的时间推进采用隐式-显式耦合格式（implicit-explicit scheme）。从计算架构层面，FLOW36采用多级并行设计：第一级并行基于消息传递接口（Message Passing Interface，MPI）；第二级并行借助OpenACC编译指令（OpenACC directives）与cuFFT库（cuFFT libraries）实现，可在面向异构计算基础设施（heterogeneous computing infrastructures）时加速代码运行。本研究详细阐述了该数值方法，并探讨了核心实现策略，尤其围绕MPI与OpenACC编译指令、代码可移植性（code portability）、性能优化及维护方案展开讨论。FLOW36采用GPLv3开源许可协议发布。