MHIT36: A phase-field code for GPU simulations of multiphase homogeneous isotropic turbulence

We present MHIT36, a GPU-tailored solver for interface-resolved simulations of multiphase turbulence. The framework couples direct numerical simulation (DNS) of the Navier–Stokes equations, which describe the flow field, with a phase-field method to capture interfacial phenomena. Simulations are performed in a cubic domain with periodic boundary conditions applied in all three spatial directions. The governing equations are discretized using a second-order finite difference scheme. The Navier–Stokes equations are integrated with an explicit fractional-step method, and the resulting pressure Poisson equation is solved using a fast Fourier transform (FFT)-based approach. The accurate conservative diffuse interface (ACDI) formulation is used to describe the transport of the phase-field variable. From a computational standpoint, MHIT36 employs a two-dimensional domain decomposition to distribute the workload across MPI tasks. The cuDecomp library is used to perform pencil transpositions and halo exchanges, while the cuFFT library and OpenACC directives are leveraged to offload the remaining computational kernels to the GPU. This parallelization strategy enables MHIT36 to achieve an excellent scaling efficiency on 1024 GPUs, while maintaining a structure that is easy to extend and modify. MHIT36 is released open source under the MIT license.

Program summary

Program Title: MHIT36

CPC Library link to program files: https://doi.org/10.17632/yb2dt99swr.1

Developer's repository link: https://github.com/MultiphaseFlowLab/MHIT36

Licensing provisions: MIT License

Programming language: Modern Fortran

Nature of problem: Solving the three-dimensional incompressible Navier-Stokes equations in a triply-periodic box. A phase-field method based on the accurate conservative diffuse interface (ACDI) formulation is used to describe the shape and topological changes of the interface.

Solution method: The system of governing equations is advanced in time using an explicit strategy while the governing equations are discretized in space using a second-order finite difference approach. A fractional step is used to solve the Navier-Stokes equations and an FFT-based method is used to solve the resulting Poisson equation for pressure. The parallelization relies on a 2D domain decomposition strategy and all intra- and inter-node communications are handled by the cuDecomp strategy. The cuFFT library and OpenACC directives are used to entirely offload code execution to GPUs.