Portable GPU-Accelerated Solver for Incompressible Two-Phase Flow with Diffuse Interface Method

This work presents a scalable and portable GPU-accelerated computational framework for simulating multiphase flows with phase change. The solver addresses incompressible two-phase flow by solving the Navier–Stokes equations and employs a direct FFT-based method for efficiently resolving the pressure field. The interface between fluids is modeled using a diffuse interface approach, wherein a smoothly varying order parameter transitions across a finite-width region and evolves according to an advection–diffusion equation. The code is developed in modern Fortran, featuring a hybrid MPI/OpenMP parallelization strategy for CPU-based execution and extended to GPUs using directive-based OpenACC acceleration. This approach enables reuse of the existing CPU code with minimal restructuring, significantly easing the porting process. Validation of the GPU implementation is carried out through quantitative comparisons with results from CPU-only runs, confirming consistency in accuracy and stability. Performance evaluations highlight strong scalability on both architectures, while also identifying key throughput bottlenecks and opportunities for further optimization.