UCNS3D

UCNS3D

UCNS3D is an open-source high-order Computational Fluid Dynamics solver for compressible flows on unstructured meshes. It is designed for flow simulations on modern parallel architectures, from MPI/OpenMP CPU clusters to selected GPU/accelerator kernels through the OpenMP target/XPU build path.

UCNS3D supports two- and three-dimensional simulations on mixed-element meshes and provides a flexible framework for high-order Finite-Volume and Discontinuous Galerkin methods, turbulence modelling, multicomponent flows, and large-scale aerodynamic flow problems.

The 2022 UCNS3D whitepaper provides a detailed description of the numerical methods, solver capabilities, and representative applications.

At a Glance

Capability Support
Flow regime Compressible inviscid and viscous flows
Meshes 2D and 3D unstructured mixed-element meshes
Elements Triangles, quadrilaterals, tetrahedra, pyramids, prisms, hexahedra
Numerics High-order finite-volume reconstruction and Discontinuous Galerkin methods
Parallelism MPI and OpenMP
Acceleration Optional OpenMP target offload, referred to in the code as the XPU path
Output ParaView, Tecplot, and VisIt-compatible files

Main Features

Repository Layout

Typical repository contents include:

src/          Fortran source code
docs/         documentation and whitepaper material
scripts/      example run scripts for local and HPC systems
tests/        representative test workflows
bin/lib/      optional third-party or platform-specific libraries

Additional documentation is provided in:

FILES.md       mesh and file-format description
PARAMETERS.md  runtime parameter description
TESTS.md       example-case description

Installation

UCNS3D can be used through Docker or compiled manually on the target machine. The canonical build entry point is src/Makefile, which provides compiler profiles selected with COMPILER=....

Docker

Install Docker Desktop. On Windows, use Windows Subsystem for Linux 2.

Build the image from the repository root:

docker build . -t ucns3d -f Dockerfile

Run the container interactively:

docker run -ti ucns3d

To mount a local working directory:

docker run -v $PWD/tmp/:/tmp/ -ti ucns3d

The current Dockerfile builds the GNU CPU/OpenMP executable with:

make -f Makefile COMPILER=gnu all

It also installs python3, which is used during compilation to derive case-specific fixed array bounds when UCNS3D.DAT is present.

Manual Build

The source code is written in Fortran and is intended to build on:

Common compiler environments include:

The solver requires mesh-partitioning libraries:

The repository build expects Tecplot, METIS, and ParMETIS libraries under bin/lib by default. Library paths can be overridden from the make command line.

CPU Build

Open a terminal in the source directory:

cd src

Select one of the Makefile compiler profiles.

GNU Fortran / MPI, CPU/OpenMP only:

make -f Makefile COMPILER=gnu clean all

Intel Fortran / Intel MPI, CPU/OpenMP only:

make -f Makefile COMPILER=intel clean all

macOS with GNU Fortran uses the same GNU profile. If libtecio, libparmetis, and libmetis are not in src, point the build at the directory containing the macOS .dylib or .a files:

make -f Makefile COMPILER=gnu MAC_LIB_ROOT=/path/to/macos/libs clean all

For an incremental rebuild, omit clean:

make -f Makefile COMPILER=gnu all

The executable is:

ucns3d_p

The folders under bin contain matching compiler-specific Makefile templates, but src/Makefile is the main build file.

Compile-Time Bounds

Several fixed work arrays use compile-time GPU_MAX_* bounds. Despite the historical name, these bounds are used by shared code paths and are relevant to both CPU and XPU builds.

The Makefile calls src/gpu_max_flags.py automatically. With no extra options, the script checks the current build directory for UCNS3D.DAT and companion files such as REALGAS.DAT, MULTISPECIES.DAT, and MULTISPECIES_DIFF.DAT.

The normal workflow is therefore:

cd src
cp /path/to/case/UCNS3D.DAT .
cp /path/to/case/REALGAS.DAT .     # if used by the case
make -f Makefile COMPILER=gnu clean all

If the input deck is outside the build directory, pass it explicitly:

make -f Makefile COMPILER=gnu GPU_MAX_CONFIG=/path/to/UCNS3D.DAT clean all

Rebuild whenever the case changes dimensionality, spatial order, number of equations, turbulence setting, passive scalars, or species configuration. If no input deck is visible at compile time, the helper emits conservative default bounds.

XPU / Accelerated Build

In UCNS3D, XPU refers to the accelerator build path enabled by the xpu preprocessor flag. It is used for selected kernels that contain OpenMP target offload regions and accelerator-specific data handling. The MPI ranks still run as normal host processes, while supported computational kernels may be offloaded to GPUs or other accelerator devices by the compiler runtime.

This is not a separate solver. It is the accelerated build of the same code base. XPU iLES path: For explicit iLES finite-volume simulations, the full runtime path is accelerated through the XPU implementation, including MPI communications.. Other solver modes may still execute routines on the host CPU where no XPU implementation is available.

On Cray systems, use the Cray compiler profile:

make -f Makefile COMPILER=cray clean all

The cray profile uses ftn, OpenMP, real64 defaults, and the -Dxpu preprocessor flag. It also uses the same automatic gpu_max_flags.py mechanism described above.

For an input deck outside the build directory:

make -f Makefile COMPILER=cray GPU_MAX_CONFIG=/path/to/UCNS3D.DAT clean all

Running a Simulation

A run directory normally contains:

Set the number of OpenMP threads:

export OMP_NUM_THREADS=N

Use N=1 for MPI-only execution.

Launch the solver with MPI:

mpirun -np M ./ucns3d_p

Here M is the number of MPI processes. At least two MPI processes are normally required.

Example Workflow

cd src
make -f Makefile COMPILER=gnu clean all

cd ../run_case
cp ../src/ucns3d_p .
export OMP_NUM_THREADS=1
mpirun -np 32 ./ucns3d_p

Machine-specific run scripts for local and HPC systems may be placed under scripts.

Visualization

UCNS3D output can be post-processed with:

Example Cases

Representative test cases are available from:

See TESTS.md for details.

License

UCNS3D is distributed under the GNU General Public License v3. See the LICENSE file for details.

Support

Questions, feedback, and suggestions can be sent to:

ucns3d@gmail.com