FEMaster

FEMaster is a C++17 finite-element solver for structural mechanics. One CMake build now covers Linux/WSL, Windows, and macOS. Intel oneMKL is supported on Linux and Windows; CUDA and optional cuDSS are supported on Linux and Windows.

Features

• Linear and geometrically nonlinear static analysis
• Eigenfrequency, buckling, and implicit Newmark analyses
• Beam, shell, solid, truss, and point-mass models
• Supports, couplings, connectors, ties, equations, and nonlinear contact
• Eigen CPU solvers with optional oneMKL/PARDISO acceleration
• Optional CUDA solvers and optional NVIDIA cuDSS
• Built-in input-deck DSL documentation

Requirements

Always required:

• CMake 3.24 or newer
• A C++17 compiler
• Eigen, Spectra, and argparse headers

Optional components:

• Ninja for Linux/macOS presets
• Visual Studio 2022 for Windows presets
• Intel oneAPI oneMKL plus the oneAPI compiler runtime for MKL presets
• CUDA Toolkit for CUDA presets
• NVIDIA cuDSS for cuDSS presets
• GoogleTest when FEMASTER_BUILD_TESTS=ON

CMake first searches installed packages and then common include directories. A custom header root containing Eigen/, Spectra/, and argparse/ can be set with:

-DFEMASTER_DEPS_INCLUDE_DIR=/path/to/include

Preset workflow

List the presets available on the current host:

cmake --list-presets

Configure and build with the same preset name:

cmake --preset linux-release
cmake --build --preset linux-release

Build presets use eight parallel jobs. CMake prints the selected compiler, OpenMP runtime, MKL threading layer, CUDA version, cuDSS version, and runtime linkage during configuration. Ninja and Visual Studio provide the per-file progress output.

Linux and WSL

CPU build with OpenMP:

cmake --preset linux-release
cmake --build --preset linux-release

For MKL, initialize oneAPI in the shell first:

source /opt/intel/oneapi/setvars.sh
cmake --preset linux-mkl
cmake --build --preset linux-mkl

linux-mkl uses threaded, statically linked MKL and links libstdc++/libgcc statically. linux-mkl-sequential selects sequential MKL. The resulting file is:

build/linux-mkl/bin/FEMaster

CUDA variants are:

cmake --preset linux-cuda
cmake --build --preset linux-cuda

cmake --preset linux-cuda-mkl
cmake --build --preset linux-cuda-mkl

For cuDSS, set its installation root before configuring:

export CUDSS_DIR=/path/to/libcudss-linux-x86_64
cmake --preset linux-cuda-cudss
cmake --build --preset linux-cuda-cudss

linux-full enables CUDA, cuDSS, and oneMKL together.

Windows

Run CMake from an Intel oneAPI command prompt for Visual Studio 2022 when using MKL. A normal Visual Studio x64 prompt is sufficient without MKL.

cmake --preset windows-mkl
cmake --build --preset windows-mkl

The executable and license files are written to:

build\windows-mkl\bin\Release\FEMaster.exe

The threaded static-MKL preset still needs Intel's OpenMP runtime DLL. CMake finds libiomp5md.dll in the oneAPI installation and copies it beside the executable automatically. Use windows-mkl-sequential if that runtime is not wanted.

CUDA variants are configured from a prompt where the CUDA Toolkit is visible:

cmake --preset windows-cuda
cmake --build --preset windows-cuda

cmake --preset windows-cuda-mkl
cmake --build --preset windows-cuda-mkl

Set CUDSS_DIR to the unpacked cuDSS root for windows-cuda-cudss. windows-full enables CUDA, cuDSS, and oneMKL together.

macOS

macOS provides CPU builds:

cmake --preset macos-release
cmake --build --preset macos-release

After installing an OpenMP runtime discoverable by CMake, use macos-openmp. Current NVIDIA CUDA toolkits do not support macOS, so CMake rejects CUDA on that platform intentionally. No macOS MKL preset is provided; the supported MKL targets for this project are Linux and Windows.

CMake options

Option	Default	Meaning
FEMASTER_ENABLE_OPENMP	OFF	Enable FEMaster OpenMP loops
FEMASTER_ENABLE_MKL	OFF	Enable oneMKL/PARDISO
FEMASTER_MKL_SEQUENTIAL	OFF	Use sequential instead of threaded MKL
FEMASTER_MKL_STATIC	ON	Link MKL component libraries statically
FEMASTER_ENABLE_CUDA	OFF	Build CUDA solver backends
FEMASTER_ENABLE_CUDSS	OFF	Use cuDSS; requires CUDA
FEMASTER_DOUBLE_PRECISION	ON	Use double precision on CPU and CUDA
FEMASTER_STATIC_RUNTIME	OFF	Statically link compiler C/C++ runtimes
FEMASTER_FULLY_STATIC	OFF	Request a fully static Linux executable
FEMASTER_BUILD_TESTS	OFF	Build GoogleTest tests
FEMASTER_TIME_REPORT	OFF	Enable compiler timing diagnostics

Options can override a preset:

cmake --preset linux-release -DFEMASTER_DOUBLE_PRECISION=OFF
cmake --build --preset linux-release

For CUDA architecture selection, use CMake's standard setting, for example:

-DCMAKE_CUDA_ARCHITECTURES=86

FEMASTER_FULLY_STATIC is separate from the distributable MKL presets. Fully static glibc builds depend on static versions of every system library and are incompatible with CUDA.

Runtime distribution

CPU presets without dynamic third-party libraries need no FEMaster-specific runtime installation. The static MKL presets embed MKL itself. On Windows, threaded MKL additionally ships the automatically copied libiomp5md.dll.

CUDA cannot be folded into a single portable executable: CUDA builds require a compatible NVIDIA driver and may require CUDA/cuDSS runtime libraries according to NVIDIA's redistribution rules. Windows cuDSS builds copy the imported cuDSS DLL beside FEMaster; other NVIDIA runtime dependencies remain external.

Running FEMaster

build/linux-release/bin/FEMaster model.inp --ncpus 8

.\build\windows-mkl\bin\Release\FEMaster.exe .\model.inp --ncpus 8

The input filename is positional. Results default to the same path with a .res suffix; --output selects another result path.

DSL documentation is available from the executable:

build/linux-release/bin/FEMaster --document --doc-list
build/linux-release/bin/FEMaster --document --doc-show LOADCASE
build/linux-release/bin/FEMaster --document --doc-search shell

Tests

Enable tests on any suitable preset, then build and run CTest:

cmake --preset linux-release -DFEMASTER_BUILD_TESTS=ON
cmake --build --preset linux-release
ctest --test-dir build/linux-release --output-on-failure

For Visual Studio builds, add -C Release to the CTest command.

License

FEMaster is licensed under the MIT License; see LICENSE.txt. Third-party components retain their own licenses; see THIRD_PARTY_NOTICES.txt and licenses/.