Building from source

This page is for licensees who have the source distribution. If you installed a binary package, Get started is the page you want.

Platform policy

Linux with Intel oneAPI is the supported platform, and the only one whose results we stand behind.

• Linux — the release platform and the only one we validate scientific results on. Intel oneAPI: ifx/icx, oneMKL, Intel MPI.
• Windows — use WSL and follow the Linux route. A native Windows preset exists as a development convenience and is not a supported configuration; not all of its I/O paths are verified.
• macOS — development only and CPU only. Useful for editing, examples and smoke checks. Do not use Intel compilers with oneMKL on macOS.

Toolchains

Platform	Compiler	Math backend	FFT backend	MPI
Linux	Intel ifx / icx / icpx	oneMKL	MKL FFT	Intel MPI or a compatible MPI
Windows (development)	Intel ifx / icx	oneMKL	MKL FFT	Intel MPI
macOS (development)	Homebrew gfortran with AppleClang for C/C++	OpenBLAS	user-installed FFTW	Homebrew OpenMPI

Caution

FFTW is never bundled into an SDK package. On macOS it is a developer-installed dependency and making the dynamic library available at run time is the developer’s responsibility. On Linux the FFTW-style MPI interface is provided by oneMKL, not by FFTW itself.

Recent oneAPI releases dropped `ifort`

Every preset uses ifx. ifort is no longer published in oneAPI 2025 and later. Note also that setvars.sh selects the latest directory for every component and exits successfully whether or not that component actually contains a Fortran compiler — check that ifx resolves before concluding the environment is good.

Configure and build

One configure step from the repository root builds everything: the licence component, the library, and the SDK examples.

Linux

source /opt/intel/oneapi/setvars.sh

cmake --fresh --preset linux-intel-debug
cmake --build --preset linux-intel-debug
ctest --preset linux-intel-debug --output-on-failure
cmake --install out/build/linux-intel-debug
cmake --build --preset linux-intel-debug --target package

macOS

brew install cmake ninja gcc openblas fftw open-mpi

cmake --fresh --preset macos-arm64-debug
cmake --build --preset macos-arm64-debug
ctest --preset macos-arm64-debug --output-on-failure
cmake --install out/build/macos-arm64-debug

Release presets (linux-intel-release, macos-arm64-release) take the same four commands.

Build options

Option	Default	Effect
MUPRO_BUILD_TDGL	ON	Build the ferroelectric TDGL solver.
MUPRO_BUILD_MAGNETIC	ON	Build the magnetic and micromagnetic solvers.
MUPRO_ENABLE_HDF5	OFF	Write fields as HDF5 in addition to the native format. Needs an HDF5 build first, and adds six test cases.

To compile-check one solver without building the rest, configure into a separate directory and build just the library target:

cmake --fresh --preset linux-intel-debug -B out/build/linux-intel-debug-tdgl \
  -DMUPRO_BUILD_TDGL=ON
cmake --build out/build/linux-intel-debug-tdgl --target mupro

Building an example against an installed SDK

examples/demo1 is a downstream project: it finds the SDK with find_package(mupro) rather than being part of the build. After installing:

cmake -S examples/demo1 -B examples/demo1/build -G Ninja \
  -DCMAKE_BUILD_TYPE=Debug \
  -DCMAKE_PREFIX_PATH="$PWD/out/install/linux-intel-debug"

cmake --build examples/demo1/build

The example enables C, C++ and Fortran, because the SDK exports MPI C dependencies and the static library contains C++ object code.

The C ABI example builds the same way:

cmake -S examples/c_api_tdgl -B examples/c_api_tdgl/build -G Ninja \
  -DCMAKE_BUILD_TYPE=Debug \
  -DCMAKE_PREFIX_PATH="$PWD/out/install/linux-intel-debug"

cmake --build examples/c_api_tdgl/build
./examples/c_api_tdgl/build/c_api_tdgl

It prints a structured MUPRO_STATUS_NOT_SETUP message: the C ABI is a skeleton, and the TDGL solve backend is not connected to it yet. That message is the expected output, not a failure.

Tests

ctest --preset linux-intel-debug --output-on-failure

The default configuration registers 127 test cases, and a Linux debug build passes all of them. Enabling MUPRO_ENABLE_HDF5 adds six more.

A test counts as passing only if it exits zero and prints its own completion marker. A program that stops early, traps a floating-point exception, or fails an I/O operation therefore fails the test even when its exit status looks clean.

Release builds do not go fully green

In a release configuration the licence gate is active, so the solver tests that run without a licence fail by design, and the floating-point-trap self-check fails because trapping is only enabled in debug builds. Use a debug build as the test gate.