python

Python Interface

The Fortran code is compiled and wrapped to a module that can be directly called from Python. The tool f2py of the NumPy package is used to wrap the interface file MagTense/python/src/magtense/lib/FortranToPythonIO.f90.

Deployment with Conda (Intel architectures)

Requirements

• Python >= 3.9

• Intel® C++ Compiler and Intel® Fortran Compiler

  • [Linux] Prepare your terminal, so that ifort compiler can be found:

    . /opt/intel/oneapi/setvars.sh

  • [Windows] Activation via VS Code extension for Intel® oneAPI Toolkits

• Required python packages

  Find available ${CUDA_LABEL} here. HINT: Use nvcc --version or nvidia-smi to detect the correct CUDA version for your system.

  conda install -y numpy matplotlib
  conda install -y -c "nvidia/label/cuda-${CUDA_LABEL}" cuda-nvcc libcusparse-dev libcublas-dev cuda-cudart-dev libnvjitlink-dev
  conda install -y -c intel mkl-static

  • [ Linux ]

    conda install -y -c intel intel-fortran-rt

  • [ Windows / MacOS ] Installation of Make utility

    conda install -y -c conda-forge make

• Additional python packages to run data creation scripts

  conda install -y h5py tqdm

Installation from source (including MacOS Intel & ARM)

Create an importable Python module from Fortran source code.

For MacOS ARM architectures, currently only magnetostatics with the gfortran compiler is supported. Navigate to folder MagTense/python/src/magtense/lib/, run make, and install the package:

cd MagTense/python/src/magtense/lib/
make
cd MagTense/python/
python -m pip install -e .

Read-in customized M-H-curve

This feature is currently only supported for soft magnetic tiles (type=2).

In iterate_magnetization(), an arbitrary number of state functions (M-H-curves) can be defined:

mu_r = 100
datapath = f'./magtense/mat/Fe_mur_{mu_r}_Ms_2_1.csv'

 ...

data_statefcn = numpy.genfromtxt(datapath, delimiter=';')
n_statefcn = 1

Here, three sample M-H-curves for Fe with different relative permeabilities and a saturation magnetization of 2.1 T are stored as CSV-files. The data format is as follows:

0; Temp0; Temp1; ...
H0-field; M0@Temp0; M0@Temp1;...
H1-field; M1@Temp0; M1@Temp1;...
.
.
H100-field; M100@Temp0; M100@Temp1; ...
.

With only one state function given, the same M-H-curve applies to all tiles of type 2.

When the soft tiles differ in their M-H-curves, multiple state function can be combined. In order to match a specific M-H-curve with the corresponding tile, the variable stfcn_index can be set.

Distribution on Anaconda

Required compilers have to be pre-installed

• Intel® C++ Compiler
• Intel® Fortran Compiler

conda install -y anaconda-client conda-build

# Add nvidia channel to find CUDA and intel libraries
# conda config --show channels
conda config --env --append channels nvidia/label/cuda-12.2.2
conda config --env --append channels intel
# On Windows
# conda config --env --append channels conda-forge

# Quick fix for error of nvcc during build on Windows
# conda install -y -c nvidia/label/cuda-12.2.2 cuda-nvcc
# cd MagTense/source/MagTenseFortranCuda/cuda/
# nvcc -c MagTenseCudaBlas.cu -o MagTenseCudaBlas.o

# Version numbers have to be set in advance in pyproject.toml
cd MagTense/python/
python scripts/dist_conda.py

Distribution on PyPI

Libraries have to be pre-build for now, and should be located in MagTense/python/compiled_libs.

# Required python packages for distribution
python -m pip install build
conda install -y twine

cd MagTense/python/
python scripts/dist_pypi.py

# Upload to pypi.org
# twine upload --repository testpypi dist/*
twine upload dist/*