Skip to content

Latest commit

 

History

History

python

README.md

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)

For MacOS ARM architectures, currently only magnetostatics with the gfortran compiler is supported.

Create an importable Python module from Fortran source code

Linux

  • New conda environment with Python >= 3.12

    conda create -y -n magtense-env 
conda activate magtense-env
conda config --env --add channels conda-forge
conda config --env --add channels nvidia/label/cuda-12.6.3
conda config --env --add channels https://software.repos.intel.com/python/conda/
conda install -y python
conda install -y numpy matplotlib meson charset-normalizer ncurses git notebook h5py tqdm

  • Required python packages for CUDA and MKL

    Available CUDA versions can be found here: https://anaconda.org/nvidia/cuda
    Note: Use nvcc --version or nvidia-smi to detect the correct CUDA version for your system.

    conda install -y cuda-nvcc libcusparse-dev libcublas-dev cuda-cudart-dev libnvjitlink-dev

    More information about the Intel Compilers: Intel® C++ Compiler and Intel® Fortran Compiler

    conda install -y mkl mkl-devel mkl-static "dpcpp_linux-64" intel-fortran-rt "ifx_linux-64"

  • Compile Fortran source files

    cd python/src/magtense/lib/
make

Windows

  • Conda environment from environment.yml

    conda env create -f python/environment.yml

  • Compile Fortran source files

    • Installation of Visual Studio 2022 / Desktop development with C++

    • Setting up customized versions of Developer PowerShell and x64 Native Tools Command Prompt for VS 2022:

      • In VS Code, these integrated terminals can be added to your profiles by editing settings.json. Further, the Python extension ensures that the correct conda environment is activated in all terminals.

        "terminal.integrated.profiles.windows": {
    "Developer PowerShell for VS 2022": {
        "source": "PowerShell",
        "icon": "terminal-powershell",
        "args": [
        "-NoExit",
        "-ExecutionPolicy",
        "ByPass",
        "-File",
        "C:/Program Files/Microsoft Visual Studio/2022/Community/Common7/Tools/Launch-VsDevShell.ps1"
        ]
    },
    "DevCmdx64": {
        "path": [
            "${env:windir}\\Sysnative\\cmd.exe",
            "${env:windir}\\System32\\cmd.exe"
        ],
        "args": [
            "/k",
            "C:/Program Files/Microsoft Visual Studio/2022/Community/Common7/Tools/VsDevCmd.bat",
            "-startdir=None",
            "-arch=amd64",
            "-host_arch=x64"
        ],
        "icon": "terminal-cmd"
    },
},

      • Alternatively, customized profiles can be set up in the Windows Terminal app with the following guide: https://learn.microsoft.com/en-us/windows/terminal/install#settings-json-file

        Further, conda has to be initialized in these terminals to have access to the make executable and other necessary packages.

    • Open a Developer PowerShell and run:

      conda activate magtense-env
cd python/src/magtense/lib/
make ps

    • Compilation with nvcc should be executed in x64 Native Tools Command Prompt for VS 2022. Otherwise, x86 will be silently used, which results in error: asm operand type size(8) does not match type/size implied by constraint 'r' in cuda_bf16.hpp.

      cd source/MagTenseFortranCuda/cuda
make

      • Note: In case the error nvcc fatal : Could not set up the environment for Microsoft Visual Studio [...] shows up, the environment path in the active conda environment prevents nvcc to work correctly. A quick fix to compile MagTenseCudaBlas is to initialize a x64 Native Tools Command Prompt for VS 2022 without conda:

        "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin\nvcc.exe" -c MagTenseCudaBlas.cu -o MagTenseCudaBlas.o

        And then only compile MagTenseCudaBlasICLWrapper.cxx before creating libCuda in the activated environment:

        cd source/MagTenseFortranCuda/cuda
make wrap

    • Linking and wrapping libraries with f2py needs to be run in x64 Native Tools Command Prompt for VS 2022 to make ifx compiler available for meson:

      conda activate magtense-env
cd python/src/magtense/lib/
make cmdx64

      • Note: In case error meson.build:1:0: ERROR: Unknown compiler(s): [['ifx']] shows up, it should help to reinitialize your conda environment to ensure having the correct environment path:

        conda deactivate
conda activate magtense-env

Install local editable magtense package

cd python/
python -m pip install -e .

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/*

Distribution on Anaconda

Required compilers have to be pre-installed

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-12.6.3
conda config --env --append channels https://software.repos.intel.com/python/conda/
conda config --env --append channels conda-forge

# Quick fix for now
# Copy pre-compiled Python extension to MagTense/python/src/magtense/lib
# Build conda seperately for each version

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

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.