PyNFFT - Pythonic bindings around the NFFT library (M1/M2/M3/M4 Mac - MacOS16.0)

"The NFFT is a C subroutine library for computing the nonequispaced discrete Fourier transform (NDFT) in one or more dimensions, of arbitrary input size, and of complex data."

The NFFT library is licensed under GPLv2.

This wrapper provides a somewhat Pythonic access to some of the core NFFT library functionalities and is largely inspired from the pyFFTW project developped by Henry Gomersall (http://hgomersall.github.io/pyFFTW/).

The documentation is hosted on pythonhosted, the source code is available on github and the Python package index page is here.

Installation for Apple Silicon Mac (macOS 12.0+, including macOS 16 Sequoia)

Step-by-step installation instructions

1. Install fftw

   1. You need gcc compiler for installing fftw with --enable-openmp. The gcc in Apple Silicon Mac (macOS > v12.0) actually points to clang

      Install gcc-12 with brew:

      brew install gcc@12
      

   2. Download FFTW 3.3.10 from fftw.org and extract the folder

   3. cd into the fftw folder and use this configuration:

      ./configure CC="gcc-12" --enable-shared --enable-threads --enable-openmp
      sudo make
      sudo make check
      sudo make install
      

   4. Check /usr/local/lib/ and /usr/local/include/ to see if fftw3 is installed.

2. Install NFFT

   1. You need autoconf automake libtool for this step. Install it with:

      brew install autoconf automake libtool
      

      Make sure you edit your .zshrc to add it to your PATH

   2. Download the 3.5.3 release of the NFFT git repository:

      https://github.com/NFFT/nfft/releases/download/3.5.3/nfft-3.5.3.tar.gz
      

   3. cd into the NFFT folder and use this configuration:

      ./bootstrap.sh
      ./configure CC="gcc-12" --enable-all --enable-openmp --enable-shared --with-fftw3-libdir=/usr/local/lib/ --with-fftw3-includedir=/usr/local/include/
      sudo make
      sudo make check
      sudo make install
      

   4. Check /usr/local/lib/ and /usr/local/include/ to see if nfft is installed.

3. Install pyNFFT

   1. Clone this repository:

      git clone https://github.com/rohandahale/pyNFFT.git
      

   2. Install ehtim, cython with pip to run this setup.py:

      pip install ehtim
      pip install Cython==0.29.32
      

   3. cd into pyNFFT folder, build and install pynfft:

      export MACOSX_DEPLOYMENT_TARGET=16.0
      python setup.py build_ext -I /usr/local/include/ -L /usr/local/lib/ -R /usr/local/lib/
      pip install . --no-build-isolation
      

   4. Important: Fix the duplicate RPATH issue that may occur with conda/micromamba environments.

      Find your Python environment path and run:

      # Replace YOUR_USERNAME and YOUR_ENV_NAME with your actual values
      for module in nfft solver util; do
          install_name_tool -delete_rpath /Users/YOUR_USERNAME/path/to/envs/YOUR_ENV_NAME/lib \
              pynfft/${module}.cpython-311-darwin.so
          install_name_tool -add_rpath /usr/local/lib \
              pynfft/${module}.cpython-311-darwin.so
      done
      

   5. Verify the installation:

      python
      >>> import pynfft
      >>> print(pynfft.__version__)
      

   6. Done.

Note for macOS 16 (Sequoia) users: The setup.py has been updated to handle the macOS 16.0 deployment target requirements. If you're using an older macOS version, you can adjust the minos value in the setup.py accordingly (e.g., 12.0, 13.0, etc.).

Usage

See the tutorial section of the documentation.

Recommended versions

• Python 3.11
• ehtim 1.2.10
• FFTW 3.3.10
• NFFT 3.5.3
• Cython 0.29.32

Contributing

See the CONTRIBUTING file.

License

The pyNFFT project is licensed under the GPLv3. See the bundled COPYING file for more details.