In the unzipped directory, build and install openblas by

make BINARY=64 FC=gfortran USE_THREAD=4
make PREFIX=$HOME/openblas install

Note that this will install libopenblas in the openblas directory at $HOME, where you are surely permitted to write and the data will not be flushed.

2. Compile numpy with OpenBlas

Download numpy source code from the GitHub repo, say https://github.com/numpy/numpy/tree/v1.21.4. Unzip it.

Uninstall numpy originally installed by pip:

pip uninstall numpy

Make a good site.cfg configuration file by renaming site.cfg.example in the directory to site.cfg and edit the file by adding the following lines:

[DEFAULT]
library_dirs = /path/to/openblas/lib
include_dirs = /path/to/openblas/include

[atlas]
atlas_libs = openblas
libraries = openblas

[openblas]
libraries = openblas
library_dirs = /path/to/openblas/lib
include_dirs = /path/to/openblas/include

If you followed the commands above, /path/to/openblas should be equal to what you get by interpreting $HOME/openblas.

Build and install numpy with:

python setup.py build --fcompiler=gnu95
python setup.py install

3. Compile Minkowski Engine

Note: this should be done with a CUDA environment. You may want to switch to a GPU node.

Setup environments:

export CPLUS_INCLUDE_PATH=$C_INCLUDE_PATH:"$HOME/openblas/include"
export C_INCLUDE_PATH=$C_INCLUDE_PATH:"$HOME/openblas/include"
export LD_LIBRARY_PATH="$HOME/openblas/lib":$LD_LIBRARY_PATH
export LIBRARY_PATH="$HOME/openblas/lib":$LIBRARY_PATH

Build: pip install -U MinkowskiEngine --install-option="--blas=openblas" --install-option="--blas_library_dirs=/path/to/openblas/lib" -v --no-deps

python setup.py install --blas=openblas --blas_library_dirs=/path/to/openblas/lib

/path/to/openblas should be specified.