Alpine Docker container with NumPy and OpenBLAS

In order to support OpenBLAS in NumPy, one need to manually build it from source using site.cfg config.

Container breakdown

• 30.5Mb Official Python Alpine-based image.
• 10.8Mb OpenBlas /opt/OpenBLAS dependency.
• 58.0Mb = 32.9Mb NumPy + 21.9Mb gfortran + misc.

Notes

• Use parallel build python setup.py build --parallel=$(nproc), it saves ~70% of time.
• In order to pass C flags to python use CFLAGS="-Ofast". It overrides original settings, see documentation.
• NumPy is installed to /usr/local/lib/python3.6/site-packages/.

Links

• What is the relation between BLAS, LAPACK and ATLAS @StackOverflow.
• Numerical Linear Algebra Packages on Linux: BLAS, LAPACK, ATLAS.
• BLAS, LAPACK, CLAPACK, LAPACKE, ATLAS, OpenBLAS tutorials in C.