Obtaining NumPy build configuration.

Contributed by Steve Hudson

NumPy is based on compiled C extensions and in turn is built upon BLAS/LAPACK libraries. The following gives some information on how to interrogate the build configuration and dependent libraries upon which NumPy is built.

Get basic configuration including BLAS/LAPACK libraries:

import numpy
numpy.__config__.show()

Note that configuration information may not be reliable in all circumstances. Dependencies of built libraries can also be checked by using the ldd command on NumPy shared libraries. E.g.

ldd /<path_to_site-packages>/numpy/core/multiarray.so

Libraries may have variations in name based on platform (e.g. multiarray.cpython-36m-x86_64-linux-gnu.so). To check the site packages directory location you can do:

$ python
Python 3.6.3 |Intel Corporation| (default, Oct 16 2017, 15:28:36) ....
>>> import numpy
>>> numpy.__path__
['/home/shudson/miniconda3/lib/python3.6/site-packages/numpy']

To get full NumPy configuration details including compiler options/environment:

import numpy.distutils
np_config_vars = numpy.distutils.unixccompiler.sysconfig.get_config_vars()

import pprint
pprint.pprint(np_config_vars)

Or to write output to a file

f = open('np_config_pprint.log','w')
pprint.pprint(np_config_vars,f)

To get specific variables you can specify them as arguments E.g:

import numpy.distutils
c_cpp_compiler = numpy.distutils.unixccompiler.sysconfig.get_config_vars('CC', 'CXX')
compile_vars = numpy.distutils.unixccompiler.sysconfig.get_config_vars('CFLAGS', 'LDSHARED')
print(c_cpp_compiler)   
print(compile_vars)

If NumPy has been installed without modification it will likely inherit the configuration from Python. You can check that with distutils.

from distutils import sysconfig
sysconfig.get_config_vars()

References

Python for High Performance Computing (Exascale Computing Project 2nd Annual Meeting) Slides William Scullin (ALCF), Matt Belhorn (OLCF), Rollin Thomas (NERSC)

stackoverflow: find-out-if-which-blas-library-is-used-by-numpy

Feedback

Any feedback/corrections are welcome: shudson@anl.gov

If this page was useful, please leave a star on the github page.

Back to main page