[Numpy-discussion] Numpy matrix multiplication slow even though ATLAS linked

[Jan-Willem van de Meent]

I resolved the issue after having another look at the build scripts. Thanks 
everyone for your help!

For reference, the problem arose from a bad site.cfg file. In the past this 
file was necessary on ArchLinux to make numpy find the atlas libs. I tried 
removing it and now numpy to compiles correctly!

(offending) site.cfg
--------------------
[atlas]
library_dirs = /usr/lib
atlas_libs   = atlas, lapack, cblas, f77blas, ptcblas, ptf77blas

For reference #2: During compilation of the improperly linked numpy, the 
script displays the info:

FOUND:
  libraries = ['atlas', 'lapack', 'cblas', 'f77blas', 'ptcblas', 'ptf77blas']
  library_dirs = ['/usr/lib']
  language = c
  define_macros = [('NO_ATLAS_INFO', 2)]
  include_dirs = ['/usr/include']

Whereas a good compile shows:

  FOUND:
    libraries = ['ptf77blas', 'ptcblas', 'atlas']
    library_dirs = ['/usr/lib']
    language = c
    define_macros = [('ATLAS_INFO', '"\\"3.8.2\\""')]
    include_dirs = ['/usr/include']

Cheers! Now I can get back to work :-)

JW