[Numpy-discussion] Installation

Bernhard Voigt bernhard.voigt at gmail.com
Fri Jan 12 09:55:41 EST 2007

Dear all,

I've recently installed the last releases of numpy and scipy (tar
archives from scipy.org).
I encountered two problems:

1) During the configuration it seems the information from my site.cfg
is not read.
2) Installation using atlas optimized lapack/blas libraries didn't work

1) My site.cfg is in numpy-1.0.1/ (same as the setup.py file) and has
the following contents:

library_dirs = ~/apps/usr/local/lib/
blas_libs = blas

library_dirs= ~/apps/usr/local/lib
lapack_libs = lapack

The output of the python setup.py config lists this:

   libraries = ['blas']
   library_dirs = ['/afs/ifh.de/user/b/bvoigt/apps/lib']
   language = f77

   libraries = ['lapack']
   library_dirs = ['/afs/ifh.de/user/b/bvoigt/apps/lib']
   language = f77

Which acutally aren't the libs from the path I quoted in site.cfg. I've
put links to ~/apps/lib in order to get it run and this works.
~/apps/libs might be take from some python standard library path since
this is where my python lives.

2) In addition to that I had another problem using ATLAS optimized blas
and lapack libraries. Installation worked the same way that I had to
put a link in ~/apps/lib/atlas. However, I got an unresolved symbold
error that ATL_dgemm was not found, when I imported numpy.linalg. I
searched for that type of failure when using altas and I found that the
order of the libraries might be of importance when the numpy fortran
wrappers are compiled:

Taken from
my problem was the order of the invication of the librarys !

if i use

   g77 dgt.f libf77blas.a libatlas.a

i get no errormessage. but if i use

   g77 dgt.f libatlas.a libf77blas.a

In this case quoted here, the compiler complained already about
unresolved symbols, so I don't know if this is the same reason for the
problem I observer.

Anyway, I sorted the list of libraries in the
numpy/distutils/system_info.py file in the method at the bottom. This
had an effect, however another symbol couldn't be resolved :-( and I
stopped playing around.

Thanks for any suggestions/help solving those problems.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20070112/ea82d447/attachment.html>

More information about the NumPy-Discussion mailing list