[SciPy-user] numpy/scipy+ATLAS on SuSE10.0/gcc4 (once more)

Nils Wagner nwagner at mecha.uni-stuttgart.de
Tue Mar 7 01:09:36 CST 2006


Christian Kristukat wrote:
> Hi,
> I've got some observations to share concerning the compilation of numpy/scipy
> with ATLAS on SuSE10.0
> In some earlier posts I told everybody to use gfortran to build
> numpy/scipy/ATLAS. Although the compilation worked without errors and
> numpy/scipy seemed to work as well, Robert prooved, that gfortran is causing
> serious faults so that scipy eventually fails during execution.
> Originally I have chosen to use gfortran because every attempt to build
> numpy/scipy/ATLAS on SuSE10.0 using gcc4 and g77 failed. I will shortly describe
> what I tried (with gcc4/g77)
>
> 1) using SUSE10.0 blas and lapack rpms
>
> Python 2.4.1 (#1, Sep 13 2005, 00:39:20)
> [GCC 4.0.2 20050901 (prerelease) (SUSE Linux)] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
>   
>>>> import numpy
>>>>         
> import linalg -> failed: /usr/lib/libblas.so.3: undefined symbol: _gfortran_filename
>   
>
> 2) with ATLAS
>
> Python 2.4.1 (#1, Sep 13 2005, 00:39:20)
> [GCC 4.0.2 20050901 (prerelease) (SUSE Linux)] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
>   
>>>> import numpy
>>>>         
> import linalg -> failed:
> /usr/lib/python2.4/site-packages/numpy/linalg/lapack_lite.so: undefined symbol:
> _gfortran_copy_string
>   
>
> 3) using self compiled lapack and blas works!!
>
> Can anyone figure out what the problem is and propose what to try next?
>
> Regards, Christian
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user
>   
Hi Christian,

It is well known that the rpm's (lapack + blas) shipped with SuSE are 
incomplete.
So I recommend to remove these rpm's and comile lapack/blas from scratch.
BTW, I have send a message to SuSE two years ago and didn't receive any 
reply.

Nils




More information about the SciPy-user mailing list