[SciPy-user] Illegal instruction in testsuite of revision 3378

David Cournapeau david@ar.media.kyoto-u.ac...
Fri Sep 28 23:53:14 CDT 2007


Xavier Gnata wrote:
> Robert Kern wrote:
>   
>> Xavier Gnata wrote:
>>   
>>     
>>> ok.
>>> I'm compiling scipy with gcc version 3.4.6 (Debian 3.4.6-6) (g77 -v).
>>> It is quite strange beacause I do have gfortran installed but I get this :
>>> Found executable /usr/bin/g77
>>> gnu: no Fortran 90 compiler found
>>> gnu: no Fortran 90 compiler found
>>> customize GnuFCompiler
>>> gnu: no Fortran 90 compiler found
>>> gnu: no Fortran 90 compiler found
>>>
>>> Is gfortran  a Fortran 90 compiler or not????
>>>     
>>>       
>> gfortran is. g77 is not. Use --fcompiler=gnu95 to tell the setup.py script to
>> use gfortran if you want it.
>>
>>   
>>     
>>> Ok : lapack, atlas and blas are compiled using g77 (3.4.6) so it could 
>>> be part of problem.
>>> Can we compile lapack, atlas and blas with gfortran 4.2 ??
>>>     
>>>       
>> I believe so.
>>
>>   
>>     
> I still have the same problem using
>
> python setup.py build --fcompiler=gnu95 
> python setup.py install
>
> --fcompiler=gnu95  works well. gfortan has been used but the bug is 
> still present.
>   
It is a bad idea to use gfortran on sid, because the default fortran 
compiler is g77, and g77 and gfortran have different ABI, making 
libraries compiled by them difficult to coexist. My advice would be 
either stick with g77, or compile everything using gfortran (eg blas, 
lapack, atlas and numpy/scipy).

cheers,

David



More information about the SciPy-user mailing list