[SciPy-user] BLAS, LAPACK, ATLAS libraries.

[fie.pye at atlas.cz]

>---------------------------------------------------------
>Od: David Cournapeau
>Přijato: 18.7.2007 3:54:48
>Předmět: Re: [SciPy-user] BLAS, LAPACK, ATLAS libraries.
>
>fie.pye at atlas.cz wrote:
>
>> Hi David, hi Vincent,
>
>>
>
>> thank you for your response. 
>
>> On my computer I have three gcc compilers: gcc4.1, gcc4.2 and gcc3.4.6. Since you mentioned possible problem with gfortran I decided to compile BLAS, LAPACK, python2.5 and numpy again with gcc3.4.6 and f77.
>
>To be precise, what is problematic is mixing fortran compilers (for 
>
>example using gfortran for BLAS and g77 for LAPACK). gfortran by itself 
>
>is not that problematic.
>
>>  
>
>> Traceback (most recent call last):
>
>>   File "<stdin>", line 1, in <module>
>
>>   File "/usr/local/python/python2.5.1/lib/python2.5/site-packages/numpy/__init__.py", line 43, in <module>
>
>>     import linalg
>
>>   File "/usr/local/python/python2.5.1/lib/python2.5/site-packages/numpy/linalg/__init__.py", line 4, in <module>
>
>>     from linalg import *
>
>>   File "/usr/local/python/python2.5.1/lib/python2.5/site-packages/numpy/linalg/linalg.py", line 25, in <module>
>
>>     from numpy.linalg import lapack_lite
>
>> ImportError: /usr/local/lib64/lapack/liblapack.so: undefined symbol: slamch_
>
>The problem is the way you build the shared library is wrong: you cannot 
>
>just use g77 -shared -o soname *.o, because LAPACK object files are in 
>
>several directories.
>
>
>
>One correct way is to use the static library .a, uncompress it in a 
>
>directory, and build the shared library from that. There is a problem 
>
>when several object files have the same name (because you cannot have 
>
>several files with the same name in a directory), though, so extra care 
>
>must be taken.
>
>>
>
>> I included all installation scripts and log files to enclosed file. 
>
>>
>
>> I also began to install from source rpm but I am not familiar with that installation therefore I stopped after building. What does it mean refblas. 
>
>I cannot use blas and lapack from officiel repository because they do 
>
>not work very well, and I use a different name to avoid name clashes. 
>
>You should build the following srpms in this order:
>
>
>
>- refblas
>
>- lapack3
>
>- python-numpy
>
>- python-scipy
>
>
>
>Building them is not really different than any other source rpms. Since 
>
>you are using a distribution I've never tested, there is a good chance 
>
>it does not work out of the box, but I would rather help you solving 
>
>those problems so that everybody can benefit from them rather than 
>
>explaining how to build blas and lapack properly (which is really error 
>
>prone and non trivial).
>
>
>
>David
>
>_______________________________________________
>
>SciPy-user mailing list
>
>SciPy-user at scipy.org
>
>http://projects.scipy.org/mailman/listinfo/scipy-user
>
>
>
>
Hello David.

Thank you for responses. I managed to install numpy and scipy with BLAS and LAPACK libraries. Now I am going to install ATLAS and again python2.5.1, numpy and scipy. I will follow two ways. The first one is to build it from *.tar.gz sources. The second one is to build it from *.src.rpm as you recommended.  I think that the closer one *.src.rpm from http://software.opensuse.org/download/home:/ashigabou to CentOS 5.0 is Fedoea 6
http://software.opensuse.org/download/home:/ashigabou/Fedora_Extras_6/src/

The reason why I started with the *.tar.gz is that I have the full control over the version of libraries and places where the libraries are located. I can easily remove them when installation fails. In case of *.src.rmp I will have to study the installation and removing procedure first. I am going to start with rpmbuild. If you have some suggestions let me know.

If everything goes well I will install BLAS, LAPACK, ATLAS, python2.5.1, numpy and scipy also on HP workstation and server with AMD Opteron 2220 and 2218 with SUSE Linux Enterprise 10. 

Best regards.
Fie Pye