[SciPy-User] Problems compiling NumPy on linux with a compiled lapack/blas

David Cournapeau cournape@gmail....
Tue Jun 19 22:57:35 CDT 2012


Le 20 juin 2012 04:36, "Russell Warren" <russ@perspexis.com> a écrit :
>
> I'm trying to compile numpy from source with both lapack and BLAS
compiled from source as well.  I've got full details below, but my
compilation method is based mostly on guidelines from these links:
> http://www.scipy.org/Installing_SciPy/BuildingGeneral
> http://www.scipy.org/Installing_SciPy/Linux
>
> I'm trying to compile in a portable way, mostly with the use
of -march=i486 -mtune=i686, since I'm trying to make this as a portable
extension for a minimal linux install based on tinycore linux.  I don't
expect/think that this is particularly relevant, but one outcome of it is
that I'm doing everything from source... ie: no debian apt packages or
other conveniences available.  It also means I'm not using ATLAS, since I
think it tunes to the local cpu by definition.
>
> I can compile numpy with no problems at all if I don't set the LAPACK
environment variable, and numpy.test("full") ends up working perfectly fine
with no errors.
>
> However, when I set the LAPACK environment variable to the path of my
generated libflapack.a file, the numpy compilation fails.  It is failing at
a strange location (when gfortran is trying to generate lapack_lite.so) and
in a strange way (it looks like the python header files are missing).
>
> Full details:
> gfortran 4.6.1
> gcc 4.6.1
> os.name: posix
> sys.platform: linux2
> sys.version: '2.7.3 (default, Jun 19 2012, 16:53:07) \n[GCC 4.6.1]'
> numpy.version: 1.6.2
> uname -a: Linux edms 3.0.21-tinycore #3021 SMP Sat Feb 18 11:54:11 EET
2012 i686 GNU/Linux
> ATLAS version:  NOT USING ATLAS...  is this a problem??
>
> I'm compiling BLAS, LAPACK, and NumPy using this command chain:
> http://static.inky.ws/syn/502
>
> And my failed setup.py output log is here:
> http://static.inky.ws/text/503
>
> What is happening here?  Do I need to use ATLAS?  Any help on getting
numpy working with lapack is greatly appreciated!

Your log appears incomplete. You definitely don't have to use atlas to
build numpy/scipy.

David
>
> _______________________________________________
> SciPy-User mailing list
> SciPy-User@scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/scipy-user/attachments/20120620/dcf9763b/attachment.html 


More information about the SciPy-User mailing list