[Numpy-discussion] Python 2.6, NumPy on CentOS 5.3
rkdelisle at gmail.com
Wed Nov 25 13:12:15 EST 2009
It does indeed work now. I also was able to find a repo package with the
atlas libraries, so I installed them as well. It appears that everything
Thank you again for your assistance.
On Tue, Nov 24, 2009 at 8:02 PM, David Cournapeau <cournape at gmail.com>wrote:
> On Wed, Nov 25, 2009 at 2:11 AM, Robert DeLisle <rkdelisle at gmail.com>
> > David Cournapeau wrote:
> > You need the *.so and not the *.so.3.1.1. The latter are enough to *run*
> > applications linked against the library, the former is necessary to link
> > against them. IOW, you need the devel packages for blas, lapack (and
> > python). If you want to do it without admin rights, there is no other
> > solution than building them by yourself.
> > David,
> > Thank you so much! That helps a lot and I don't know why I didn't think
> > it myself. As root I'm able to install blas-devel and lapack-devel which
> > gives me new libraries in /usr/lib64. Specifically, libblas.a,
> > and liblapack_pic.a. Now when I run setup.py from the numpy directory,
> > says that it finds the libraries in the appropriate locations.
> > A few more issues:
> > The configuration complains that there are no mkl, vml, guide libraries.
> > Also, atlas libraries are not found. I cannot seem to find these at this
> > point.
> That's ok. Our build process is too verbose, and is a bit confusing.
> MKL and Atlas are just optimized BLAS/LAPACK, but they are not
> necessary to build scipy. They can significantly improve scipy speed
> for linear algebra stuff, but building Atlas or linking against the
> Intel MKL is complicated.
> NumPy-Discussion mailing list
> NumPy-Discussion at scipy.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the NumPy-Discussion