[SciPy-User] building scipy with
Agile
agile.aspect at gmail.com
Sat Aug 8 21:46:41 EDT 2009
Hi - I've built Python 2.5.4 with gcc 4.1.2 and with numpy 1.3.0 on 32
bit machine running
CentOS 5.3 using the following instructions for ATLAS officially
sanctioned by SciPy:
http://www.scipy.org/Installing_SciPy/Linux#head-40c26a5b93b9afc7e3241e1d7fd84fe9326402e7
That is, I'm using the DavidCournapeau method. This is the method
where one ends up having just
2 libraries installed in /usr/lib/atlas/sse2, namely
libblas.so.3.0
liblapack.so.3
Numpy works fine, and using Nose, it was tested with 2030 tests and
finished with the following
message
[OK] (KNOWNFAIL=1)
I'm not really sure what this means, but it suggests that 1 out of
2030 tests failed - I'm going
to ignore the failure for now (since there was no clue as to which
tested failed anyway.) Hence
I can safely say Numpy is working properly for at 2029 tests.
The problem is I can't even begin to build Scipy because it craps out
immediately claiming it
can't find the ATLAS libraries (and suggests setting BLAS_SRC and
LAPACK_SRC environment
variables for a non-optimized build.)
My question is, since it's recommend method of building ATLAS for
Scipy, what do I have to do
to make it work with Scipy?
Here's the current Numpy's site.cfg.
[DEFAULT]
library_dirs = /usr/devtools/lib:/usr/lib/atlas/sse2
include_dirs = /usr/devtools/include:/usr/include
[blas_opt]
#libraries = f77blas, cblas, atlas
libraries = blas
[lapack_opt]
#libraries = lapack, f77blas, cblas, atlas
libraries = lapack
[fftw]
libraries = fftw3
Any help would greatly appreciated.
-- Agile
More information about the SciPy-User
mailing list