[SciPy-user] Question about ATLAS usage in SciPy

[Fernando Perez]

Hi all,

I have a question concerning the use of atlas static libraries. I built 
atlas on my system as per their standard instructions, resulting in a set 
of static libraries (this is on a RedHat 8.0 machine). I put these in 
/usr/local/lib, and ran ldconfig.

When I start the scipy build, it correctly seems to detect atlas:

root[scipy]# python setup.py build
atlas_info:
  FOUND:
    libraries = ['lapack', 'f77blas', 'cblas', 'atlas']
    library_dirs = ['/usr/local/lib']
    include_dirs = ['/usr/local/include']

/usr/bin/python 
/usr/local/installers/python/scipy/scipy/linalg/setup_atlas_version.py 
build_ext --inplace --force
/usr/bin/python -c "import atlas_version"
ATLAS version 3.4.1


However, my system _also_ has dynamical versions of blas and lapack 
provided by redhat in /usr/lib:

root[lib]# ls /usr/lib/*blas*so*
/usr/lib/libblas.so@    /usr/lib/libblas.so.3.0@    
/usr/lib/libgslcblas.so.0@
/usr/lib/libblas.so.3@  /usr/lib/libblas.so.3.0.3*  
/usr/lib/libgslcblas.so.0.0.0*
root[lib]# ls /usr/lib/*lapack*so*
/usr/lib/liblapack.so@    /usr/lib/liblapack.so.3.0@
/usr/lib/liblapack.so.3@  /usr/lib/liblapack.so.3.0.3*
root[lib]#


My question is: is the scipy build process smart enough to link the static 
atlas libraries, or will the default behavior of the compiler (dynamic 
linking) slip through? After going through the work of building/installing 
atlas, I don't want to end up with scipy using the stock unoptimized 
blas/lapack for everything.

This question also extends to f2py: if I wrap a simple fortran code with 
f2py on this system, will the linking pick up the atlas versions or will 
the resulting .so module end up using the dynamic (unoptimized) versions 
provided by redhat?

Thanks for any help.

Cheers,

f