[SciPy-User] Building SciPy on Debian with Intel compilers

Klonuo Umom klonuo@gmail....
Wed Aug 17 09:28:11 CDT 2011

I got new Ubuntu 11.04 PC, and actually I accidentally found older
blog post which helped me pass this old sparsetools problems. Blog is
here: http://marklodato.github.com/

In brief, solution was to first run custom: `python setup.py config`
and then link sparsetools with icpc by hand:

for x in csr csc coo bsr dia; do
    icpc -xHost -O3 -fPIC -shared \
        build/temp.linux-x86_64-2.6/scipy/sparse/sparsetools/${x}_wrap.o \
        -o build/lib.linux-x86_64-2.6/scipy/sparse/sparsetools/_${x}.so
icpc -xHost -O3 -fPIC -openmp -shared \
    build/temp.linux-x86_64-2.6/scipy/interpolate/src/_interpolate.o \
    -o build/lib.linux-x86_64-2.6/scipy/interpolate/_interpolate.so

Paths are dependant on Intel tools and Scipy version, and it's trivial
to correct them

So I build Numpy and Scipy with latest Intel Parallel Studio XE 2011
update 2 + SparseSuite (AMD and UMFPACK), but then running test I got

*** libmkl_p4p.so *** failed with error :
undefined symbol: i_malloc

*** libmkl_def.so *** failed with error :
undefined symbol: i_malloc

MKL FATAL ERROR: Cannot load neither libmkl_p4p.so nor libmkl_def.so

Workaround is this:

export LD_PRELOAD=/opt/intel/mkl/lib/ia32/libmkl_core.so:/opt/intel/mkl/lib/ia32/libmkl_sequential.so

Now I run tests again

Numpy: FAILED (KNOWNFAIL=3, SKIP=4, failures=4)
more info: http://pastebin.com/raw.php?i=m3sns5xU

Scipy: FAILED (KNOWNFAIL=12, SKIP=35, errors=1, failures=3)
more info: http://pastebin.com/raw.php?i=tvqg8PJ1

I wish someone reply about this 'export LD_PRELOAD' workaround, and
also maybe correct online documentation about building Numpy/Scipy
with Intel compilers - at least with earlier David's corrections in
this thread about 'scipy/spatial/qhull/src/qhull_a.h', as if user does
not understand what are C++ templates, he/she could hardly figure what
to do. About sparsetools I'm happy I got it working, and this issue is
open as of Numpy 1.3.0 at least it seems


More information about the SciPy-User mailing list