MKL FATAL ERROR: Cannot load libmkl_avx.so or libmkl_def.so
It seems like boost.python + numpy triggers an MKL bug:
$ python2 dist/Basis.py
('Created PointsKey with', 10, 'points')
Points locations:
[-1. -0.91953391 -0.73877387 -0.47792495 -0.16527896 0.16527896
0.47792495 0.73877387 0.91953391 1. ]
('\nIntegral of sin(x) over [-1,1] =', 1.0408340855860843e-16)
MKL FATAL ERROR: Cannot load libmkl_avx.so or libmkl_def.so.
although:
$ echo $LD_LIBRARY_PATH | grep -i
mkl
[snip]:/opt/intel/13.0.1/mkl/lib/intel64:[snip]
and
$ ll /opt/intel/13.0.1/mkl/lib/
intel64/libmkl_avx.so 1 ↵
-r-xr-xr-x 1 root root 27M 3. Jul 2015 /opt/intel/13.0.1/mkl/lib/intel64/
libmkl_avx.so
according to stackoverflow, this this seems to be a bug in MKL: https://stackoverflow.com/questions/13951953/linking-intels-mkl-within-python-c-c-fortran-2003-program
the LD_PRELOAD workaround works for me. Until Intel fixes MKL we will probably have to live with this, so i am immediately closing this issue again, keeping it for others that might run into it.