New Intel MKL Issue

There is a new issue with all releases concerning the Intel MKL.
https://github.com/devsim/devsim/issues/68

Newer versions of the Anaconda mkl package install a versioned dynamic library, which will not be found by devsim. If you encounter this issue with a new Anaconda environment, try installing an older version of the MKL using this command:

conda install mkl==2020.2 numpy

The next release will link against the newest version of the Intel MKL, whose name should hopefully not change very often.

Hi Juan,

I installed version 2.0.1 using the Linux binary on Ubuntu, but when I run a simulation libmkl_rt.so.2 cannot be found:

from .devsim_py3 import *
ImportError: libmkl_rt.so.2: cannot open shared object file: No such file or directory

I have anaconda version 4.11.0 but this contains only libmkl_rt.so.1

Probably I need do to an update, but I cannot find which.

Thanks for the help,
Regards,

Johan

If you create a new conda environement and install mkl, that should work.

conda create  -y --name my_test_environment python=3 mkl
conda activate my_test_environment
find $CONDA_PREFIX -iname 'libmkl_rt.so*'

There are not any new software features in version 2.0.1, so you can continue to use version 2.0.0 with libmkl_rt.so.1

please check CHANGES.md for notes. This version was to have a release with all of the documentation file changes for the JOSS publication.

Hopefully, Intel will not do new versioned releases very often. The last time I checked, numpy does not utilize the newer version, and instead resolves to the openblas.

>>> import numpy
>>> print(numpy.show_config())
blas_mkl_info:
  NOT AVAILABLE
blis_info:
  NOT AVAILABLE
openblas_info:
    libraries = ['openblas', 'openblas']
    library_dirs = ['/Users/jsanchez/anaconda/envs/test99/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None)]

while numpy is not required to use devsim, it demonstrates that not all of the Anaconda packages have been updated to use the latest version of the MKL.

Hopefully in a future release we can make devsim more resilient in terms of what versions of the MKL are available.

Thanks a lot for your quick reply.
I will continue to use 2.0.0

DEVSIM version 2.1.0 will attempt to use the version of Intel MKL available in your Anaconda environment.