For some time now the Travis CI builds for probcomp/bayesliteprobcomp/cgpm, and probcomp/Venturecxx have all been failing with the following error.
Intel MKL FATAL ERROR: Cannot load libmkl_avx2.so or libmkl_def.so.
Some even started failing without any commits having been pushed up to master which would suggest that the failures are due to a shift in the dependencies being pushed by conda.
MKL, also known as the Intel® Math Kernel Library, is a proprietary math library for us on Intel hardware. Conda, our build tool, provides versions of our dependencies that include MKL and others that do not. Our (my and @fsaad's) hypothesis here is that some shift in the underlying build environment caused our packages to start being build with MKL whereas the test environment does not have MKL available.
This pull request adds a dependency on the nomkl package, which appears to be the recommended way of opting out of MKL support for numpy and other dependencies of ours.
For some time now the Travis CI builds for
probcomp/bayesliteprobcomp/cgpm, andprobcomp/Venturecxxhave all been failing with the following error.Some even started failing without any commits having been pushed up to
masterwhich would suggest that the failures are due to a shift in the dependencies being pushed byconda.MKL, also known as the Intel® Math Kernel Library, is a proprietary math library for us on Intel hardware. Conda, our build tool, provides versions of our dependencies that include MKL and others that do not. Our (my and @fsaad's) hypothesis here is that some shift in the underlying build environment caused our packages to start being build with MKL whereas the test environment does not have MKL available.
This pull request adds a dependency on the
nomklpackage, which appears to be the recommended way of opting out of MKL support fornumpyand other dependencies of ours.