Updating MKL without rebuilding application

[mvrht__u3425923]

Hello,

I am using NumPy/SciPy manually built with MKL as BLAS. I am wondering is it possible to update from MKL 2018 update 3 to MKL 2019 without rebuilding NumPy/SciPy?

Thanks in advance for providing information!

[mecej4 · Accepted solution]

In what follows, I assume that you are building on Windows. If not, replace "DLL" by "shared"/"shared library".

If you linked your NumPy with the static MKL libraries, updating MKL will have no effect on NumPy. If you want NumPy to use the more recent BLAS in MKL 2019, you will have to rebuild NumPy. If you don't care, or there is no announced improvement in the BLAS, your NumPy will continue to work as it did before you updated MKL.

If you linked NumPy using the DLL MKL libraries, you do not have to rebuild NumPy at all. However, if you wish NumPy to use the MKL 2019 DLLs, you have to replace the path to the MKL 2018 DLLs with the path to the MKL 2019 DLLs, or make the path to the new DLLs appear first.

[mvrht__u3425923]

Hi, mecej4,

sorry for not providing full information. Yes, it is windows build linked with DLLs and there will be no problem with path as I am going to uninstall old version. Thank you for answering!

[Gennady_F_Intel]

btw, in the case if you will use IDP package ( Intel Distribute Package for Python) then all latest version of MKL would be available. NumPy and SciPy from IDP are already optimized by Intel MKL and DAAL as well.