Add dependency on MKL_jll.jl

[andreasnoack]

That would make the package work automatically and make it easy to enable CI. MKL.jl overwrites OpenBLAS which currently can cause some issues when using other binary dependencies which require BLAS, e.g. Arpack so we might want to wait a bit. MKL.jl also causes a regression in REPL compile latency but hopefully, I'll be able to fix that soon.

We could probably add MKL.jl as a test dependency here as soon as it has been registered.

[andreasnoack]

Well. We probably don't need to MKL.jl to be registered to use it for testing.

[KristofferC]

But this package already works automatically (as long as you set the env variables with the mklvars script).

[andreasnoack]

Yeah. I wasn't precise. I meant without a preinstalled MKL.

[arnavs]

Just a question on this issue.

An RA here is working on fleshing out coverage in MKLSparse. The MKL downloads for Mac on Intel's site are broken, so setting the environment variables is impossible. Instead, we decided to install MKL.jl, and then nuke the piece of code that manually links to it (i.e., the __init__().)

Is this correct? (Basically, if you added MKL.jl as a dep, would that init function be irrelevant?)

[KristofferC]

Update here: What should be done is to depend on MKL_jll and use that (no need to use MKL.jl).

[ywa136]

Hi guys, a question.

I'm new to Julia, MKL, and MKLSparse. I'm having trouble getting MKLSparse to work because it could not load the library "libmkl_rt". I understand that you mentioned to depend the package on MKL or MKL_jll. But how that exactly works? More specifically, how do I depend MKLSparse to one of those?

Many thanks!

[pablosanjose]

@ywa136 , what you need to do is install the MKL library (from Intel) in your system, and load the env variables before launching Julia (source /opt/intel/mkl/bin/mklvars.sh intel64 ilp64 or somesuch). What these gentlemen are discussing here is not anything the user needs to do, it is a mechanism to allow, in the future, to download the library automatically so the user doesn't have to. Also, please direct your questions to e.g. discourse.julialang.org, so we can keep github focused on development issues.