Update MKL to 2021.1.1

[ViralBShah]

No description provided.

[giordano]

Changes look good, but it seems they once again changed the files structure in the Window package...

[ViralBShah]

Looks like the windows now has the same paths for the licenses.

[ViralBShah]

What's the reason to make MKL a lazy artifact? Shouldn't we always install it when required?

[giordano]

Oh yeah, like it was last year: 6c03359. They seem to like moving files back and forth

[giordano]

What's the reason to make MKL a lazy artifact? Shouldn't we always install it when required?

MKL is proprietary, it's large, it's a dependency of FFTW.jl and it isn't even used by default: JuliaMath/FFTW.jl#133

[ViralBShah]

Ah it is an optional dependency there. Makes sense. My thought was that in a package that explicitly depends on MKL, it felt odd to see the download at runtime. For example, SuiteSparse upstream recommends against openblas, and I recently ran into a major performance issue in Circuitscape.jl, which had to be resolved by building SuiteSparse with MKL. That's why I have a renewed interest in MKL.

[ViralBShah]

I wonder why it can't find mkl_core. There is a file in the distribution called mkl-core.1.dll. Is it the .1 that is a problem?

[ Info: Found a valid dl path mkl_vml_mc3.1.dll while looking for libmkl_core, mkl_core
[ Info: Could not locate libmkl_core, mkl_core inside ["/agent/_work/1/s/M/MKL/build/x86_64-w64-mingw32/jIJ1g1rr/destdir/bin"]
┌ Error: Built MKL but libmkl_core still unsatisfied:
└ @ BinaryBuilder /depot/packages/BinaryBuilder/57NRU/src/AutoBuild.jl:761
##[error]Built MKL but libmkl_core still unsatisfied:

[giordano]

Yes, it should be called mkl-core.1:

julia> using Base.BinaryPlatforms

julia> parse_dl_name_version("mkl-core.1.dll", "windows")[1] 
"mkl-core.1"