MKL_jll does not function properly on Mac in Julia 1.4

[KristofferC]

Based on issue JuliaMath/IntelVectorMath.jl#46, I could repro the hang on Mac with Julia 1.4 using for example:

using MKL_jll
gamma1(x) = (x = copy(x); ccall((:vdTGamma, libmkl_rt), Cvoid, (Int32, Ptr{Float64}, Ptr{Float64}), length(x), x, x); x);
gamma1([1.0, 2.0])

Note that this works in Julia 1.3, Julia 1.4-RC2 and locally built Julia. When I used my local installation of MKL it works:

julia> ENV["MKLROOT"]
"/opt/intel/compilers_and_libraries_2020.0.166/mac/mkl"
julia> const libmkl_rt = "@rpath/libmkl_rt.dylib";
julia> gamma1(x) = (x = copy(x); ccall((:vdTGamma, libmkl_rt), Cvoid, (Int32, Ptr{Float64}, Ptr{Float64}), length(x), x, x); x);
julia> gamma1([1.0, 2.0]) # works
2-element Array{Float64,1}:
 1.0
 1.0

I then looked at what the difference was between the MKL_jll provided MKL and my local installation. One file had a different hash: libmkl_intel_thread.dylib. Copying the one over from my local MKL to the MKL_jll artifact library made the code start working.

So my conclusion is that the libmkl_intel_thread.dylib that is provided by MKL_jll is somehow faulty.

[KristofferC]

Alternatively, the things creating these logs

 update_linkage_libmkl_intel_thread.dylib_libiomp5.dylib.log
 update_rpath_libmkl_intel_thread.dylib_libiomp5.dylib.log

messes something up.