Octave gives wrong results when diagonalizing large matrices with MKL

Octave gives wrong results when manipulating large matrices, when the package `intel-mkl` is installed, and BLAS, LAPACK are configured with `intel-mkl` (from default ubuntu repositories). With MKL, Octave seems to have a 2x speedup when diagonalize certain matrices, but the results are totally wrong (when the matrix is large, say 1000x1000).

Code:
```
for a = 1:500
 for b = 1:500
  c(a,b) = sin(a + b^2);
 endfor
endfor

g = eig(c);
max(real(g))
```
The correct result is `ans = 16.915` (I checked this before installing intel-mkl, and also in Ubuntu 18.04 in the same computer, where intel-mkl was not installed)
With MKL in Ubuntu 20.04, I get random numbers of order 10^5 - 10^6, which change every time the code is run.

lsb_release -rd
Description: Ubuntu 20.04 LTS
Release: 20.04
(It is Xubuntu 20.04)

apt-cache policy octave
octave:
  Installed: 5.2.0-1
  Candidate: 5.2.0-1
  Version table:
 *** 5.2.0-1 500
        500 http://archive.ubuntu.com/ubuntu focal/universe amd64 Packages
        100 /var/lib/dpkg/status

apt-cache policy intel-mkl
intel-mkl:
  Installed: 2020.0.166-1
  Candidate: 2020.0.166-1
  Version table:
 *** 2020.0.166-1 500
        500 http://archive.ubuntu.com/ubuntu focal/multiverse amd64 Packages
        100 /var/lib/dpkg/status