IIRC per prior feedback by Intel support and comments by Intel users at its forums, gfortran + Intel MKL + OpenMP PARALLEL DO is suboptimal.
Given Intel releases oneAPI Toolkit, free, Fortran 2018, @implicitall 's use of ifort -qopenmp -mkl test.f90 && ./a.out
is what makes sense if using the code posted in @implicitall 's analysis upthread, and not gfortran + MKL
.