Compile error with Intel 2019

[dftd4]

Compile error with Intel 2019

[awvwgk]

Unfortunately, this is an issue with the latest Intel compiler, the concurrent iteration counters are correctly declared as private, but this clause is simply ignored by the Intel compiler. To fix it the offending code pieces must be rewritten entirely.

Was fixed in xtb by rewriting the D4 implementation to not use the offending construct anymore, maybe I can just copy it over here.

[mkrompiec]

This will pop up more often as more people will move to Intel 2019... Would it make sense to report this to Intel as a bug?
If you can copy that part from xtb, then perhaps this is the easiest way to fix this.

[awvwgk]

The usual sed -i 's/(none)/(shared)/g' source/dftd4.f90 can be used as workaround or to generate the necessary patches, until I have created a better fix for this.

[awvwgk]

This will pop up more often as more people will move to Intel 2019... Would it make sense to report this to Intel as a bug?

Actually, this is an Intel 2020 issue, it works fine with Intel 2019, but they bumped the version from 19.0 to 19.1 for the 2020 release for whatever reason.

This construct is working with 19.0.3.199 20190206 and is failing with 19.1.1.217 20200306: ifort -qopenmp mwe.f90

implicit none
integer :: i, j
integer :: vec(10)

vec(:) = 0
!$omp parallel do default(none) reduction(+:vec) private(i, j)
do i = 1, size(vec)
   do concurrent(j = 1:size(vec))
      vec(i) = vec(i) + j * j
   end do
end do
end

Dropping the concurrency to workaround the issue is straight-forward if not tedious.

[mkrompiec]

Seems like this change is mentioned in Intel's changelog: https://software.intel.com/content/www/us/en/develop/articles/intel-fortran-compiler-191-for-linux-release-notes-for-intel-parallel-studio-xe-2020.html#changed_comp_behaviour