Segfault when structure is large about dftd4

[awvwgk]

Did you set the stacksize for the system as well ulimit -s unlimited?

[iloncaric]

I tested it also with executable: "dftd4 --func r2scan --grad dftd4.txt POSCAR" and it also gives segfault. Is there a hardcoded limit to the number of atoms in the unit cell?

[e-kwsm]

Segmentation fault occurs at

dftd4/src/dftd4/ncoord.f90

Lines 171 to 173 in 3844dc1

	!$omp parallel do default(none) reduction(+:cn, dcndr, dcndL) &
	!$omp shared(mol, trans, cutoff2, rcov, en) &
	!$omp private(jat, itr, izp, jzp, r2, rij, r1, rc, countf, countd, sigma, den)

even if OMP_NUM_THREADS is set to 1, and OMP_STACKSIZE to 1G (gfortran 13.2.1, ifort 2021.10.0, and ifx 2023.2.0).

[iloncaric]

When I noticed this problem "ulimit -s unlimited" was set, but then when I was making minimal test on PC it was not set. So minimal test could actually work if "ulimit -s unlimited" is set. But if you try to increase the size of the system a bit more, then again there is the segfault. This can be apparently solved by increasing OMP_STACKSIZE.
Since OMP_STACKSIZE cannot be set to unlimited, one might get segfaults until OMP_STACKSIZE is large enough.

You might want to check if there is another way of managing memory to avoid the need for a very large OMP_STACKSIZE. For 10 000 atoms cell OMP_STACKSIZE has to be on the order of OMP_STACKSIZE=3G with 8 threads.

In any case, thanks a lot for the prompt help!