Wrong computation of global number in BORD3D in parallel
In BORD3D subroutine, there is a computation of global number index of node. In sequential, it is OK, using NBOR3 array:
N=NBOR3%I(IBORD)
There is a bug in parallel as NPOIN2 is the number of 2D nodes in the subdomain, not the the total number of 2D nodes of the mesh before partioning:
N=MESH2D%KNOLG%I(NBOR2%I(K))+(NP-1)*NPOIN2
By chance, the computed variable N is only sent to SL3, VIT3 or TR3 functions and not used but a user who may modify the functions or the associated USER subroutines (USER_SL3, USER_VIT3 or USER_TR3) and use N would fall on a bug in parallel.