Covariant dipoles in tech-master broken on GPUs after last changes

[sangallidavide]

Below the changes which are buggy with CUDAF.
Probably due to slicing of wave-functions

-     call WF_symm_kpoint((/1,Dip%ib(2)/),ikbz,Xk,WF_symm)
-     !  
-     call WF_shift_kpoint((/1,Dip%ib(2)/),ikbz,WF_shifts(ikbz,:),Xk,WF_symm,WF_ik)
+     do ib=1,Dip%ib(2)
+       !
+       call WF_symm_kpoint((/ib,ib/),ikbz,Xk,WF_symm)
+       !  
+       call WF_shift_kpoint((/ib,ib/),ikbz,WF_shifts(ikbz,:),Xk,WF_symm,WF_ik(:,:,ib:ib,:))
+       !
+     enddo

and

-         call WF_shift_kpoint((/1,Dip%ib(2)/),ikbzp,shift,Xk,WF_symm,WF_ikp)
-         !
-         call WF_free(WF)
-         !
-         do ib=1,Dip%ib(2)
-           if(.not.PAR_IND_VAL_BANDS_OVLP%element_1D(ib)) cycle
-           do ibp=1,Dip%ib(2)
-             if(.not.PAR_IND_CON_BANDS_OVLP%element_1D(ibp)) cycle
+         do ibp=1,Dip%ib(2) 
+           if(.not.PAR_IND_CON_BANDS_OVLP%element_1D(ibp)) cycle
+           call WF_symm_kpoint((/ibp,ibp/),ikbzp,Xk,WF_symm)
+           call WF_shift_kpoint((/ibp,ibp/),ikbzp,shift,Xk,WF_symm,WF_ikp)
+           do ib=1,Dip%ib(2)
+             if(.not.PAR_IND_VAL_BANDS_OVLP%element_1D(ib)) cycle
              DIP_S(ib,ibp,id+(istep-1)*3,ikbz,i_sp_pol)= &
-             &  Vstar_dot_V(wf_ng_overlaps*n_spinor,WF_ik(:,:,ib,i_sp_pol),WF_ikp(:,:,ibp,i_sp_pol))
+             &  Vstar_dot_V(wf_ng_overlaps*n_spinor,WF_ik(:,:,ib,i_sp_pol),WF_ikp(:,:,1,i_sp_pol))
            enddo ! ibp
          enddo ! ib

[sangallidavide]

Fixed in branch tech-bse presently under pull request with commit 092f32e

[attacc]

Thank you

[palful]

The covariant dipoles on GPU were also not working for us (but since forever, not just due to recent changes). If this change works then it's very convenient also for the GW@DFT+U part!