Optimization of dynspg_ts with key_vvl

[nemo_user]

In the time splitting case with variable volume, total depths are updated at the end of each barotropic time step by taking the sum of updated vertical scale factors. This implies several jpk loops inside the barotropic loop. A way to make this routine really faster (and to avoid the definition of 3d arrays) is to replace the following lines:

            CALL dom_vvl_sf( zsshun_e, 'U', zfse3un_e ) ;   CALL dom_vvl_sf( zsshvn_e, 'V', zfse3vn_e )

            ! Ocean depth at U- and V-points
            hu_e(:,:) = 0.e0
            hv_e(:,:) = 0.e0

            DO jk = 1, jpk
               hu_e(:,:) = hu_e(:,:) + zfse3un_e(:,:,jk) * umask(:,:,jk)
               hv_e(:,:) = hv_e(:,:) + zfse3vn_e(:,:,jk) * vmask(:,:,jk)
            END DO

by:

hu_e(:,:) = hu_0(:,:) + zsshun_e(:,:) 
hv_e(:,:) = hv_0(:,:) + zsshvn_e(:,:) 

where hu_0 and hv_0 would be new arrays representing total depths at equilibrium, filled once for all at the begining of the run.

Jerome