Restoring restartability of WED025 and ISOMIP+ with the current trunk and nn_hls=1

[acc]

Context

The restartability and reproducibility of 4.2_RC (current trunk) was broken by changes a r15084

Analysis

Adding an lbc_lnk on pqfwf in isfcav.F90 fixes the issue and can be justified thus:

pqfwf is set everywhere in a call to isfcav_mlt but the SETTE tests use:

        cn_gammablk = 'vel'     ! scheme to compute gammat/s (spe,ad15,hj99)

which employs a zustar value based on putbl, pvtbl ! velocity in the losch top boundary layer. These velocities are averaged onto T-points and are not set for i=1 and j=1. Comments in isftbl.F90 are aware of this:

         DO_2D( nn_hls-1, nn_hls, nn_hls, nn_hls )
            pvarout(ji,jj) = 0.5_wp * (zvarout(ji,jj) + zvarout(ji-1,jj))
         END_2D
         ! lbclnk not needed as a final communication is done after the computation of fwf
.
.
         DO_2D( nn_hls, nn_hls, nn_hls-1, nn_hls )
            pvarout(ji,jj) = 0.5_wp * (zvarout(ji,jj) + zvarout(ji,jj-1))
         END_2D
         ! lbclnk not needed as a final communication is done after the computation of fwf

But presumably the 'final communication' has subsequently been removed and needs to be reinstated. Running with:

      cn_gammablk = 'spe'

which uses a constant, specified exchange coefficient, is restartable and reproducible without the lbc_lnk thus proving that the gammats_vel is responsible. Exchanging pqfwf fixes the issue with one exchange but it may be more robust to exchange both zutbl and zvtbl at source in isftbl.F90. I will commit the former for now.

Recommendation

isfcav.F90

@@ -169 +169 @@
          ! set temperature content
          ptsc(ji,jj,jp_tem) = zqh(ji,jj) * r1_rho0_rcp
       END_2D
+      CALL lbc_lnk( 'isfmlt', pqfwf, 'T', 1.0_wp)
       !
       ! output fluxes
       CALL isf_diags_flx( Kmm, misfkt_cav, misfkb_cav, rhisf_tbl_cav, rfrac_tbl_cav, 'cav', pqfwf, zqoce, zqlat, zqhc)

[acc]

In 15143:

Error: Failed to load processor CommitTicketReference

No macro or processor named 'CommitTicketReference' found

[acc]

Closed by changeset: 15143

[mathiot]

Exchanging pqfwf fixes the issue with one exchange but it may be more robust to exchange both zutbl and zvtbl at source in isftbl.F90. I will commit the former for now.

By doing so, you have 2 com (one for utbl and one for vtbl) instead of 1, this was the main thinking when I move the lbclnk all at the end (isfcav). Maybe moving it on zustar will do the job and make it clearer why this call is needed.

It is possible (not tested) this call is not needed if nn_hls == 2.