non reproductible NP folding

[smasson]

Context

The model is not reproductible with jperio = 4 or 6 :-)

Analysis

I had the stupid idea to test jperio = 4 or 6 in a very simple configuration: square box, flat bottom, no continents, constant temperature and salinity, flat ssh, no currents.

• with jpni = 1 and jperio = 4 or 6, we get the expected result: constant temperature everywhere except for the first line that is set to 0 because of the closed southern boundary.

• if jpni > 1 and jperio = 4 or 6, there is some points with temperature set to 0 around the upper left/right corners of the domain! These points are visible in outputs and restart files. Their exact localisation is depending on the value of jperio and ln_nnogather. see attached image.

As points located near the NP are usaully (must be) masked, we never saw this problem...

It is a real bug?

Fix

...

[acc]

I think this is a bug but a benign one since the poles should always be in land. Looking at the ln_nnogather= false option, It looks as if this sequence from mpp_nfd_generic.F90:

         ztab(:,:,:,:,:) = 0._wp
         DO jr = 1, ndim_rank_north         ! recover the global north array
            iproc = nrank_north(jr) + 1
            ildi  = nldit (iproc)
            ilei  = nleit (iproc)
            iilb  = nimppt(iproc)
            DO jf = 1, ipf
               DO jl = 1, ipl
                  DO jk = 1, ipk
                     DO jj = 1, ipj
                        DO ji = ildi, ilei
                           ztab(ji+iilb-1,jj,jk,jl,jf) = znorthgloio(ji,jj,jk,jl,jf,jr)
                        END DO
                     END DO
                  END DO
               END DO
            END DO
         END DO
         DO jf = 1, ipf
            CALL lbc_nfd( ztab(:,:,:,:,jf), cd_nat LBC_ARG, psgn LBC_ARG )   ! North fold boundary condition
         END DO
         !
         DO jf = 1, ipf
            DO jl = 1, ipl
               DO jk = 1, ipk
                  DO jj = nlcj-ipj+1, nlcj             ! Scatter back to ARRAY_IN
                     ij = jj - nlcj + ipj
                     DO ji= 1, nlci
                        ARRAY_IN(ji,jj,jk,jl,jf) = ztab(ji+nimpp-1,ij,jk,jl,jf)
                     END DO
                  END DO
               END DO
            END DO
         END DO

can leave the global edges set to zero (ztab is set from ildi to ilei for each processor, then folded and scattered back to 1 to nlci). Not sure why lbc_nfd doesn't fill them (too many options to work through on paper). That's probably where to monitor first (if it is necessary).

[smasson]

In 9805:

Error: Failed to load processor CommitTicketReference

No macro or processor named 'CommitTicketReference' found

[smasson]

I agree, see [9805]
jperio, 0, 4, 5, 6 and 7 are now ok with or without ln_nnogather. I still have an issue with jperio = 3. Strange...

[smasson]

because of the the characteristics of the NP folding at T point, jperio = 3 requires that the first and the last 2 columns are masked. So there is no issue with jperio = 3 as I originally though. The ticket can be closed.

[smasson]

In 9850:

Error: Failed to load processor CommitTicketReference

No macro or processor named 'CommitTicketReference' found

[smasson]

In 9908:

Error: Failed to load processor CommitTicketReference

No macro or processor named 'CommitTicketReference' found

[smasson]

trunk [9908] : bugfix following [9805]