Missing subroutines for C1D case in lbclnk

[hadcv]

Compilation of the C1D configuration at version 3.5 beta fails due to missing dummy subroutines lbc_bdy_lnk_3d, lbc_bdy_lnk_2d and lbc_lnk_2d_e in lbclnk.F90.

These need to be specified for the 'key_c1d' case in addition to the 'else' case.

[rbourdal]

Hi,

The problem comes from the addition of new lbc routines.
This problem can be solve just by changing the order of soubroutines and cpp_key

at this time the order is:
#if defined key_mpp_mpi
...
#else
# if defined key_c1d

SUBROUTINE lbc_lnk_3d_gather
SUBROUTINE lbc_lnk_3d
SUBROUTINE lbc_lnk_2d

#else

SUBROUTINE lbc_lnk_3d_gather
SUBROUTINE lbc_lnk_3d
SUBROUTINE lbc_bdy_lnk_3d
SUBROUTINE lbc_bdy_lnk_2d
SUBROUTINE lbc_lnk_2d
SUBROUTINE lbc_lnk_2d_e

#endif
#endif

You erase this problem by changing in this way:
#if defined key_mpp_mpi
...
#else
# if defined key_c1d

SUBROUTINE lbc_lnk_3d_gather
SUBROUTINE lbc_lnk_3d
SUBROUTINE lbc_lnk_2d

#else

SUBROUTINE lbc_lnk_3d_gather
SUBROUTINE lbc_lnk_3d
SUBROUTINE lbc_lnk_2d

#endif

SUBROUTINE lbc_bdy_lnk_3d
SUBROUTINE lbc_bdy_lnk_2d
SUBROUTINE lbc_lnk_2d_e

#endif

But I don't close this ticket because:
This modification is OK for NEMO's trunk before the modification of iom/xio interface in the trunk (Ticket #1122).
But with the new version of the trunk (NEMO revision 3956) et the xios associated (rev 445 and more) there is a conflict with the key_mpp_mpi:
The 1D configuration sould NOT use the key_mpp_mpi to apply the 1D-specific lbclnk routines and 1D configuration must use key_iomput (line 490 of
nemogcm.F90) for the outputs. The problem is that the new xios serveur (rev445) and NEMO (with key_iomput) must be compiled with the key_mpp_mpi !
Maybe a good solution is remplementing the lbc 1D-specific routines in the mpp case... ?

Regards,
Romain

[hadcv]

Hi Romain,

When you say that XIOS requires compilation with key_mpp_mpi, are you referring to the following line in iom.F90?

CALL xios_context_initialize(TRIM(clname), mpi_comm_opa)

This is also a block to running C1D in version 3.5 since mpi_comm_opa isn't defined without key_mpp_mpi. I have fixed this in my working copy by replacing with:

#if defined key_mpp_mpi

CALL xios_context_initialize(TRIM(clname), mpi_comm_opa)

#else

CALL xios_context_initialize(TRIM(clname), 0)

#endif

This and a definition of the lbc_bdy subroutines outside the C1D cpp key allow the model to run, and I have tested it with the recent XIOS changes. What do you think?

Daley

[rbourdal]

Hi Daley,

My first idea was to be able running also with key_mpp_mpi but I think you're right.
In a first step, we should be able running with "classical" option (without key_mpp_mpi).
I purpose you the following modifications:

domzgr.F90 fix to compile with check bounds option (pb of write in s-coord case !)
iom.F90 Correct the call to xio if not key_mpp_mpi (your solution)
lbclnk.F90 change the order of routines to be OK in 1D case.
solmat.F90 add lbclnk in 1D case (just pb at initialization) because bmask=/0 everywhere and gcdmat not defined everywhere (pb can be correct also with creation of nemo_init_1d with just the usefull init. !)
step_c1d.F90 Add iom_init like in step.F90

It's OK for you ?

Romain

[hadcv]

Hi Romain,

That seems ok to me, it covers the three bugs I addressed in my local copy (lbclnk, step_c1d and iom). In step_c1d.F90, should we also change:

CALL iom_setkt( kstp )

to

CALL iom_setkt ( kstp - nit000 + 1 )

for consistency with step.F90?

Daley

[rbourdal]

Yes I have changed also this modified call.
I have also added a call to diahth to access at more MLD diagnotics...

Romain