Context

Branch dev_merge_2017: zdf_iwm_init reads from a namelist whose name does not match that of the namelist supplied in the namelist_ref file.

Analysis

As a consequence updated values for the wave-driven mixing parameters are not not used

Recommendation

Change every occurrence of namzdf_iwm_new to namzdf_iwm to match entries in the reference namelist ...

Context

A 2018 development to add an adaptive-implicit vertical advection based on Shchepetkin 2015 (as used in CROCO).

Proposal

• Option to be selected via a logical set in the namzdf name list (ln_zad_Aimp)
• New routine to compute Courant number criteria and partition wn into wn + wi where wi is to be treated implicitly (hopefully at infrequent and sparse locations). New routine added to sshwzv.F90 and called from step
• wi treated implicitly by adding appropriate terms to the tridiagonal solvers in tra_zdf and dyn_zdf
• include diagnostic output to track activity of the scheme

Development branch:

​http://forge.ipsl.jussieu.fr/nemo/browser/NEMO/branches/2018/dev_r9956_ENHANCE05_ZAD_AIMP

Context

Reorganisation of nemogcm.F90 and mppini.F90 to better separate domain decomposition functions from the rest of the initialisation. This reorganisation (thanks to Sebastien) also fixes issues with land suppression and offers the opportunity to tidy up some long standing issues relating to halo use in prt_ctl (prtctl.F90). A few other changes have been made in the process which remove redundant operations or add new features. These are detailed here separately and have been submitted as separate changesets for clarity

Stages

1. Main reorganisation

Relocation of partition functions (nemo_partition, factorise and nemo_ndfcom) from nemogcm.F90 to mppini.F90 and associated move of allocations from dom_oce.F90 to mppini.F90. mpp_init is now called before nemo_alloc and contains all code related to setting the individual domain sizes. The algorithm has been corrected to correctly set domain sizes and identify neighbours in all cases including land suppression. layout.dat now labels entries with processor rank rather than narea. Modules affected are:

NEMOGCM/NEMO/OFF_SRC/nemogcm.F90
NEMOGCM/NEMO/OPA_SRC/nemogcm.F90
NEMOGCM/NEMO/SAO_SRC/nemogcm.F90
NEMOGCM/NEMO/SAS_SRC/nemogcm.F90
NEMOGCM/NEMO/OPA_SRC/DOM/dom_oce.F90
NEMOGCM/NEMO/OPA_SRC/LBC/mppini.F90

Submitted at Changeset: [9436]

2. Code update:

Line 2396 of iom.F90 contained the statement:

      ! do we read the overlap 
      ! ugly patch SM+JMM+RB to overwrite global definition in some cases
      llnoov = (jpni * jpnj ) == jpnij .AND. .NOT. lk_agrif

with the algorithm fixes there should no longer be a need to read overlaps in the land suppressed case. The statement has been rationalised to:

      llnoov = .NOT. lk_agrif

subject to deeper testing beyond SETTE. Module affected is:

NEMOGCM/NEMO/OPA_SRC/IOM/iom.F90

Submitted at Changeset: [9437]

3. New feature:

Currently a bad ctl_opn call only reports an error if the master processor raises the error first. Code has been added to ctl_opn so that any other processor raising the error reports to stderr. Module affected:

NEMOGCM/NEMO/OPA_SRC/LBC/lib_mpp.F90

Submitted at Changeset: [9438]

4. Code update:

Line 546 of sbcmod.F90 contains a redundant lbc_lnk call for emp. This is an old bug relating to non-symmetric forcing data across the north-fold (i.e. gridcells that are meant to be the same point receiving different emp fluxes - leading to divergence errors). Modern datasets and/or interpolation on the fly should negate the need for this exchange. It has been commented subject to deeper testing beyond SETTE. Module affected:

NEMOGCM/NEMO/OPA_SRC/SBC/sbcmod.F90

Submitted at Changeset: [9439]

5. New feature/ reduction of variants:

In the mono processor case, prt_ctl can be used to print mean values for a sub-domain (a psuedo-mpp domain). Code has been added to produce a layout.dat-like file (layout_prtctl.dat) to detail the sub domain parameters. At the same time, it is now risky to allow sums over the whole sub-domain since many routines (zdftke, for example) only compute interior values. The optional argument 'overl' which enabled this has been suppressed and all references to it removed elsewhere. Main module affected is:

NEMOGCM/NEMO/OPA_SRC/IOM/prtctl.F90

Other modules where the use of overl has been suppressed:

NEMOGCM/NEMO/OFF_SRC/dtadyn.F90
NEMOGCM/NEMO/OPA_SRC/TRA/eosbn2.F90
NEMOGCM/NEMO/SAS_SRC/sbcssm.F90
NEMOGCM/NEMO/OPA_SRC/DYN/sshwzv.F90
NEMOGCM/NEMO/OPA_SRC/TRD/trdmxl.F90
NEMOGCM/NEMO/OPA_SRC/ZDF/zdfddm.F90
NEMOGCM/NEMO/OPA_SRC/ZDF/zdfgls.F90
NEMOGCM/NEMO/OPA_SRC/ZDF/zdfiwm.F90
NEMOGCM/NEMO/OPA_SRC/ZDF/zdfmxl.F90
NEMOGCM/NEMO/OPA_SRC/ZDF/zdftke.F90

Submitted at Changeset: [9940]

6. New feature:

Add netcdf version of the run.stat file to stpctl.F90. This contains the same information written to the run.stat text file. The current implementation is basic and will overwrite any existing run.stat.nc file. Output is flushed to disk (NF90_SYNC) every 100 timesteps. This may need adjusting if this output is being used for run-time monitoring. Modules affected are:

NEMOGCM/NEMO/OPA_SRC/stpctl.F90 NEMOGCM/NEMO/SAS_SRC/stpctl.F90 NEMOGCM/CONFIG/TEST_CASES/CANAL/MY_SRC/stpctl.F90

Submitted at Changeset: [9441]

7. Local architecture changes:

Changes to local architecture and batch files to update compiler and XIOS versions:

NEMOGCM/SETTE/BATCH_TEMPLATE/batch-X64_MOBILIS
NEMOGCM/ARCH/arch-X64_MOBILIS.fcm

Submitted at Changeset: [9442]

SETTE notes

Full SETTE testing was carried out on the NOCS Mobilis platform. Tests were successful with the Intel 17.0.4 compiler (-i4 -r8 -O3 -fp-model source -xAVX) but there are problems with the older Intel 14.0/2013_sp1.3.174 compiler which has been reliable previously. Recommend discontinuing use of v14 compilers.

SETTE validation report :  @ r9442
WSAS_ST               ice restarts are IDENTICAL  passed :  20180327
WGYREPIS_ST           run.stat    restartability  passed :  20180327
WGYREPIS_ST           tracer.stat restartability  passed :  20180327
WORCA2LIM3PIS_ST      run.stat    restartability  passed :  20180327
WORCA2LIM3PIS_ST      tracer.stat restartability  passed :  20180327
WORCA2OFFPIS_ST       tracer.stat restartability  passed :  20180327
WAMM12_ST             run.stat    restartability  passed :  20180327
WISOMIP_ST            run.stat    restartability  passed :  20180327
WORCA2AGR_ST          run.stat    restartability  passed :  20180327
WGYREPIS_ST           run.stat    reproducibility passed :  20180327
WGYREPIS_ST           tracer.stat reproducibility passed :  20180327
WORCA2LIM3PIS_ST      run.stat    reproducibility passed :  20180327
WORCA2LIM3PIS_ST      tracer.stat reproducibility passed :  20180327
WORCA2OFFPIS_ST       tracer.stat reproducibility passed :  20180327
WAMM12_ST             run.stat    reproducibility passed :  20180327
WISOMIP_ST            run.stat    reproducibility passed :  20180327
WORCA2_LIM3_OBS_ST    run.stat    reproducibility passed :  20180327
WORCA2AGR_ST          run.stat    reproducibility FAILED :  20180327
AGRIFNOZ             AGRIFNO  AGRIF: run.stat unchanged - test  passed :  20180327 20180327

The WORCA2AGR_ST reproducibility test is reporting a failure but only diverges after 70 time steps which is as good as I've even been able to get with this test on this platform. The tests included and additional ORCA2LIM3PISCES reproducibility test with a 60 processor, 8x8 decomposition to test land suppression. Results were identical to the standard 4x8 and 8x4 fully populated cases over 1000 time steps.