| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o verification/offline_exf_seaice: replace adjoint experiment. |
| 8 |
|
o verification/tutorial_tracer_adjsens, pkg/ptracers and OpenAD: |
| 9 |
|
- add OpenAD setup to this exp and make ptracers package work with OpenAD |
| 10 |
|
o pkg/seaice/seaice_evp.F: mask the asymmetric contribution of the ice-ocean |
| 11 |
|
stress (when the turningAngle is non-zero) over open water to avoid |
| 12 |
|
stripes in the EVP-solutions after many EVP-substeps. This does not change |
| 13 |
|
the verification experiment (because turningAngle=0 in that experiment), but |
| 14 |
|
will affect other EVP solutions with turningAngle.ne.0 |
| 15 |
|
o verification, AD & TLM output: |
| 16 |
|
- new TAF version 2.3.10 (previously 2.3.8) add more parenthesis in multiple |
| 17 |
|
additions and multiplications which affect results at machine truncation |
| 18 |
|
level. Update few AD output (1D_ocean_ice_column, bottom_ctrl_5x5, |
| 19 |
|
lab_sea.noseaicedyn & offline_exf_seaice) and TLM output |
| 20 |
|
(1D_ocean_ice_column & tutorial_tracer_adjsens.som81) which had only 12 |
| 21 |
|
digits of matching. |
| 22 |
|
o pkg/seaice: |
| 23 |
|
modify lsr-algorithm so that it vectorizes AND converges by using a |
| 24 |
|
"zebra" method: split vector loop into two with step size 2; this |
| 25 |
|
solves a general convergence problem for the JFNK solver on a vector |
| 26 |
|
computer. It has the potential to even improve convergence for |
| 27 |
|
non-vector machines, but it does change the results so it is turned |
| 28 |
|
off by default. To turn on define CPP-flags SEAICE_VECTORIZE_LSR and |
| 29 |
|
SEAICE_VECTORIZE_LSR_ZEBRA in SEAICE_OPTIONS.h |
| 30 |
|
|
| 31 |
|
checkpoint64e (2013/03/05) |
| 32 |
|
o pkg/ecco and several verification experiments: |
| 33 |
|
Remove special versions of the_model_main.F, the_main_loop.F, forward_step.F |
| 34 |
|
from ecco package, and revert to unified routine in model/src/ |
| 35 |
|
Adjust verification experiments accordingly, especially gloal_oce_llc90 |
| 36 |
|
Should conclude long-standing work by Gael to reconsile the_main_loop with |
| 37 |
|
ECCO and will enable OpenAD's revolve loop in ECCO setup. |
| 38 |
|
o pkg/generic_advdiff: |
| 39 |
|
- implement compressible flow method for multi-dim advection (similar to |
| 40 |
|
gad_som_advect.F); add new option "GAD_MULTIDIM_COMPRESSIBLE" (since |
| 41 |
|
TAF generates many recomputations) to use it; no yet coded with implicit |
| 42 |
|
vertical advection. |
| 43 |
|
- change experiment advect_cs (new output) and advect_xy (same output) |
| 44 |
|
to test compressible flow multi-dim advection. |
| 45 |
|
o pkg/generic_advdiff: |
| 46 |
|
- fix SOM advection for unusual cases (linear Free-Surf in r*, other |
| 47 |
|
cases where linear Free-Surf is not @ k=1, e.g., useShelfIce) |
| 48 |
|
o pkg/seaice: |
| 49 |
|
- fix bug in the no-slip implementation in seaice_calc_strainrates.F, this |
| 50 |
|
bug affects two verification experiments (lab_sea and lab_sea.hb87), |
| 51 |
|
recover old (wrong) results by setting noSlipFac = 2. in |
| 52 |
|
seaice_calc_strainrates.F; |
| 53 |
|
- add another averaging method for etaZ (etaZmethod=3), a mix of 2 and 0 |
| 54 |
|
- add runtime parameter SEAICE_JFNKepsilon for that finite difference |
| 55 |
|
gradient in the Jacobian times vector operation (s/r seaice_jacvec) |
| 56 |
|
- fix line search in seaice_jfnk.F (s/r seaice_jfnk_update) |
| 57 |
|
o pkg/seaice: reorganize seaice_lsr.F to be able to reuse more code: |
| 58 |
|
- new subroutines seaice_lsr_tridiagu/v |
| 59 |
|
- introduce loop bounds i/jmin/max to be able to use a restricted |
| 60 |
|
additive Schwarz method, but not used for now. |
| 61 |
|
- rename SEAICE_CALC_LSR_COEFFS into SEAICE_LSR_CALC_COEFFS |
| 62 |
|
- use new subroutines seaice_lsr_tridiagu/v in s/r seaice_preconditioner |
| 63 |
|
o OpenAD support |
| 64 |
|
Reorganize S/R call for overall time-stepping loop to facilitate |
| 65 |
|
OpenAD revolve loop replacement |
| 66 |
|
(new S/R main_do_loop.F) |
| 67 |
|
o model/src: |
| 68 |
|
- add 3-D array 3rd dimension to argument list of S/R ADAMS_BASHFORTH2/3; |
| 69 |
|
- pkg/cheapaml: remove local S/R ADAMS2d and use instead standard S/R |
| 70 |
|
ADAMS_BASHFORTH2. |
| 71 |
|
o pkg/cheapaml: |
| 72 |
|
- add new parameter (default=F, same as before) to disable full domain |
| 73 |
|
(including land) periodicity in X & Y dir. |
| 74 |
|
- new S/R CHEAPAML_COPY_EDGES to copy relaxation field into edges of |
| 75 |
|
prognostics field ; merge the 2 S/R cycle_2d_tracer and timestep_2d_tracer |
| 76 |
|
into single S/R CHEAPAML_TIMESTEP. |
| 77 |
|
|
| 78 |
|
checkpoint64d (2013/02/17) |
| 79 |
|
o pkg/monitor (angular momentum): |
| 80 |
|
- do not add Eta contribution to the total AM if using rigid-lid; |
| 81 |
|
- correct for Adams-Bashforth in zonal-wind part (instead of in AM_eta). |
| 82 |
|
- horizontal discretisation: compute Zonal wind contribution separately for |
| 83 |
|
each component (instead of from cell centered averaged) and use the 4 |
| 84 |
|
grid-cell corner averaged value of Omega*(R*cos_Phi)^2 in AM_eta part. |
| 85 |
|
- Check that AM is conserved with linear dynamics (momAdvection=F and linear |
| 86 |
|
FreeSurf) if using vectorInvariant momentum. |
| 87 |
|
o model/src (horizontal grid): |
| 88 |
|
- move initialisation of all horizontal grid arrays to null value |
| 89 |
|
from specific horizontal grid S/R to main S/R INI_GRID; |
| 90 |
|
- add 2 new arrays to project model C-grid velocity on to zonal direction |
| 91 |
|
(similar to AngleCosC & -AngleSinC but at U and V point location) and |
| 92 |
|
compute them in S/R CALC_GRID_ANGLES. |
| 93 |
|
o pkg/seaice: replace a hard coded parameter by a runtime parameter |
| 94 |
|
SEAICE_JFNK_tolIter |
| 95 |
|
o pkg/seaice: introduce a restricted additive Schwarz (RAS) method in |
| 96 |
|
preconditioner for JFNK (run LSR into the overlaps as far as possible, |
| 97 |
|
i.e OLx/y-2, but overwrite the solution in the overlaps) |
| 98 |
|
- extend loop range in seaice_calc_viscosity |
| 99 |
|
- add new parameters SEAICE_OLx/y (default = OLx/y-2) |
| 100 |
|
- add flexible loop ranges to seaice_preconditioner, seaice_calc_lsr_coeffs |
| 101 |
|
- slightly modify loops in tridiagonal solver lsr-preconditioner |
| 102 |
|
- by default remove the extra exchange in seaice_preconditioner with a |
| 103 |
|
replacement of an undocumented CPP flag by SEAICE_PRECOND_EXTRA_EXCHANGE |
| 104 |
|
- modify offline_exf_seaice so that old results without RAS are reproduced |
| 105 |
|
for output.dyn_jfnk.txt |
| 106 |
|
in seaice_fgmres.F (unrelated to RAS) |
| 107 |
|
- add a cpp-flag option for a re-ordered mapping (potentially more |
| 108 |
|
efficient), but because this changes the results, turn it off by |
| 109 |
|
default |
| 110 |
|
- replace the save statement by local common blocks |
| 111 |
|
o utils/matlab: |
| 112 |
|
- fix tile-offset in rdmnc.m when output was produced using EXCH2. |
| 113 |
|
- update cs_grid/bk_line scripts to use load_grid.m (in utils/matlab) |
| 114 |
|
instead of local script "load_cs.m". |
| 115 |
|
o pkg/seaice: |
| 116 |
|
- fix a serious bug in seaice_fgmres.F, s/r seaice_scalprod, connected |
| 117 |
|
to the scalarProductMetric and the unrolling. By chance this does not |
| 118 |
|
change the verification experiment, because of the particular domain |
| 119 |
|
decomposition |
| 120 |
|
o pkg/diagnostics: |
| 121 |
|
- change default missing_value from UNSET_RL to -999. |
| 122 |
|
o pkg/layers: improve vectorization of layers_fluxcalc.F |
| 123 |
|
- step 1: split loops, make TatU/V 2D arrays. This alone reduces the cpu |
| 124 |
|
time spent in this routine by more than a factor of 3 on an NEC-SX8 |
| 125 |
|
- step 2: replace search algorithm by a bisection algorithm found in |
| 126 |
|
Press et al. Numerical Recipes in Fortran. This algorithm can be |
| 127 |
|
vectorized and finds the correct bin in approx. log2(Nlayers) steps. |
| 128 |
|
For now, use this algorithm only for TARGET_NEC_SX defined. |
| 129 |
|
- moved the search algorithm into a separate subroutine, that is kept |
| 130 |
|
within the same file for automatic inlining. |
| 131 |
|
- modified catching potential errors |
| 132 |
|
o pkg/ecco: |
| 133 |
|
- add compile switches ALLOW_SHALLOW_ALTIMETRY and ALLOW_HIGHLAT_ALTIMETRY. |
| 134 |
|
o model/src, pkg/autodiff, pkg/ctrl, pkg/grdchk : |
| 135 |
|
- modifications for ALLOW_GENTIM2D_CONTROL |
| 136 |
|
-- forward_step.F, the_main_loop.F : pass options and arrays for adjoint tapes. |
| 137 |
|
-- load_fields_driver.F : add call to CTRL_MAP_GENTIM2D( myTime, myIter, myThid ) |
| 138 |
|
-- add xx_gentim2d0/1 in ctrl_ad_check_lev1_dir.h checkpoint_lev1_directives.h etc |
| 139 |
|
-- use ctrl_swapffields.F (new) instead of exf_SwapFFields.F in ctrl_get_gen.F |
| 140 |
|
o model/src: |
| 141 |
|
- account for moist thermodynamics in (dry) potential temperature equation; |
| 142 |
|
update results of experiments: aim.5l_cs (x2), fizhi-cs (std + aqualev) |
| 143 |
|
and cpl_aim+ocn. |
| 144 |
|
- also switch to standard compiler (gfortran, was using ifort previously) |
| 145 |
|
for fizhi-cs experiments reference output. |
| 146 |
|
o pkg/ctrl: |
| 147 |
|
- 1st step toward generic time-varying control variable (2D) via CPP option |
| 148 |
|
ALLOW_GENTIM2D_CONTROL |
| 149 |
|
- note new namelist ctrl_nml_genarr in data.ctrl |
| 150 |
o verification/global_ocean.cs32x15: |
o verification/global_ocean.cs32x15: |
| 151 |
- switch off OLD_THSICE_CALL_SEQUENCE (both forward and AD code) to use new |
- switch off OLD_THSICE_CALL_SEQUENCE (both forward and AD code) to use new |
| 152 |
thsice calling sequence and update "output.icedyn.txt" output file. |
thsice calling sequence and update "output.icedyn.txt" output file. |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch