| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o pkg/seaice/seaice_evp.F: fix store directives and key computations, since |
| 8 |
|
there is no adjoint test for EVP, we don't know what it does to the adjoint |
| 9 |
|
solution, but at least there are no more recomputation warnings left |
| 10 |
|
o pkg/seaice/seaice_evp.F: slight modification of averaging so that EVP stands |
| 11 |
|
a chance of giving the same results as a fully converged VP (JFNK) solver |
| 12 |
|
o tools/adjoint_options/adjoint_diva: |
| 13 |
|
- fix bug (typo) in Tang-Linear FLAGS setting for MPI + DIVA |
| 14 |
|
o pkg/streamice: |
| 15 |
|
- spotted bug in streamice_driving_stress using dyG where should have been dxG |
| 16 |
|
o pkg/exf: |
| 17 |
|
- remove STOP statement that prevented to use wind or (wind-)stress |
| 18 |
|
interpolation for constant-in-time forcing fields (wind/stress period = 0). |
| 19 |
|
o pkg/seaice: |
| 20 |
|
- replace SEAICE_EVP_USE_ADAPTATION by run time parameters, requires |
| 21 |
|
5 additionals 2D fields and 6 new store directives |
| 22 |
|
- change logic in seaice_readparms.F: EVP code can now be turned on in |
| 23 |
|
various ways, setting SEAICE_deltaTev is no longer required, |
| 24 |
|
alternatively set SEAICE_evpAlpha/Beta, SEAICEuseEVPstar, SEAICEuseEVPrev, |
| 25 |
|
or SEAICEaEVPcoeff |
| 26 |
|
o pkg/shelfice/shelfice_thermodynamics.F |
| 27 |
|
- redefining uLoc/vLoc to u/v, instead of 2u/v |
| 28 |
|
o pkg/seaice/seaice_evp.F: introduce "adaptive EVP" by Madlen Kimmritz |
| 29 |
|
- this is an intermediate state of a truely converging EVP that has |
| 30 |
|
the potential of outperforming implicit VP methods, I do not recommend |
| 31 |
|
using it yet. |
| 32 |
|
- for now, "adaptive EVP" is turned on by setting SEAICE_EVP_USE_ADAPTATION |
| 33 |
|
and all parameters are hardwired |
| 34 |
|
- for convergence it still requires very smooth regularisations, that means, |
| 35 |
|
if SEAICE_DELTA_SMOOTHREG is defined, deltaCreg=deltaC+deltaMin, which is |
| 36 |
|
different from the VP case with deltaCreg=sqrt(deltaC**2+deltaMin**2); also |
| 37 |
|
specific averaging appears to be required, still subject to tuning |
| 38 |
|
- still needs cleaning and I need to turn parameter into runtime parameters |
| 39 |
|
- add code to compute and print residuals of iteration, compile with |
| 40 |
|
defining ALLOW_SEAICE_EVP_RESIDUAL to enable it |
| 41 |
|
o pkg/seaice/seaice_evp.F: fix a bug (factor of four in front of e12Csq) |
| 42 |
|
that entered with r1.54 on Sep04, 2015, -> update results |
| 43 |
|
|
| 44 |
|
checkpoint65o (2015/09/14) |
| 45 |
|
o pkg/mom_vecinv & mom_common: |
| 46 |
|
- add deep-atmosphere geometry factor in mom_calc_visc.F + in routines |
| 47 |
|
called from there. |
| 48 |
|
- add anelastic and deep-atmosphere geometry factor in pkg/mom_vecinv ; this |
| 49 |
|
allows to use Vector-Invariant form in deep atmos and anelastic formulation |
| 50 |
|
o pkg/seaice: introduce fast ice parameterisation following Itkin et al. (2015) |
| 51 |
|
and Koenig-Beatty+Holland (2012) |
| 52 |
|
- two new parameters SEAICE_tensilFac and SEAICE_tensilDepth |
| 53 |
|
- global 2D field tensileStrength, computed in seaice_calc_ice_strength |
| 54 |
|
- requires an extra input argument for S/R seaice_calc_viscosities |
| 55 |
|
- tensileStrength's depth dependence is different from Itkin et al. (2015) |
| 56 |
|
(to be changed and played with, for now exponential) |
| 57 |
|
o pkg/seaice: introduce new parameter SEAICEscaleSurfStress (default = .FALSE.) |
| 58 |
|
- if TRUE scale ice-ocean and ice-atmosphere stress acting on sea ice |
| 59 |
|
by AREA according to Connelly et al. (2004), JPO. |
| 60 |
|
- For EVP and the LSR solver, the implicit part of the drag term is |
| 61 |
|
the only non-zero term in the denominator when there is no ice (which |
| 62 |
|
prevented running the model with zero ice-ocean drag). If |
| 63 |
|
SEAICEscaleSurfStress = .TRUE., this results in a division by zero |
| 64 |
|
(or zero main diagonals BU/BV) which need to be caught. A practical |
| 65 |
|
consequence is that for open water, the momentum equation reduce |
| 66 |
|
to trivially 0 = 0 + 0 (for EVP). For LSR, BU/BV are reset to 1, if |
| 67 |
|
they happen to be zero, often leading to a non-zero solution over |
| 68 |
|
open water. |
| 69 |
|
o pkg/seaice: revise EVP code (seaice_evp.F) in preparation for a more |
| 70 |
|
efficient EVP method. |
| 71 |
|
- introduce area-weighted averages for e12 (analogously to |
| 72 |
|
seaice_calc_viscosities.F); this totally changes lab_sea.hb87 (2 digits |
| 73 |
|
remain, but also changing the order of terms in the previous average |
| 74 |
|
changed the results as much) -> update experiment |
| 75 |
|
- add SEAICE_DELTA_SMOOTHREG code |
| 76 |
|
- disentangle computation of zetaC/deltaC and zetaZ/deltaZ, also for |
| 77 |
|
TEM-code |
| 78 |
|
- adjust loop bounds so that only required fields are computed |
| 79 |
|
- add new averaging code for zetaZ/deltaZ (again similar to |
| 80 |
|
seaice_calc_viscosities.F), inactive by default for now (will change |
| 81 |
|
results again), but will replace old code eventually |
| 82 |
|
o eesupp/src: |
| 83 |
|
- add an other alternative for multi-proc GLOBAL_SUM (in global_sum_tile.F) |
| 84 |
|
to always cumulate tile local-sum in the same order |
| 85 |
|
(option #define GLOBAL_SUM_ORDER_TILES in CPP_EEOPTIONS.h). |
| 86 |
|
Still uses MPI_Allreduce (like standard code) but on array of tiles |
| 87 |
|
and should be faster than GLOBAL_SUM_SEND_RECV code. |
| 88 |
|
- switch almost all verification experiments that use GLOBAL_SUM_SEND_RECV |
| 89 |
|
(defined in CPP_EEOPTIONS.h_mpi) to GLOBAL_SUM_ORDER_TILES instead. |
| 90 |
|
- add CPP_EEOPTIONS.h_mpi with GLOBAL_SUM_ORDER_TILES defined in 14 more exp.: |
| 91 |
|
aim.5l_cs, cheapAML_box, deep_anelastic, dome, fizhi-cs-32x32x40, |
| 92 |
|
fizhi-cs-aqualev20, fizhi-gridalt-hs, global_ocean_ebm, matrix_example, |
| 93 |
|
natl_box, solid-body.cs-32x32x1, tutorial_advection_in_gyre, |
| 94 |
|
tutorial_baroclinic_gyre and tutorial_held_suarez_cs. |
| 95 |
|
|
| 96 |
|
checkpoint65n (2015/07/29) |
| 97 |
|
o pkg/thsice: |
| 98 |
|
- add a warning when used with pkg/aim but without updating SST |
| 99 |
|
(stepFwd_oceMxL=F & useCoupler=F & tauRelax_MxL<>-1 ). |
| 100 |
|
o pkg/offline: |
| 101 |
|
- change where Offline-Fields are loaded, from load_fields_driver.F to |
| 102 |
|
directly within FORWARD_STEP, just after updating iter number and time; |
| 103 |
|
- move setting of GM and KPP diffusivity out of OFFLINE_FIELDS_LOAD into new |
| 104 |
|
S/R OFFLINE_GET_DIFFUS that is called towards the end of DO_OCEANIC_PHYS; |
| 105 |
|
These changes allow: |
| 106 |
|
a) to recover exactly on-line results using deltaToffline=deltaTClock, |
| 107 |
|
tested (using offlineIter0= nIter0 -1 and offlineTimeOffset= -deltaT/2) |
| 108 |
|
both with Lin & Non-Lin Free-Surf, both with staggerTimeStep & synchronous |
| 109 |
|
time-step, both with GM & KPP files and with recomputing these fields; |
| 110 |
|
b) no needs to switch on staggerTimeStep anymore; |
| 111 |
|
c) simplify rescaling of offline horiz. velocity with Non-Lin Free-Surf. |
| 112 |
|
- update output of experiment tutorial_dic_adjoffline (only 3 digits left). |
| 113 |
|
o pkg/offline: |
| 114 |
|
- change (rarely used) offset-iteration parameter to an offset time (in s): |
| 115 |
|
can easily recover previous offset-iter effect by setting offset-time |
| 116 |
|
(offlineTimeOffset) to offset-iter x deltaToffline; |
| 117 |
|
- do not turn-off Temp & Salinity surface relaxation when using KPP (used |
| 118 |
|
in KPP as part of surface buoyancy forcing) in offline run. |
| 119 |
|
- fix ptracers KPP non-local term in the case where KPP is re-computed. |
| 120 |
|
o OpenAD |
| 121 |
|
- Adding preliminary ggl90 test to OpenAD verification. |
| 122 |
|
Config. with ggl90 currently diverges after 64 time steps |
| 123 |
|
so will need more work to become more stable |
| 124 |
|
- Adding preliminary kpp test to OpenAD verification. |
| 125 |
|
required small tweakes in kpp_routines.F |
| 126 |
|
- Merge and update from Krishna Narayanan's contrib area: |
| 127 |
|
* Add one variable for preliminary OpenAD DIVA |
| 128 |
|
* Add replacement time-stepping for OpenAD DIVA |
| 129 |
|
* genmake2 flag -diva (but only for OpenAD; preliminary) |
| 130 |
|
* required modifs for OAD_support |
| 131 |
|
successful preliminary test for tutorial_global_oce_biogeo/ |
| 132 |
|
o pkg/seaice: |
| 133 |
|
- initialise deltaC in seaice_lsr, cosmetic changes in seaice_lsr |
| 134 |
|
- add cpp-flag SEAICE_DELTA_SMOOTHREG for option of regularising |
| 135 |
|
delta with a smooth function in s/r seaice_calc_viscosity |
| 136 |
|
(no effect on EVP) |
| 137 |
|
- rename local suffix "sqr" into "sq" for more consistent variable names |
| 138 |
|
o pkg/seaice: |
| 139 |
|
add new parameter SEAICE_deltaMin that is used ***only*** |
| 140 |
|
for regularizing Delta (and nothing else, like the parameter |
| 141 |
|
which is also used for all sorts of things SEAICE_EPS). |
| 142 |
|
Defaults to SEAICE_EPS for backward compatibility only. |
| 143 |
|
o pkg/seaice: replace erroneously storing e12Csqr on the wrong tape with |
| 144 |
|
the wrong key by inialising it before use -> fixes a recomputation |
| 145 |
|
problem but does not fix the gradient |
| 146 |
|
o pkg/layers: |
| 147 |
|
- move resetting of flux arrays to outside the iLa loop |
| 148 |
|
(this caused layers_diapycnal to not work with more than one layer coord.) |
| 149 |
|
- fix bug in calculation of layers_recip_delta in layers_init_fixed.F |
| 150 |
|
o pkg/layers: |
| 151 |
|
- remove from common block 3D layers fields (used to diagnose transport |
| 152 |
|
in tracer coords) and declare them as local arrays in layers_calc.F |
| 153 |
|
o pkg/seaice: |
| 154 |
|
- Rename SEAICE_VECTORIZE_LSR_ZEBRA to SEAICE_LSR_ZEBRA and define it |
| 155 |
|
in lab_sea forward experiment (changes results -> updated) |
| 156 |
|
- update adjoint experiment lab_sea after cleaning up seaice_lsr_tridiagu/v |
| 157 |
|
o pkg/seaice: modify seaice_lsr.F in order to improve the gradient |
| 158 |
|
computations; for SEAICE_VECTORIZE_LSR |
| 159 |
|
- move the loops over which the tridiagonal solvers (seaice_lsr_tridiagu/v) |
| 160 |
|
sweep to the outside, |
| 161 |
|
- remove store directives and add "CADJ loop sequential" directives |
| 162 |
|
in analogy to model/src/solve_tridiagonal.F |
| 163 |
|
- replace many "#ifdef SEAICE_VECTORIZE_LSR_ZEBRA" by variable loop |
| 164 |
|
boundaries to yield more readable code. This has also the charming side |
| 165 |
|
effect that your can use SEAICE_VECTORIZE_LSR_ZEBRA without |
| 166 |
|
SEAICE_VECTORIZE_LSR (i.e. adjoint without recomputations in |
| 167 |
|
seaice_lsr_tridiagu/v now requires either of these flags, vectorization |
| 168 |
|
still requires SEAICE_VECTORIZE_LSR with SEAICE_VECTORIZE_LSR_ZEBRA as |
| 169 |
|
an additional option) |
| 170 |
|
The tridiagonal solvers are now completely analoguous to solve_tridiagonal.F |
| 171 |
|
but the adjoint gradients (which are affected by this change) still explode. |
| 172 |
|
o coupler (atm & ocn _compon_interf): |
| 173 |
|
- avoid using Exch2 IO mapping if not suitable. |
| 174 |
|
|
| 175 |
|
checkpoint65m (2015/06/15) |
| 176 |
|
o pkg/layers: |
| 177 |
|
- clean-up old parameters setting (from Sep 2012): |
| 178 |
|
+ remove old parameters setting from common block (declared locally); |
| 179 |
|
+ stop if any old parameter is mixed with new parameters; |
| 180 |
|
+ check for valid or consistent parameter setting (e.g., layers_name) |
| 181 |
|
- check that layers_bounds for Pot.Density are in the right range; |
| 182 |
|
this is supposed to catch un-updated setting after recently shifting |
| 183 |
|
potential density field "prho" by -1000 (now in "sigma" scale). |
| 184 |
o pkg/layers: |
o pkg/layers: |
| 185 |
- Added missing call to layers_fill for keeping track of tottend in |
- Added missing call to layers_fill for keeping track of tottend in |
| 186 |
pkg/diagnostics/diagnostics_fill_state.F |
pkg/diagnostics/diagnostics_fill_state.F |
| 198 |
thermodynamics.F, impldiff.F) and gad to save the necessary fields |
thermodynamics.F, impldiff.F) and gad to save the necessary fields |
| 199 |
for layers_thermodynamics calculations |
for layers_thermodynamics calculations |
| 200 |
o pkg/ptracers: |
o pkg/ptracers: |
| 201 |
- add capability of stepping a passive tracer forward |
- add capability of stepping a passive tracer forward without advection |
| 202 |
without advection by setting PTRACERS_advScheme to 0. |
by setting PTRACERS_advScheme to 0. |
| 203 |
o pkg/diagnostics: |
o pkg/diagnostics: |
| 204 |
- add few (threads) _BARRIER around any update of "diag_pkgStatus". |
- add few (threads) _BARRIER around any update of "diag_pkgStatus". |
| 205 |
o pkg/layers: |
o pkg/layers: |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch