4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
5 |
============================== |
============================== |
6 |
|
|
7 |
|
o pkg/seaice |
8 |
|
- change (once again) the discretization for the evp solver: rather than |
9 |
|
squaring the averages of strain rates, average the squares to give |
10 |
|
a better approximation to VP behavior (in simple test); this breaks |
11 |
|
verification experiments that use EVP (lab_sea and lab_sea.hb87) |
12 |
|
--> update outputs |
13 |
|
o add CPP-flag (in mom_calc_visc.F) and TAF directive (in gmredi_slope_limit) |
14 |
|
to improve vectorization of adjoint code on TARGET_NEC_SX |
15 |
|
o verification/vermix: |
16 |
|
- change vermix to a more "classical" 1d column with high vertical |
17 |
|
resolution, seasonal heat flux forcing and constant wind forcing |
18 |
|
--> update outputs |
19 |
|
o pkg/seaice |
20 |
|
- fix diagnostics of principal stress components for EVP |
21 |
|
- add optional diagnostics of non-linear convergence for LSOR solver |
22 |
|
- make maximum zeta a runtime parameter |
23 |
|
o model/src (3-d solver): |
24 |
|
Undo modif from Dec 12, 2005 (solve_for_pressure.F cvs version v.49), |
25 |
|
which consisted in removing the Surf. Non-Hydrostatic pressure from the |
26 |
|
whole column and adding it to Eta (inspired from Casualli, 1999): |
27 |
|
Only affects 3-D solver initial guess and only if using ExactConserv, |
28 |
|
but tests did show a degradation of solver convergence. For this reason, |
29 |
|
decide to disable this piece of code. |
30 |
|
to recover previous version, just set zeroPsNH = exactConserv (line 78). |
31 |
|
Update output of exp. global_ocean.cs32x15.viscA4 & hs94.cs-32x32x5.impIGW |
32 |
|
o pkg/ggl90: |
33 |
|
- fix previous modif (unfortunatly has not been documented here). |
34 |
|
o pkg/seaice/autodiff: |
35 |
|
- add CPP flag to include clipping of zeta in EVP code; |
36 |
|
although clipping of zeta is not necessary for EVP and not recommended |
37 |
|
it was done by default, now clipping is disabled by default |
38 |
|
- modify computation and store zetaC/Z in seaice_evp for TAF |
39 |
|
- add CPP-flag SEAICE_ALLOW_CLIPVELS (undefined by default) |
40 |
|
- change seaice_calc_viscosities/strainrates for better |
41 |
|
vectorization after AD by TAF |
42 |
|
- rearrange common blocks in SEAICE.h once again for easier dumps of |
43 |
|
adjoint output, adjust adcommon.h and g_common.h |
44 |
|
- change flux form of default advection scheme for better vectorization |
45 |
|
of adjoint code |
46 |
|
- modify lab_sea/input.lsr to test the flux form of default advection |
47 |
|
scheme |
48 |
|
o autodiff: fix seaice fields in g_common.h (left over from 3 to 1 level |
49 |
|
transition in checkpoint61r) |
50 |
|
|
51 |
|
checkpoint61x |
52 |
|
o adjoint: adding two new verification experiments |
53 |
|
* tutorial_global_biogeo: an online dic sensitivity experiment |
54 |
|
(work with G. Forget) |
55 |
|
* tutorial_dic_adjoffline: an offline dic sensitivity experiment |
56 |
|
(an update of the Dutkiewicz et al., 2006 paper) |
57 |
|
Both use the same CO2FLUX cost function, but different controls |
58 |
|
Gradient checks look good, but some recomp. could still be traightened out |
59 |
|
Should add xx_dic control variable to test at some point |
60 |
|
o model/src: |
61 |
|
add run-time parameters to check and/or mask initial Temp & Salt |
62 |
|
o pkg/ptracers: |
63 |
|
apply zonal filter (if used) to all passive tracers |
64 |
|
|
65 |
|
checkpoint61w |
66 |
|
o adjoint: |
67 |
|
* update and finish work on adjoint of dic package (G. Forget) |
68 |
|
* start adjoint of offline package |
69 |
|
* bug fix for cube-sphere TLM code (active_file_g.F) |
70 |
|
o tutorial_global_oce_optim: |
71 |
|
now tested with 4 tiles (change data.grdchk and update output_adm.txt) |
72 |
|
and add SIZE.h_mpi for MPI testing |
73 |
|
o pkg autodiff |
74 |
|
- fix bug in autodiff_restore |
75 |
|
- make genmake2 remove comments and empty lines from ad_input_code.f in |
76 |
|
order to reduce file size (new script file remove_comments_sed) |
77 |
|
o allows to specify vertical profile of vertical viscosity (viscArNr) |
78 |
|
o pkg autodiff |
79 |
|
autodiff_ini_model_io: comment out writing of grid information to netcdf |
80 |
|
files, because at this time the grid fields are a bunch of zeros |
81 |
|
(this is analogous to ini_mnc_vars.F) |
82 |
|
o pkg seaice/autodiff: |
83 |
|
left-over from removing the 3-time-level fields: |
84 |
|
- small rearrangement of common blocks (SEAICE.h) |
85 |
|
- fix field declaration in order to get meaningful output in |
86 |
|
adseaice.*.nc (adcommon.h) |
87 |
|
o pkg obcs: |
88 |
|
- go through NH code only if nonHydrostatic=T |
89 |
|
- fix restart for Orlanski (Hydrostatic; still to fix NH+Orlanski restart) |
90 |
|
- move call to OBCS_CALC before SEAICE_MODEL which needs seaice-obc fields |
91 |
|
=> Update output of experiment seaice_obcs |
92 |
|
- useOBCSbalance: code is wrong with multi-tile / proc: add a stop. |
93 |
|
- fix NH+Orlanski restart (add wvel OBC value in pickup file) |
94 |
|
- fix restart for useOBCSprescribe + useEXF |
95 |
|
o pkg ctrl |
96 |
|
- make obcs as control parameter work also with useSingleCPUio |
97 |
|
- replace a few sny and snx by Ny and Nx to be consistent with |
98 |
|
ctrl_set_globfld_x/yz.F |
99 |
|
o pkg seaice |
100 |
|
- re-introduce global field frWtrAtm for the case of |
101 |
|
ALLOW_MEAN_SFLUX_COST_CONTRIBUTION defined (SEAICE.h, seaice_growth.F) |
102 |
|
- catch the case of an empty filename in seaice_cost_init_fixed.F |
103 |
|
o pkg ecco: |
104 |
|
- small bug fixes in the_main_loop |
105 |
|
. fix order of monitor arguments |
106 |
|
. change SEAICE_MULTILEVEL to SEAICE_MULTICATEGORY |
107 |
|
. move include DYNVARS.h up outside the ALLOW_AUTODIFF_TAMC block so |
108 |
|
that ALLOW_ECCO_FORWARD_RUN works with NONLIN_FRSURF |
109 |
|
- replace cos(yc*deg2rad) by _rA in cost_mean_heat/saltflux.F, fix the |
110 |
|
imbalance diagnostics for wmean_h/sflux = 0 |
111 |
|
- add actual values to f_obcsn/s/w/e in ecco_cost_final.F |
112 |
|
- catch the case of empty filenames in ecco_cost_init_fixed.F |
113 |
|
and ecco_readparms.F |
114 |
|
- fix the logic of wsalt=0.and.wsalt2=0 to wsalt=0.or.wsalt2=0 (and theta) |
115 |
|
to make this statement meaningfull in ecco_cost_weights.F |
116 |
|
o pkg zonal_filt: |
117 |
|
- change ZONAL_FILTER S/R interface (allows to filter 2-D fields) |
118 |
|
- when using rStar, filter uVel*hFacW instead of uVel |
119 |
|
(tested without topography: fix a Pb of growing instability with NLFS) |
120 |
|
o pkg/aim/thsice: |
121 |
|
-Add capacity to read a Q-Flux to be applied in slab-mixed layer thsice_slab_ocean.F |
122 |
|
(works as SST restoring) |
123 |
|
o pkg/thsice: |
124 |
|
-add specific parameter for salt restoring in slab mixed-layer |
125 |
|
(by default equal to the temperature restoring) |
126 |
|
|
127 |
|
checkpoint61v |
128 |
|
o pkg/kpp: |
129 |
|
-add double diffusive contributions as a hack, turned off by default |
130 |
|
for now, and the code can be excluded with a CPP-flag |
131 |
|
EXCLUDE_KPP_DOUBLEDIFF just as EXCLUDE_KPP_SHEAR_MIX |
132 |
|
-add a verification experiments to "vermix" to test this code |
133 |
|
o pkg/mom_common: |
134 |
|
Move computation of length scales to mom_init_fixed.F (avoid recomputation |
135 |
|
at each time-steps*levels). |
136 |
|
o pkg/layers: |
137 |
|
Created a new package for computing volume fluxes in temperature or |
138 |
|
[not yet implemented] density layers, e.g. for calculating residual |
139 |
|
overturning circulations. |
140 |
|
o pkg/diagnostics: |
141 |
|
implement a RS type version for the set of DIAGNOSTICS-FILL subroutines |
142 |
|
o fix few RS/RL type mismatch (check with ifort 11.1 "-warn all" option) |
143 |
|
o pkg/mdsio (& pkg/rw): rework high level S/R interface: |
144 |
|
To avoid mixing type (RS/RL) of input/output array argument, |
145 |
|
replace single mixed-type array with a pair from each type (RS/RL). |
146 |
|
o setting pickupSuff force to start from a pickup (even if nIter0=0): this |
147 |
|
is now implemented for all pkgs (previously only done for the main pickup) |
148 |
|
|
149 |
|
checkpoint61u |
150 |
|
o pkg/obcs: |
151 |
|
- change OB?eta from _RS to _RL and put them into a separate commen block |
152 |
|
- add code to read OB?eta from a file via obcs_prescribe_read: the code |
153 |
|
compiles and does not destroy any other experiments, the data is read |
154 |
|
properly, but the code is not tested in actual applications |
155 |
|
o ini_mnc_vars.F: if available (ALLOW_CAL), use startTime_1 and 2 to |
156 |
|
create a meaningful unit attribute for the time coordinate variable |
157 |
|
o pkg/ebm: add plenty of missing "_d 0" ; |
158 |
|
update results of experiment global_ocean_ebm (fwd + adjoint tests) |
159 |
|
Notes: several pieces of code only works for (very) specific set up |
160 |
|
(domain size, lat range, etc ...); should be documented & stop if one |
161 |
|
of those assumptions is not meet. |
162 |
|
o pkg/seaice: add missing "_d 0" in HSNOW initialisation |
163 |
|
update results of experiments: lab_sea (3 fwd tests + 2 adjoint tests) |
164 |
|
and offline_exf_seaice (seaicetd + adjoint). |
165 |
|
o add missing "_d 0" in ShortWave flux penetration (S/R SWFRAC): |
166 |
|
update results of experiments: lab_sea (4 fwd tests + 3 adjoint), |
167 |
|
seaice_obcs, global_ocean.cs32x15 (icedyn & thsice) and natl_box (x2). |
168 |
|
o few changes to pass when compiling with strick checking of arguments |
169 |
|
across S/R (allow to find few bugs in S/R calls). |
170 |
|
o pkg/seaice: |
171 |
|
- seaice_evp: move if-statement out of loop (TEM) at the cost of 11 new |
172 |
|
2d-fields, for slightly better performance |
173 |
|
|
174 |
|
checkpoint61t |
175 |
o pkg/seaice: |
o pkg/seaice: |
176 |
- completely remove code for leap-frog time stepping for second order |
- remove unused variables in seaice_evp |
177 |
advection |
- reduce number of SQRT in seaice_evp and seaice_calc_viscosities |
178 |
- clean up the 3-time levels of UICE,VICE,HEFF,AREA: |
o eeset_parms.F, open_copy_data_file.F & ini_parms.F: |
179 |
1. introduce fields for time step (N-1) |
named file replacing SCRATCH files (for TARGET_BGL & TARGET_CRAYXT): |
180 |
2. do not use levels 2 and 3 anymore (but keep them for now) |
each processor opens a different file (a fix from Matt). |
181 |
|
o cg2d: |
182 |
|
use simple EXCH (overlap size = 1 and ignore corners), like in cg3d.F : |
183 |
|
this reduces number of EXCH calls by 2 when using exch2. |
184 |
|
For now, keep previous size array for gc2d_nsa.F to avoid Pb with store |
185 |
|
directives and missing simple EXCH (exch_s3d_rx) adjoint code. |
186 |
|
o pkg/thsice: |
187 |
|
add code for Near Infra-Red albedo (from Jeff) |
188 |
|
(turned on with run-time flag: thSIce_calc_albNIR, in data.ice). |
189 |
|
o ini_model_io: |
190 |
|
move (from packages_readparms.F & initialise_fixed.F) to ini_model_io.F |
191 |
|
MNC & MONITOR initialisation calls; move MNC calls from {pkg}_readparms.F |
192 |
|
to {pkg}_init_fixed.F for {pkg}= kpp, seaice & shelfice. |
193 |
|
|
194 |
|
checkpoint61s |
195 |
|
o pkg/exch2 (exch2_rxN_cube*): |
196 |
|
remove last BARRIER (no need to synchronise after getting data from shared |
197 |
|
buffer (get) as long as any change to buffer (put,recv) is between BARRIER) |
198 |
|
o change tiling of natl_box (from 2 to 4 tiles) to get it tested with MPI+MTH |
199 |
|
and generate new output (for the 2 tests) using default rotationPeriod. |
200 |
|
o pkg/exch2: |
201 |
|
- always call exch2_*_cube, not exch-1 anymore, if useCubedSphereExchange=F |
202 |
|
(was already the case with AUTODIFF_EXCH2 defined) |
203 |
|
- add bj in exch2 arrays and S/R. |
204 |
|
- can now use pkg/exch2 for regular (non CS) set-up; switch test experiment |
205 |
|
MLAdjust to exch2 (identical results except with MPI -> truncation diff); |
206 |
|
o add pkg/longstep for ptracer timestep a multiple of U/V/T/S timestep |
207 |
|
- adds deltaT argument to the subroutines TIMESTEP_TRACER DWNSLP_APPLY |
208 |
|
GAD_ADVECTION GAD_CALC_RHS GAD_IMPLICIT_R GAD_SOM_ADVECT |
209 |
|
- verification experiment natl_box.longstep |
210 |
|
o build_options: ifort+mpi_aces optfile (following Constantinos advise): |
211 |
|
allow different versions of ifort (8 or 9) and different versions |
212 |
|
of mpi (mpich or mpich2) to be used with the same optfile. |
213 |
|
requires to load a netcdf module and set MPI_INC_DIR (see comments inside |
214 |
|
optfile). Change testing scripts (example_scripts/ACESgrid) accordingly. |
215 |
|
o pkg/seaice: |
216 |
|
fix previous modif (3-level field removed) for old-pickup reading |
217 |
|
|
218 |
|
checkpoint61r |
219 |
|
o pkg/seaice: |
220 |
|
- clean up the 3-time levels of UICE,VICE,HEFF,AREA in three steps: |
221 |
|
1. introduce fields for time step (N-1) |
222 |
|
2. do not use levels 2 and 3 anymore (but keep them) |
223 |
|
3. replace 3D versions of UICE,VICE,HEFF,AREA by 2D versions. This |
224 |
|
includes changes to pkg/thsice/thsice_get_velocity.F, |
225 |
|
model/src/external_forcing_surf.F, pkg/obcs/obcs_apply_seaice.F, and |
226 |
|
pkg/obcs/obcs_apply_uvice.F. |
227 |
|
- completely remove code for leap-frog time stepping for second order |
228 |
|
advection |
229 |
|
- adjusted adjoint accordingly; |
230 |
|
gradient checks have improved by an order of magnitude! |
231 |
|
|
232 |
checkpoint61q |
checkpoint61q |
233 |
o pkg/seaice: fix bug in free slip boundary conditons for LSOR solver |
o pkg/seaice: fix bug in free slip boundary conditons for LSOR solver |
234 |
- the boundary conditions along open boundaries are unclear, because the |
- the boundary conditions along open boundaries are unclear, because the |