4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
5 |
============================== |
============================== |
6 |
|
|
7 |
|
o pkg/seaice adjoint of LSR: |
8 |
|
- substantially rework STORE dirs. |
9 |
|
- add new flag SEAICE_LSR_ADJOINT_ITER and parameter SOLV_MAX_FIXED |
10 |
|
this flag will do SOLV_MAX_ITERS but only store SOLV_MAX_FIXED fields |
11 |
|
and only perform SOLV_MAX_FIXED adjoint iters. |
12 |
|
- now produces non-zero LSR-related gradients, and acceptable gradients for |
13 |
|
An's zonal (x-z) test case (to be checked in) |
14 |
|
- lab_sea: produces acceptable gradients for lab_sea with forward/backward LSR |
15 |
|
(but these are screwed up by the thermodynamics in the first place); |
16 |
|
reduce from 4 to 3 timesteps to avoid t.d. issue as temporary solution; |
17 |
|
output_adm.txt also reflects changes to seaice_lsr |
18 |
|
- global_ocean.cs32x15: |
19 |
|
* the "default" input_ad.seaice has been re-defined to always exclude dynamics |
20 |
|
(output_adm.seaice.txt was updated) |
21 |
|
* new input_ad.seaice_dynmix tests a forward LSR / backward free-drift |
22 |
|
(output_adm.seaice_dynmix.txt) |
23 |
|
o test-exp. exp4+stevens: |
24 |
|
- switch off pTracers and nonHydrostatic (restart problems with NH); update output |
25 |
|
o pkg/seaice (lsr solver): |
26 |
|
- put coriolis term into FORCEX,Y (as in seaice_evp.F; previously in rhsU,V only); |
27 |
|
update output (change truncation) of FWD test exp lab_sea & global_ocean.cs32x15 |
28 |
|
- fix SEAICEuseFREEDRIFT for new OBCS version (ndef OBCS_UVICE_OLD) |
29 |
|
- new parameter "LSR_mixIniGuess" allows (if=2,4) to mix free-drift solution |
30 |
|
into initial guess for linearised solver; use weight ~ (1/err)^2,4. |
31 |
|
- compute (with new S/R) and print initial (+ free-driff sol. if available) |
32 |
|
and final residual of linearised system. |
33 |
|
o pkg/pp81: add diagnostics for vertical eddy viscosity and diffusivity |
34 |
|
(requires new subroutines pp81_init_fixed.F/pp81_diagnostics_init.F) |
35 |
|
o pkg/obcs: Stevens BCs (still only for east and west) |
36 |
|
- fix Stevens BCs by using intermediate velocities of previous |
37 |
|
time step as "baroclinic" contribution rather velocity of one grid cell |
38 |
|
inside the boundaries; more consistent but makes the entire treatment |
39 |
|
more complicated, requires new pickup files and two new S/Rs: |
40 |
|
obcs_copy_uv_n and obcs_save_uv_n |
41 |
|
keep old version with flag OBCS_STEVENS_USE_INTERIOR_VELOCITY |
42 |
|
(undefined by default) |
43 |
|
- silently move computation of some grid factors into separate loops to |
44 |
|
save some recomputations |
45 |
|
- add a hack for non-linear free surface |
46 |
|
- add a verification experiment to exp4 that tests this code |
47 |
|
o make gluemnc/xplodemnc work with (moron) non-GNU shell tools |
48 |
|
|
49 |
|
checkpoint63d (2011/10/25) |
50 |
|
o pkg/obcs + seaice-dynamics: |
51 |
|
- new implementation calls OBCS_APPLY_UVICE (in seaice_dynsolver.F) before |
52 |
|
solving for uIce,vIce and prevent SEAICE_LSR or SEAICE_EVP to modify OB |
53 |
|
values (consistent with an implicit method). |
54 |
|
- add option OBCS_UVICE_OLD (in OBCS_OPTIONS.h, defined by default) to |
55 |
|
return to older implementation (same results as before); note that none of |
56 |
|
the "experimental" OBCS_SEAICE options are implemented in the new version. |
57 |
|
- switch test exp. seaice_obcs to use new version (update output). |
58 |
|
o pkg/seaice: |
59 |
|
- add few stop (B-grid & CS-exch, B-grid & obcs, AUTODIFF & SEAICE_maskRHS) |
60 |
|
- add parenthesis in Water-Ice drag calculations (evp & lsr); |
61 |
|
- in seaice_lsr.F, simplify and fix AA3 setting in case sNx or sNy=1; |
62 |
|
and fix case SEAICE_waterDrag=0 (but unstable if no or little seaice) |
63 |
|
- update output of test exp lab_sea that uses evp or lsr (fwd+ad) |
64 |
|
and global_ocean.cs32x15 (icedyn & seaice fwd+ad). |
65 |
|
o pkg/layers: |
66 |
|
- Include potential density as new coordinate (Thanks to David Munday) |
67 |
|
o pkg/dic: |
68 |
|
- Remove subroutine CALC_PCO2_APPROX_CO3 from carbon_chem.F and add |
69 |
|
carbonate computation/output to CALC_PCO2_APPROX |
70 |
|
o pkg/generic_advdiff |
71 |
|
- make TAF generate code that vectorizes: |
72 |
|
gad_dst3fl_adv_*.F, gad_u3c4_impl_r.F |
73 |
|
o model/src: |
74 |
|
- improve vectorization also for adjoint code of |
75 |
|
solve_tridiagonal and solve_pentadiagonal. As this requires up to |
76 |
|
5 new 3D-fields, this code is only turned on for the combination of |
77 |
|
ALLOW_AUTODIFF_TAMC and TARGET_NEC_SX. Also remove error message |
78 |
|
for singular matrix check (only for TARGET_NEC_SX). |
79 |
|
- remove hard-wiring of CPP-flag ALLOW_SOLVERS_KLOOPINSIDE to |
80 |
|
ALLOW_AUTODIFF. This flag now needs to be set outside of |
81 |
|
solve_tri/pentadiagonal.F |
82 |
|
o pkg/exf: |
83 |
|
- fix previous modif in exf_readparms.F (revision 1.66: treatment of |
84 |
|
optional namelist EXF_NML_OBCS). |
85 |
|
- add many debug-print (to trace down where pkg/cal error msg comes from) |
86 |
|
o pkg/mom_vecinv/mom_vi_hdissip.F |
87 |
|
- rearrange loops so that vectorization is improved and in particular |
88 |
|
the adjoint routine vectorizes. The code is a little longer now, but I |
89 |
|
think easier to read. |
90 |
|
Note: CPP-flag MOM_VI_ORIGINAL_VISCA4 is not tested. |
91 |
|
o pkg/seaice: |
92 |
|
- allow to specify initial seaice velocity from files (uIceFile,vIceFile) |
93 |
|
o pkg/exch2: |
94 |
|
- alternative code (without tile x tile matrix) for cumulated (global) sum; |
95 |
|
used as default; keep using matrix version in adjustment.cs-32x32x1. |
96 |
|
o tools/build_options: |
97 |
|
- after merging some MPI and/or multi-threads specific optfiles so that |
98 |
|
the modified default optfile support both built (using genmake2 -mpi |
99 |
|
for the MPI built, using genmake2 -omp for OpenMP built), |
100 |
|
move now obsolete optfiles to tools/build_options/unsupported/ |
101 |
|
(add copy of email sent to support list in doc/optfiles_changes.txt) |
102 |
|
|
103 |
|
checkpoint63c (2011/09/28) |
104 |
|
o pkg/ebm: |
105 |
|
- fix ebm_zonalmean.F for several procs in Ydir (nPy > 1); update output of |
106 |
|
test exp. global_ocean_ebm (affects machine truncation, 12 digits for cg2d) |
107 |
|
o pkg/seaice (SItracer pickup) |
108 |
|
* Simple code for pickup of SItracer fields (#define ALLOW_SITRACER) |
109 |
|
To be done: |
110 |
|
- thorough testing of missing fields in S/R seaice_read_pickup |
111 |
|
- thorough matching between generic tracers and specific fields used |
112 |
|
(e.g. volume-weighted age, area-weighted age, salinity, ...) |
113 |
|
o pkg/grdchk: |
114 |
|
- all procs calculate and print finite-difference gradient (for testreport) |
115 |
|
- change description output used by testreport; |
116 |
|
- AD testing: also compare and report Finite-difference gradient; |
117 |
|
=> update all output_adm*.txt |
118 |
|
o test exp. global_ocean_ebm: |
119 |
|
- switch back useRealFreshWaterFlux (no problem anymore with run-off); |
120 |
|
and turn on exactConserve; update (fwd) output. |
121 |
|
o pkg/seaice (LSR): |
122 |
|
- print to STDOUT number of solver iterations at SEAICE_monFreq interval. |
123 |
|
- modify solver to deal cleanly with u/v switch at face edges (CS-grid): |
124 |
|
* identical results with non-CS topology (but save up to half of EXCH calls) |
125 |
|
* no much changes in cs-32 verification tests since no ice at face edges |
126 |
|
(differences due to convergence criteria, since nb of iterations changed) |
127 |
|
* expect some differences on lat-lon-cap grid. |
128 |
|
update FWD results of global_ocean.cs32x15 test exp. (icedyn & seaice) |
129 |
|
o pkg/exch2: |
130 |
|
- add new CPP option to disable Cumul-Sum matrix (to save storage when using |
131 |
|
many tiles); until alternative coding is added, just add a stop. |
132 |
|
o pkg/kpp: |
133 |
|
- move ptracer/salt correction to KPP diffusivity from calc_3d_diffusivity.F |
134 |
|
to S/R KPP_CALC_DIFF_PTR (+ add tracer-index to argument list). |
135 |
|
o optfiles: |
136 |
|
- merged few MPI specific (*+mpi*) and multi-threads specific (*+mth) optfiles |
137 |
|
into corresponding standard one, which now support different build: |
138 |
|
(when using "genmake2 -mpi" for MPI built, "genmake2 -omp" for OpenMP built) |
139 |
|
- add "darwin_amd64_gfortran" optfile (initially from Constantinos, tested by Gael) |
140 |
|
o pkg/obcs (seaice): |
141 |
|
- add missing EXCH for uIce,vIce after applying OBCS; |
142 |
|
- remove S/R EXF_SWAPFFIELDS_X/Y , disable S/R EXF_SET_OBCS_X/Y by adding |
143 |
|
argument = number of levels to S/R EXF_SWAPFFIELDS_XZ/YZ and S/R |
144 |
|
EXF_SET_OBCS_XZ/YZ ; update obcs_exf_load.F; |
145 |
|
- fix EXF_SET_OBCS_XZ/YZ for multi-threaded (assuming loaded arrays are shared) |
146 |
|
- verification exp. seaice_obcs: split domain in 2 tiles ; add SIZE.h_mpi |
147 |
|
and eedata.mth for MPI and mutlti-threaded testing. |
148 |
|
o pkg/flt: |
149 |
|
- make FLT EXCH working with multi-threads (and MPI + multi-threads); |
150 |
|
- use buffer for floats IO (trajectories & profiles); size set in FLT_SIZE.h; |
151 |
|
- fix pickup and trajectories + profiles output with multi-threads; |
152 |
|
- add parameter to select group of fields to write along trajectory and |
153 |
|
profile (default unchanged); |
154 |
|
- only call PORT_RAND* function when flt_noise <> 0; |
155 |
|
- add multi-threaded test for verification exp. flt_example. |
156 |
|
o eesupp/src: |
157 |
|
- fix S/R ALL_PROC_DIE for MPI+multi-threads. |
158 |
|
- USE_OMP_THREADING: check that number of threads (from eedata) matches |
159 |
|
actual number (from env.var. setting). |
160 |
|
o pkg/ebm: |
161 |
|
- fix multi-threading in ebm_area_t.F ; clean-up EBM.h header file. |
162 |
|
- add multi-threaded test for verification exp. global_ocean_ebm |
163 |
|
|
164 |
|
checkpoint63b (2011/08/29) |
165 |
|
o pkg/ebm: |
166 |
|
- fix sign of runoff; |
167 |
|
- update directly forcing fields (Qnet,EmPmR) to use standard |
168 |
|
external_forcing_surf.F (remove ebm_forcing_surf.F); |
169 |
|
- avoid loading forcing fields a 2nd time (1rst time from ini_forcing.F); |
170 |
|
- set either tauThetaClimRelax (in data) or tauThetaZonRelax (in data.ebm) |
171 |
|
and apply relaxation towards corresponding SST field (2-D map or ZonAver). |
172 |
|
- update global_ocean_ebm parameters file data and update output (fwd+ad). |
173 |
|
o verification/matrix_example: |
174 |
|
- split domain in 8 tiles (2x4); add SIZE.h_mpi and eedata.mth for |
175 |
|
MPI and mutlti-threaded testing. |
176 |
|
o pkg/matrix: |
177 |
|
- change name of S/R MATRIX_INIT to S/R MATRIX_INIT_VARIA ; |
178 |
|
- fix for multi-tiles (add bi,bj indices to exp/impMatrixCounter); |
179 |
|
- fix for multi-threaded (remove DATA statement); |
180 |
|
- use standard DIFFERENT_MULTIPLE function; |
181 |
|
- add MATRIX_OPTIONS.h |
182 |
|
o verification/matrix_example: |
183 |
|
To allow to split the domain in several tiles, change the domain size |
184 |
|
from 31x31 to 32x32 by adding 1 row (on Southern side) + 1 column (on |
185 |
|
Western side) of land points. |
186 |
|
o (*_)write_pickup: |
187 |
|
- fix few bugs in mypackage_read/write_pickup.F |
188 |
|
- in MDS_WR_METAFILES calls, replace "myTime" by a local copy into |
189 |
|
array of length 1 (was caught by compiler syntax-check). |
190 |
|
o pkg/diagnostics (velocity potential solver): |
191 |
|
- fix initialisation of main diagonal matrix (for overlap values) |
192 |
|
o tools/build_options: |
193 |
|
- towards single optfile for MPI and non-MPI built: merge few +mpi optfile |
194 |
|
into previously non MPI one (linux_am64 pgf77,open64,pathf90,ifort11) |
195 |
|
- linux_amd64_gfortran: merge +mpi_generic version into standard optfile: |
196 |
|
* move IEEE flags to DEVEL section; take IEEE flags from +mpi_generic ; |
197 |
|
* FAST case: for now, keep -O3 with ini_masks_etc.F in NOOPTFILES list; |
198 |
|
and use "-devel" testreport option for standard testing on baudelaire. |
199 |
|
o pkg/grdchk: |
200 |
|
- Assume that all gradient-check position are within the same tile |
201 |
|
corresponding to global indices (iGloTile,jGloTile) and that |
202 |
|
positions (i,j indices or nbeg) are local to this tile; |
203 |
|
- set tile local-indices and proc. number according to tile global-indices; |
204 |
|
- remove grdchkwhichproc from namelist (retired param); |
205 |
|
- send adjoint gradient to all procs so that all procs print it to STDOUT; |
206 |
|
- switch on MPI test for global_ocean.90x40x15 AD test experiment. |
207 |
|
o adding package bbl (Bottom Boundary Layer) |
208 |
|
description in MITgcm/pkg/bbl/bbl_description.tex |
209 |
|
example/test experiment in MITgcm_contrib/bbl |
210 |
|
o adjoint verifications |
211 |
|
- changed 5 experiments to multi-tiles |
212 |
|
* global_ocean.90x40x15 |
213 |
|
* obcs_ctrl |
214 |
|
* offline_exf_seaice |
215 |
|
* tutorial_dic_adjoffline |
216 |
|
* tutorial_tracer_adjsens |
217 |
|
changed reference output for those that crossed tile boundaries |
218 |
|
(changed choices of gradient check positions) |
219 |
|
- with same tiling, switch on MPI test for the last 4 (add SIZE.h_mpi) |
220 |
|
- temporary bug fix in grdchk_loc.F for obcs |
221 |
|
o genmake2 & testreport |
222 |
|
- add target "obj" (just for OBJFILES files) to Makefile; |
223 |
|
- add option to specify FEXTRAFLAGS on command line. |
224 |
|
o pkg/mypackage: |
225 |
|
- remove bi,bj arguments from argument list of S/R MYPACKAGE_CALC_RHS |
226 |
|
|
227 |
|
checkpoint63a (2011/08/02) |
228 |
|
o pkg/diagnostics (velocity potential): |
229 |
|
- fix velocity-potential calculation when using OBCS: balance net inflow |
230 |
|
at each level (new S/R OBCS_DIAG_BALANCE). |
231 |
|
- fix some of the problems with OBCS in stream-function calculation. |
232 |
|
o pkg/exf/exf_readparms.F |
233 |
|
- make treatment of optional namelists (EXF_NML_04, EXF_NML_SGRUNOFF, |
234 |
|
EXF_NML_OBCS) more consistent, so that you need not have these lists |
235 |
|
in data.exf when they are not used. |
236 |
|
o optim/optim_readdata.F: |
237 |
|
- add some output about obcs variables that might help debugging |
238 |
|
- add a check about nvarlength |
239 |
|
o pkg/ctrl: once again, fix computation of endrec in s/r ctrl_init_rec |
240 |
|
o pkg/diagnostics (velocity potential): |
241 |
|
- fix mask for OBCS (still problems in stream-function with OBCS); |
242 |
|
- add specific parameter (default = main code CG2D params) for solver; |
243 |
|
- in case of poor convergence, use solution corresponding to lowest residual. |
244 |
|
o pkg/seaice/seaice_growth.F: small changes that permit vectorizable |
245 |
|
adjoint code. In general statements like tmpscal=fld(i,j) are problematic |
246 |
|
and should be avoided where possible. |
247 |
|
o model/src (rStar): |
248 |
|
- include r* effect on reference-state (PhiRef) in totPhiHyd |
249 |
|
(so that it's really the potential anomaly at the cell center); |
250 |
|
affect solution only if z* with full pressure in EOS. |
251 |
|
- test-exp global_ocean.90x40x15: reduce viscA4 (was not stable) and |
252 |
|
update output (because of z* in totPhiHyd and viscA4 changes); |
253 |
|
- keep the previous diagnostics as "PHIHYDcR" (closer to potential anomaly |
254 |
|
@ constant r) until interpolation at constant r is implemented. |
255 |
|
o pkg/diagnostics |
256 |
|
- fix RCENTER diagnostics in case cell center is not at interface middle |
257 |
|
o add code for TEOS-10 (www.teos-10.org, McDougall et al. 2011). Use |
258 |
|
this eos with eosType = 'TEOS10', in data (PARM01). This eos implies |
259 |
|
that THETA and SALT are "conservative temperature" and "absolute |
260 |
|
salinity" |
261 |
|
o pkg/ctrl |
262 |
|
- fix computation of endrec in s/r ctrl_init_rec |
263 |
|
- ctrl_pack/unpack_*: remove unused variables and some obsolete code |
264 |
|
o pkg/seaice/seaice_growth.F |
265 |
|
- fix recomputation with FENTY_AREA_EXPANSION_CONTRACTION code, |
266 |
|
remove then obsolete STORE directives |
267 |
|
- fix STORE directives for tice and tices, so that only the |
268 |
|
appropriate part of the fields are stored (and not the entire |
269 |
|
field); update global_ocean.cs32x15.seaice AD output. |
270 |
|
- cosmetic changes: adjust CPP flags and indentation for better |
271 |
|
legibilty |
272 |
|
o pkg/autodiff: add mnc-output of adBottomDragFld (code contributed by |
273 |
|
Chris Wilson) |
274 |
|
o pkg/gmredi: |
275 |
|
Add Sub-Meso Eddies parameterisation from Baylor (for now, disabled with TAF), |
276 |
|
and test it in verification exp. "front_relax.bvp" (update output.bvp.txt) |
277 |
|
|
278 |
|
checkpoint63 (2011/07/11) |
279 |
|
o pkg/exch2: |
280 |
|
- rename+move: nTiles in W2_EXCH2_PARAMS.h --> exch2_nTiles in W2_EXCH2_TOPOLOGY.h |
281 |
|
- exch2 version of Cumul-Sum routine: S/R W2_SET_MAP_CUMSUM set up the |
282 |
|
(2*nTiles)x(nTiles) Cumul-Sum matrix and Master Proc collects all tiles X,Y |
283 |
|
increments, calculates Cumul-Sum @ tile origin and sends results to each tile. |
284 |
|
o pkg/diagnostics: |
285 |
|
- deal with pairs of post-processed diags which are computed together; in |
286 |
|
particular, re-order list of diags within 1 file to save 2nd computation. |
287 |
|
- add S/R to compute velocity stream-function (called after velocity-potential |
288 |
|
calculation) from divergence free transport -> new diagnostics "PsiVEL". |
289 |
|
- add parameter to select grid-point location where PsiVEL == 0. |
290 |
|
o eesupp/src: |
291 |
|
new S/R to calculate cumulated sum on tiled array, corner grid-cell location |
292 |
|
(specific S/R needed with exch2 topology - for now, just print an error msg) |
293 |
|
o pkg/shelfice |
294 |
|
- Implement friction velocity-dependent transfer coefficients following |
295 |
|
Holland and Jenkins, JPO, 1999 |
296 |
|
Original code by M. Losch with small modifs. |
297 |
|
Enable with CPP option SHI_ALLOW_GAMMAFRICT |
298 |
|
and runtime flag SHELFICEuseGammaFrict |
299 |
|
o verification lab_sea & global_ocean.cs32x15: |
300 |
|
- use tools/xmakedepend (instead of cyrus_makedepend, in genmake_local): |
301 |
|
does not miss some dependencies (but is slower) and also works with many files |
302 |
|
(default makedepend limitation for AD built). |
303 |
|
- use cyrus_makedepend in adjustment.128x64x1 (so that it still tested). |
304 |
|
o pkg/diagnostics: |
305 |
|
- add run-time parameter "useMissingValue" (def=False) to fill land-point |
306 |
|
(i.e., where mask=0) with MissingValue ; used only in MNC output file. |
307 |
|
This replaces CPP-option DIAGNOSTICS_MISSING_VALUE. |
308 |
|
- Extend the use of "missing_value" setting from data.diagnostics for the case |
309 |
|
of counter-diagnostics (where fraction-weight is zero). This applies both to |
310 |
|
MNC and MDS output files. |
311 |
|
o pkg/seaice/seaice_growth.F: |
312 |
|
- add 'a priori' cap of latent heat flux based on actual mass considerations |
313 |
|
when SEAICE_ADD_SUBLIMATION_TO_FWBUDGET is used in conjunction with the |
314 |
|
non-legacy seaice_solve4temp. |
315 |
|
- add diagnostic of the residual sublimation freshwater flux (the freshwater |
316 |
|
flux taken from the ocean if the total actual latent heat fluxes exceed |
317 |
|
the maximum) |
318 |
|
- remove the CPP switch around the *_FWbySublim unit change and move diagnostic |
319 |
|
fill for SIatmQnt, SIfwSubl, SIatmFW to the end of seaice_growth.F. |
320 |
|
- add diagnostic of the actual sublimation freshwater flux (that is 0. ifndef |
321 |
|
SEAICE_ADD_SUBLIMATION_TO_FWBUDGET) and of the latent heat flux (evap+sublim). |
322 |
|
- bug fix : frWtrAtm needs a_FWbySublim ifdef SEAICE_ADD_SUBLIMATION_TO_FWBUDGET. |
323 |
|
- bug fix : missing part to Martin's r_FWbySublim pathological case fix. |
324 |
|
o verification 1D_ocean_ice_column: |
325 |
|
made model domain truly 1x1x23 instead of 5x5x23 with only center cell (3,3,:) wet |
326 |
|
update results |
327 |
|
o pkg/diagnostics: |
328 |
|
- Implement setting of "Post-Processed" diagnostics (corresponding to gdiag(5)='P') |
329 |
|
which are not filled-up but computed from other diags ; In this case, the mate |
330 |
|
diag points to the primary (filled-up) diag to used for post processing. |
331 |
|
- Update Velocity-potential computation using post-processed diag framework. |
332 |
|
==> Change the ordering of fields and time-periods in MDS output file |
333 |
|
when using periodic averaging: now writes one time record with the full set |
334 |
|
of fields for this time period, then the next time reccord (it used to be |
335 |
|
one field with the full set of time records then the next field). |
336 |
|
Note: this is now similar to NetCDF (MNC) output file. <== |
337 |
|
|
338 |
|
checkpoint62z (2011/06/21) |
339 |
o adjoint monitor: |
o adjoint monitor: |
340 |
- Add adjoint monitor output for packages exf, seaice |
- Add adjoint monitor output for packages exf, seaice |
341 |
and update relevant testreport output |
and update relevant testreport output |