| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o verification/testreport: |
| 8 |
|
- add option '-tlm' for Tangent-Linear test (-adm,-oad,-tlm are all exclusive); |
| 9 |
|
- can also use compressed (gzip) reference output file; |
| 10 |
|
- change format (in pkg/grdchk) of TLM output (+ add forward gradient) |
| 11 |
|
to be easier to collect by testreport. |
| 12 |
|
- update existing TLM output (half out-of-date by more than 3 yrs). |
| 13 |
|
- report NETCDF & LAPACK settings (from genmake2) in output dir. |
| 14 |
|
- report run-time in each individual summary (previously only for AD test); |
| 15 |
|
for TLM test, add also TAF version. |
| 16 |
|
- add daily TLM test on (old) aces cluster + weekly test on baudelaire. |
| 17 |
|
|
| 18 |
|
checkpoint63r (2012/08/15) |
| 19 |
|
o tools/genmake2: |
| 20 |
|
- fix Makefile for target ftlall (+ update files to clean after make ftlall) |
| 21 |
|
o pkg/ecco/ecco_cost_final.F : |
| 22 |
|
- fix fc print statements (to stdout and costfunction0000) so |
| 23 |
|
that the printed value is always after the global sum. |
| 24 |
|
o pkg/seaice/seaice_dynsolver.F : |
| 25 |
|
- fix ice strength assymptote at SEAICE_area_max<1 |
| 26 |
|
(no change for the standard case of SEAICE_area_max=1) |
| 27 |
|
o some cleaning: |
| 28 |
|
- pkg source files: ensure that all *.F files from each pkg included the |
| 29 |
|
corresponding ${PKG}_OPTIONS.h file (except stand-alone src file which |
| 30 |
|
only includes CPP_EEOPTIONS.h); fix S/R CTRL_SMOOTH ; update AD experiment |
| 31 |
|
code_ad/*.F src files; add new option-file in pkg/admtlm + corresponding |
| 32 |
|
includes; |
| 33 |
|
- model/src: add PACKAGES_CONFIG.h wherever ALLOW_AUTODIFF[_TAMC] is used. |
| 34 |
|
- ALLOW_DIFFKR_CONTROL: add a stop in S/R CTRL_CHECK in case ALLOW_3D_DIFFKR |
| 35 |
|
is undef; also define /DYNVARS_DIFFKR/ common block in DYNVARS.h when |
| 36 |
|
ALLOW_DIFFKR_CONTROL is defined, to allow to compile and reach this stop. |
| 37 |
|
- update option-file CPP_OPTIONS.h of all AD experiments: |
| 38 |
|
* no longer include PACKAGES_CONFIG.h ; |
| 39 |
|
* includes ECCO_CPPOPTIONS.h if PACKAGES_CONFIG.h has been included ; |
| 40 |
|
* def/undef ALLOW_3D_DIFFKR independently of ALLOW_DIFFKR_CONTROL (before |
| 41 |
|
including ECCO_CPPOPTIONS.h). |
| 42 |
|
o pkg/cost, pkg/ctrl, pkg/ecco option-file: |
| 43 |
|
- clean-up and split content of old pkg/ecco/ECCO_CPPOPTIONS.h file into |
| 44 |
|
separated pkg option-file (done according to old doc from ECCO_CPPOPTIONS.h, |
| 45 |
|
not necessary accurate). |
| 46 |
|
- add a template for option-file ECCO_CPPOPTIONS.h, mostly empty, in |
| 47 |
|
pkg/autodiff, i.e., the 1rst pkg to compile for adjoint built, with lines to |
| 48 |
|
deliberately break compilation (a remainder that it needs to be customized) |
| 49 |
|
- in pkg/ecco, includes ECCO_OPTIONS.h instead of COST_OPTIONS.h |
| 50 |
|
- rename COST_CPPOPTIONS.h to COST_OPTIONS.h |
| 51 |
|
- rename CTRL_CPPOPTIONS.h to CTRL_OPTIONS.h |
| 52 |
|
o tools/genamke2: |
| 53 |
|
- check for lapack. |
| 54 |
|
o pkg/cost and pkg/ecco : attempt to sort out some of the limiting intricacies. |
| 55 |
|
- pkg/cost.cost_final.F : allow for use of pkg/ecco and pkg/cost cost functions |
| 56 |
|
jointly. The exclusive ifdefs were removed, to add costs terms (when defined) |
| 57 |
|
from both packages. |
| 58 |
|
- the fc global sum in pkg/ecco/ecco_cost_final.F was removed accordingly. |
| 59 |
|
- for objf_atl we use either the pkg/ecco or the pkg/cost term depending on CPPs. |
| 60 |
|
- pkg/ecco and model/src : add CPP brackets around pkg/cost and pkg/ctrl includes |
| 61 |
|
of "CTRL_SIZE.h" "ctrl.h" "ctrl_dummy.h" "optim.h" / "cost.h" accordingly. |
| 62 |
|
o pkg/cost : add comments to explain special and/or untested setups. |
| 63 |
|
o pkg/exf: |
| 64 |
|
- use EXF_OPTIONS.h to set pkg/exf options except if done in ECCO_CPPOPTIONS.h |
| 65 |
|
(previously, ALLOW_AUTODIFF_TAMC was controling this switch). |
| 66 |
|
o verification/*/code_ad: |
| 67 |
|
mimimal cleaning of AD experiments ECCO_CPPOPTIONS.h option-file: |
| 68 |
|
- reccord when this file is included (define ECCO_CPPOPTIONS_H) |
| 69 |
|
- remove the most obvious obsolete features (e.g., #define ALLOW_COST not |
| 70 |
|
allowed since Oct 2003 in any option-file except in PACKAGES_CONFIG.h) |
| 71 |
|
o pkg/seaice : |
| 72 |
|
- added possibility to change ice strength/PRESS0 relationship to |
| 73 |
|
HEFF, according to the following run time parameters |
| 74 |
|
SEAICEpresPow0 :: HEFF exponent for ice strength below SEAICEpresH0 (integer, >0) |
| 75 |
|
SEAICEpresPow1 :: HEFF exponent for ice strength above SEAICEpresH0 (integer, >0) |
| 76 |
|
SEAICEpresH0 :: HEFF threshold for ice strength (m) |
| 77 |
|
o pkg/ggl90 : |
| 78 |
|
- added run time flag mxlSurfFlag to include the code that ensure |
| 79 |
|
mixing between first and second level (previously included as a comment) |
| 80 |
|
o verification/1D_ocean_ice_column/input_ad/data.autodiff : |
| 81 |
|
- add test of newly added dumpAdByRec=T option where |
| 82 |
|
we write all records to one file per variable. |
| 83 |
|
o verification/global_ocean.cs32x15/code_ad/CPP_OPTIONS.h : |
| 84 |
|
- switch to #define ALLOW_BALANCE_RELAX and ALLOW_BALANCE_FLUXES |
| 85 |
|
to check that this code is ok with taf and compiles in adjoint. |
| 86 |
|
o pkg/autodiff : |
| 87 |
|
- added run time param dumpAdByRec : ad dump format switch. |
| 88 |
|
If true then write all records to one file per variable. |
| 89 |
|
Else write one file per record (remains the default). |
| 90 |
|
- revise seaice_ad_dump.F and exf_adjoint_snapshots_ad.F accordingly. |
| 91 |
|
o pkg/seaice: |
| 92 |
|
- Added SEAICEuseTILT run time switch : if set to |
| 93 |
|
false then omit the tilt term in seaice dynamics. |
| 94 |
|
- Fix free drift calculation. |
| 95 |
|
o pkg/shelfice: |
| 96 |
|
When initializing SHELFICEDragQuadratic, set default to be |
| 97 |
|
consistent with shiCdrag in case SHELFICEuseGammaFrict is used, |
| 98 |
|
but retain option to set it independently |
| 99 |
|
(inconsistenty spotted by Veronique Dansereau) |
| 100 |
|
o eesupp/inc: |
| 101 |
|
- in EEPARAMS.h, add real-type constant for some frequently used simple number |
| 102 |
|
(0,1,2,1/2), with name explicitly referring to RS/RL type. |
| 103 |
|
o model/src: |
| 104 |
|
- move balancing of relaxation terms to new S/R balance_relax.F. |
| 105 |
|
o pkg/autodiff: |
| 106 |
|
re-write active I/O high-level (control) S/R: |
| 107 |
|
- rename S/R ACTIVE_READ/WRITE_RL to ACTIVE_READ/WRITE_3D_RL |
| 108 |
|
- add 1 more argument (useCurrentDir) to replace the "_LOC" 2nd set of S/R. |
| 109 |
|
- use standard pkg/mdsio I/O routines (MDS_READ/WRITE_SEC_XZ/YZ and |
| 110 |
|
MDS_READ/WRITE_FIELD); change loops to multi-threads. |
| 111 |
|
- update and merge active_file_control_slice.flow into active_file_control.flow |
| 112 |
|
- update the calls in lower level S/R (active_file.F, active_file_g.F and |
| 113 |
|
active_file_ad.F) keeping the argument list and function unchanged. |
| 114 |
|
- also update lower-level "_LOC" version (call same S/R with "useCurrentDir |
| 115 |
|
=.TRUE.") and remove high-level (control) _loc version (not called anymore). |
| 116 |
|
- new low-level active I/O routines, one for each type (RL/RS), for active var |
| 117 |
|
of various (XY,XZ,YZ) shape and level number (arguments: arrShape, myNr); |
| 118 |
|
fix ctrl_init_wet.F & ecco_cost_weights.F by calling the new S/R with the |
| 119 |
|
right type. |
| 120 |
|
o autodiff and RS=real*4: |
| 121 |
|
- call the right type (_RS instead of _RL) of S/R in few output S/R (pkg/exf |
| 122 |
|
ad-monitor and snapshots); disable arrType='RS' in mdsio_gl*.F ; |
| 123 |
|
- change argument type of 3-D mask argument to _RS to match type in the call |
| 124 |
|
(in seaice_cost*.F, ctrl_*_gen.F and *ctrl_bound.F); to be continued. |
| 125 |
|
o pkg/thsice: |
| 126 |
|
- re-anable balanceEmPmR when useSEAICE=T with useThSIce=T. |
| 127 |
|
- option to correct ocean fresh-water flux for imbalance of Atmos+Land |
| 128 |
|
fresh-water flux: uniform correction or proportional to local precip. |
| 129 |
|
o pkg/ctrl: |
| 130 |
|
- new header file CTRL_SIZE.h for parameters maxCtrlArr2D, maxCtrlArr3D |
| 131 |
|
- redefine CPP option ALLOW_BOTTOMDRAG_CONTROL to |
| 132 |
|
ALLOW_BOTTOMDRAG_CONTROL_NONGENERIC |
| 133 |
|
- verification/hs94.1x64x5/ tests ALLOW_GENARR{2D,3D}_CONTROL |
| 134 |
|
|
| 135 |
|
checkpoint63q (2012/07/30) |
| 136 |
|
o pkg/autodiff: |
| 137 |
|
note about monitor and snap-shot of ad-variables: |
| 138 |
|
a) need to apply ADEXCH before monitor or output of ad-vars to include |
| 139 |
|
overlap contributions (was done for adjDump but missing for monitor). |
| 140 |
|
b) since ADEXCH implies adding overlap values to interior array, machine |
| 141 |
|
truncation error is expected when adding an ADEXCH call in the middle of |
| 142 |
|
the time-stepping (although mathematically identical). |
| 143 |
|
modifs: |
| 144 |
|
- fix some incorrect type (RL/RS, UV/scalar) in ADEXCH calls and ad-vars |
| 145 |
|
declaration (adEddyPsiX,adEddyPsiY in adcommon.h). |
| 146 |
|
- add new run-time params (dumpAdVarExch & mon_AdVarExch) to select how to |
| 147 |
|
apply ADEXCH before monitor or snap-shot of ad-variables: |
| 148 |
|
=0 : no ADEXCH ; =1 : do ADEXCH ; =2 : apply ADEXCH on a local copy. |
| 149 |
|
- implement new method (copy+adexch) for snap-shot and monitor of main model |
| 150 |
|
ad-vars (initially not used by default -> output unchanged). |
| 151 |
|
- change default (set to 2) to use new method and update few output_adm.txt |
| 152 |
|
with accurate ad-monitor (global_ocean.90x40x15, tutorial_global_oce_biogeo |
| 153 |
|
and both tutorial_tracer_adjsens tests). |
| 154 |
|
- switch 2 global_ocean.cs32x15 AD-tests to AB-2 (seaice_dynmix & thsice) |
| 155 |
|
and turn off momDissip_In_AB in the 2 other (needed for stable AB-3). |
| 156 |
|
o pkg/ctrl: |
| 157 |
|
enable the use of a variable number (maxCtrlArr2D, maxCtrlArr3D) of |
| 158 |
|
generic control variables via |
| 159 |
|
#define ALLOW_GENARR2D_CONTROL and/or #define ALLOW_GENARR3D_CONTROL |
| 160 |
|
Only required customization should be S/R ctrl_map_ini_genarr.F |
| 161 |
|
(routine in repository contains example) |
| 162 |
|
o pkg/exf: |
| 163 |
|
- fix and refine checking of surface-relaxation setting when using EXF |
| 164 |
|
o verification (adjoint exp): |
| 165 |
|
- reduce length of simulation by half for experiment global_ocean.cs32x15 |
| 166 |
|
(std, seaice & seaice_dynmix) and exp tutorial_dic_adjoffline; update output. |
| 167 |
|
o pkg/mom_common (mom_u/v_bottomdrag), pkg/shelfice (shelfice_u/v_drag) |
| 168 |
|
- replace if statements KE.ne.0 with KE.gt.0 to avoid sqrt of negative numbers |
| 169 |
|
of KE that can occur with aggressive optimization |
| 170 |
|
+ put quadratic bottom drag calculation in separated loops (only done if |
| 171 |
|
using bottomDragQuadratic) |
| 172 |
|
o model/src: CG2D_NSA |
| 173 |
|
- clean-up: stop if cg2dNormaliseRHS=T and AUTODIFF |
| 174 |
|
and replace GLOBAL_SUM calls with GLOBAL_SUM_TILE. |
| 175 |
|
- in test-exp that uses CG2D_NSA (bottom_ctrl & tutorial_tracer_adjsens), |
| 176 |
|
change cg2d residual target to be expressed in Wunits (=> do not normalise |
| 177 |
|
RHS ; this fix global_max Pb with TAF); update output. |
| 178 |
|
o verification/tutorial_tracer_adjsens: |
| 179 |
|
improve parameter consistency (and get rid of warnings): |
| 180 |
|
- use deltaTFreeSurf=deltaTtracer (better with NonLin FreeSurf); |
| 181 |
|
- switch Temp Adv Scheme from 81 to 80 (no limiter since theta < 0 is physical) |
| 182 |
|
- use staggerTimeStep (for stability) in som81; use default rotation period. |
| 183 |
|
o pkg/exch2: |
| 184 |
|
- with empty facet: fix compact-format (W2_mapIO=1) definition (was previously |
| 185 |
|
reset to 0 = 1 long line in X) |
| 186 |
|
o pkg/mdsio: |
| 187 |
|
- fix reading of dimensions from meta file for large-size domain |
| 188 |
|
o model/src: |
| 189 |
|
- fix saltIsActiveTr setting for IdealGas eosType (eosType is 6 character long); |
| 190 |
|
- to refine CS-grid check, add logical flag "hasWetCSCorners": true if using |
| 191 |
|
Cubed-Sphere Exch with CS-corners inside the domain; |
| 192 |
|
make new S/R INI_GLOBAL_DOMAIN from code in ini_linear_phisurf.F to calculate |
| 193 |
|
globalArea and to set hasWetCSCorners. |
| 194 |
|
- add warning if usingCurvilinearGrid and momAdvection in flux-form (missing |
| 195 |
|
metric terms); stop if hasWetCSCorners=T and momAdvection in flux-form. |
| 196 |
|
- re-activate the stop if useCDscheme and hasWetCSCorners=T (previously if |
| 197 |
|
useCDscheme and useCubedSphereExchange). |
| 198 |
|
o pkg/seaice/seaice_growth.F : |
| 199 |
|
- compute the SEAICE_DISABLE_HEATCONSFIX term accordingly |
| 200 |
|
for the (.NOT.useRealFreshWaterFlux).OR.(nonlinFreeSurf.EQ.0) |
| 201 |
|
when the previous code was not correct. |
| 202 |
|
- in all cases diagnoze the boundary condition mismatch to SIaaflux |
| 203 |
|
- in the case of useRealFreshWaterFlux.AND.(nonlinFreeSurf.GT.0) add |
| 204 |
|
that term to QNET (essentially to set the melt/freeze water to 0degC). |
| 205 |
|
when SEAICEheatConsFix is true, so that the ocean+ice budget heat is closed. |
| 206 |
|
o verification/global_ocean.cs32x15/input.seaice : |
| 207 |
|
- add SEAICEheatConsFix in data.seaice, ensuring closed ocean+ice heat budget. |
| 208 |
|
- add diagnostics (commented out) for budget test using my matlab codes (gcmfaces). |
| 209 |
|
- update results (output.seaice.txt) accordingly. |
| 210 |
|
o verification/lab_sea/input/data.seaice : |
| 211 |
|
- remove SEAICEheatConsFix, which is now be tested |
| 212 |
|
in global_ocean.cs32x15. The presently coded fix |
| 213 |
|
is only consistent with real fresh water + nlfs. |
| 214 |
|
- update results (output.txt) accordingly. |
| 215 |
|
o pkg/salt_plume: added diagnotic for salt tendency (oceSPtnd). |
| 216 |
|
o AD_CONFIG.h file: |
| 217 |
|
- in genmake2: remove "#undef ALLOW_ECCO_OPTIMIZATION" from AD_CONFIG.h; |
| 218 |
|
- remove #include "AD_CONFIG.h" from all included OPTIONS files and |
| 219 |
|
add it explicitly in each fortran src file where it is needed. |
| 220 |
|
motivation: header file AD_CONFIG.h is generated/modified after "make depend" |
| 221 |
|
thus introducing potentially wrong dependency; therefore, better to include |
| 222 |
|
it only where it's really needed, and after any other included OPTIONS file |
| 223 |
|
(less tempting to bring conditional includes according to ALLOW_ADJOINT_RUN |
| 224 |
|
or ALLOW_TANGENTLINEAR_RUN). |
| 225 |
|
|
| 226 |
|
checkpoint63p (2012/07/06) |
| 227 |
|
o pkg/autodiff: |
| 228 |
|
- rename S/R INI_AUTODIFF to AUTODIFF_INIT_VARIA ; |
| 229 |
|
- create new header file "AUTODIFF_PARAMS.h" for pkg/autodiff parameters |
| 230 |
|
which are now read from new parameter file "data.autodiff" (inAdExact |
| 231 |
|
previously in "data", use{KPP/GMRedi/SEAICE}inAdMode previously in data.pkg) |
| 232 |
|
- add file "data.autodiff" in all verification/*/input_ad/ directories. |
| 233 |
|
- add run-time parameter to turn On/Off SEAICEuseFREEDRIFT in adjoint/forward |
| 234 |
|
mode (used in global_ocean.cs32x15.seaice_dynmix test exp.) |
| 235 |
|
o verification/tutorial_tracer_adjsens/ |
| 236 |
|
- add extra sub-verification which uses "SOM" advection scheme |
| 237 |
|
(2nd Order-Moment Advection Scheme, Prather Limiter) with |
| 238 |
|
#define GAD_ALLOW_TS_SOM_ADV in GAD_OPTIONS.h |
| 239 |
|
and in data: |
| 240 |
|
multiDimAdvection=.TRUE., |
| 241 |
|
tempAdvScheme=81, |
| 242 |
|
saltAdvScheme=81, |
| 243 |
|
o model/src: |
| 244 |
|
- introduce ALLOW_BALANCE_RELAX which allow the removal |
| 245 |
|
of the global mean of relaxation terms by setting |
| 246 |
|
balanceThetaClimRelax and balanceSaltClimRelax |
| 247 |
|
- disable balanceEmPmR and balanceQnet in the case when useSeaice. |
| 248 |
|
This case is now treated appropriately in seaice_growth.F |
| 249 |
|
o pkg/seaice: |
| 250 |
|
- do balanceEmPmR and balanceQnet based on the net fluxes |
| 251 |
|
(SItflux) that enter the ocean+ice system rather than QNET |
| 252 |
|
and EMPMR. Works best in real fresh water. |
| 253 |
|
- SIatmFW, SIatmQnet, and SItflux computations had to be moved |
| 254 |
|
out of the diagnotics bloc, and stored in global arrays. |
| 255 |
|
|
| 256 |
|
checkpoint63o (2012/06/29) |
| 257 |
|
o vertical mixing schemes: |
| 258 |
|
- remove unused S/R CALC_BUOYANCY (left from branch-atmos merging) |
| 259 |
|
- replace computation of density (in Function STATE1, opps_calc.F) by a call |
| 260 |
|
to S/R FIND_RHO_SCALAR |
| 261 |
|
- pass sigmaR to S/R PP81_CALC, MY82_CALC (but not yet used) and to GGL90_CALC |
| 262 |
|
and use it for N^2 calculation in ggl90_calc.F (save 2 density computations); |
| 263 |
|
affects output (vermix.ggl90) at truncation level. |
| 264 |
|
o testreport: |
| 265 |
|
- change adjustment of SIZE.h for mpi+mth testing: preserve enough tiles |
| 266 |
|
(nSx,nSy) for the number of threads (from eedata.mth) |
| 267 |
|
o model/src: |
| 268 |
|
- remove 2nd set of calls at the top of initialise_varia.F (Adjoint part) |
| 269 |
|
related to Non_Lin Free-Surf or Bottom-Control (but not longer necessary). |
| 270 |
o verification/tutorial_tracer_adjsens/ |
o verification/tutorial_tracer_adjsens/ |
| 271 |
- compile with ALLOW_CG2D_NSA |
- compile with ALLOW_CG2D_NSA |
| 272 |
- run with nonlinFreeSurf=3 |
- run with nonlinFreeSurf=3 |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch