| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
o ECCO-CLIVAR merge: |
o pkg/autodiff: |
| 8 |
1. seaice AREA relaxation |
- rename S/R INI_AUTODIFF to AUTODIFF_INIT_VARIA ; |
| 9 |
|
- create new header file "AUTODIFF_PARAMS.h" for pkg/autodiff parameters |
| 10 |
|
which are now read from new parameter file "data.autodiff" (inAdExact |
| 11 |
|
previously in "data", use{KPP/GMRedi/SEAICE}inAdMode previously in data.pkg) |
| 12 |
|
- add file "data.autodiff" in all verification/*/input_ad/ directories. |
| 13 |
|
- add run-time parameter to turn On/Off SEAICEuseFREEDRIFT in adjoint/forward |
| 14 |
|
mode (used in global_ocean.cs32x15.seaice_dynmix test exp.) |
| 15 |
|
o verification/tutorial_tracer_adjsens/ |
| 16 |
|
- add extra sub-verification which uses "SOM" advection scheme |
| 17 |
|
(2nd Order-Moment Advection Scheme, Prather Limiter) with |
| 18 |
|
#define GAD_ALLOW_TS_SOM_ADV in GAD_OPTIONS.h |
| 19 |
|
and in data: |
| 20 |
|
multiDimAdvection=.TRUE., |
| 21 |
|
tempAdvScheme=81, |
| 22 |
|
saltAdvScheme=81, |
| 23 |
|
|
| 24 |
|
checkpoint63o (2012/06/29) |
| 25 |
|
o vertical mixing schemes: |
| 26 |
|
- remove unused S/R CALC_BUOYANCY (left from branch-atmos merging) |
| 27 |
|
- replace computation of density (in Function STATE1, opps_calc.F) by a call |
| 28 |
|
to S/R FIND_RHO_SCALAR |
| 29 |
|
- pass sigmaR to S/R PP81_CALC, MY82_CALC (but not yet used) and to GGL90_CALC |
| 30 |
|
and use it for N^2 calculation in ggl90_calc.F (save 2 density computations); |
| 31 |
|
affects output (vermix.ggl90) at truncation level. |
| 32 |
|
o testreport: |
| 33 |
|
- change adjustment of SIZE.h for mpi+mth testing: preserve enough tiles |
| 34 |
|
(nSx,nSy) for the number of threads (from eedata.mth) |
| 35 |
|
o model/src: |
| 36 |
|
- remove 2nd set of calls at the top of initialise_varia.F (Adjoint part) |
| 37 |
|
related to Non_Lin Free-Surf or Bottom-Control (but not longer necessary). |
| 38 |
|
o verification/tutorial_tracer_adjsens/ |
| 39 |
|
- compile with ALLOW_CG2D_NSA |
| 40 |
|
- run with nonlinFreeSurf=3 |
| 41 |
|
- update reference output_adm.txt |
| 42 |
|
o tools/genmake2: |
| 43 |
|
- add "-DHAVE_NETCDF" to the DEFINES list when NetCDF is available ; |
| 44 |
|
refine usePROFILES switching off (only when NetCDF is not available). |
| 45 |
|
o verification (atmospheric set-up): |
| 46 |
|
- change eosType='IDEALGAS' to eosType='IDEALG' to match variable declaration |
| 47 |
|
o pkg/profiles : refine profiles and files handling. |
| 48 |
|
o model/src and pkg/ecco : introduce run-time switch usePROFILES. |
| 49 |
|
o model/src/the_main_loop.F, forward_step.F and pkg/exf/EXF_OPTIONS.h : |
| 50 |
|
- changes to use model/src/the_main_loop.F rather than the ecco version |
| 51 |
|
in ecco type adjoint runs. |
| 52 |
|
o model/src: |
| 53 |
|
- move computation of recip of grid distance & area from INI_MASKS_ETC to INI_GRID |
| 54 |
|
- move setting of h0Fac[C,W,S] from INI_LINEAR_PHISURF to INI_MASKS_ETC |
| 55 |
|
- update_masks_etc.F: rename SMOOTH*_R4,R8 function to the corresponding type (RS,RL) |
| 56 |
|
o model/src: |
| 57 |
|
- fix S/R UPDATE_CG2D for ALLOW_DEPTH_CONTROL case (was empty before); |
| 58 |
|
update output of test experiment bottom_ctrl_5x5. |
| 59 |
|
o pkg/generic_advdiff: |
| 60 |
|
- implement vertical bi-harmonic diffusion (for now, only for T & S and |
| 61 |
|
only explicitly). |
| 62 |
|
o verification/bottom_ctrl_5x5/ |
| 63 |
|
update reference output (results changed quite some time ago, |
| 64 |
|
not sure when; but new reference results look ok in terms of grdchk) |
| 65 |
|
o verification/1D_ocean_ice_column/ |
| 66 |
|
For verification purposes, the exact adjoint rather than an approx. |
| 67 |
|
should be used (e.g., to comare against F.D. or TLM). |
| 68 |
|
removed autodiff_inadmode*_ad.F routines in code_ad/ |
| 69 |
|
Update results/ accordingly. |
| 70 |
|
o pkg/cal: |
| 71 |
|
- change cal_getdate.F so that output date is function of input time only |
| 72 |
|
(less ambiguous, not mixed up with input iteration number). |
| 73 |
|
- fix calendarDumps for non-zero phase-lag (S/R CAL_TIME2DUMP). |
| 74 |
|
o pkg/diagnostics: |
| 75 |
|
- implement calendarDumps also for stats-diags |
| 76 |
|
- implement calendarDumps also for snap-shot output (negative freq) |
| 77 |
|
- for now, ignores calendarDumps if non-zero phase-lag. |
| 78 |
|
|
| 79 |
|
checkpoint63n (2012/06/04) |
| 80 |
|
o model/src: |
| 81 |
|
- recompute rSurfW & S to ensure consistent column thickness (needed for rStar) |
| 82 |
|
between Sum_k(hFac*drF) and rSurf-rLow, for now only when using pkg/shelfice; |
| 83 |
|
- use h0FacC, W & S to compute hFac_surfC, W & S in calc_surf_dr.F (more |
| 84 |
|
similar to calc_r_star.F); update results of experiment internal_wave and |
| 85 |
|
tutorial_global_oce_in_p (affected @ truncation level). |
| 86 |
|
o pkg/shelfice: |
| 87 |
|
- add parameterisation of advective-diffusive flux into the ice shelf |
| 88 |
|
according to Holland and Jenkins (1999), eq.22-33, as suggested by Jan De Ryd |
| 89 |
|
- simply eps3/3a computation |
| 90 |
|
o eesupp: |
| 91 |
|
- add new CPP option (DISCONNECTED_TILES) to disconnect tiles (no exchange |
| 92 |
|
between tiles, just fill-in edges assuming locally periodic subdomain) and |
| 93 |
|
add corresponding exchange subroutines (EXCH0); |
| 94 |
|
- new CG-solver version (_EX0) for disconnected-tiles special case. |
| 95 |
|
o model/src (CG-solvers): |
| 96 |
|
- modify Single-Reduction version (CG2D_SR) in case cg2dMaxIters is reached |
| 97 |
|
(similar final residual output and max-Iter number as with standard version); |
| 98 |
|
- allows to store and use lowest-residual solution (standard & _SR versions); |
| 99 |
|
- cleaning up (remove old, commented out code, LETS_MAKE_JAM ...). |
| 100 |
|
- by default, also compile CG2D_SR version (define ALLOW_SRCG in CPP_OPTIONS.h) |
| 101 |
|
o eesup/src: |
| 102 |
|
- global_vec_sum.F cleaning: add a BARRIER ; update comments ; remove ndef |
| 103 |
|
ALWAYS_USE_MPI (no longer needed after fixing usingMPI settings - March 2012) |
| 104 |
|
o pkg/salt_plume added to lab_sea/input_ad.noseaicedyn |
| 105 |
|
|
| 106 |
|
checkpoint63m (2012/04/27) |
| 107 |
|
o pkg/shelfice: |
| 108 |
|
- account for partical cells in surface level below ice-shelf for hydrostatic |
| 109 |
|
pressure calculion (set uniformFreeSurfLev to F); update isomip experiments. |
| 110 |
|
o ad test experiments: |
| 111 |
|
- update and rename to "ad_optfile.local" local AD-optfile specific to a |
| 112 |
|
particular adjoint verification experiment (in verification/*/code_ad/) |
| 113 |
|
o pkg/seaice & pkg/exf: |
| 114 |
|
- Part two of ECCO-CLIVAR code merges. |
| 115 |
|
Completes merge of seaice area relax. code. |
| 116 |
|
Requires unified CPP option EXF_ALLOW_SEAICE_RELAX |
| 117 |
|
to be set in EXF_OPTIONS.h since this pkg controls its I/O |
| 118 |
|
(and avoids second CPP option in SEAICE_OPTIONS.h) |
| 119 |
|
- change default value of costIceStart & costIceEnd to a valid date. |
| 120 |
|
- add snowPrecip to HSNOW in seaice_growth |
| 121 |
|
o pkg/ptracers: |
| 122 |
|
- move ptracers summary and setup check into separated S/R ; add a check for |
| 123 |
|
valid PTRACERS_Iter0 value (error+stop or just warning). |
| 124 |
|
o pkg/cal (+ pkg/ecco): |
| 125 |
|
- avoid calling a S/R with 2 times the same arg (FWD: unless both are only used |
| 126 |
|
as input; and for AD, unless both are not differentiable (e.g., k index)). |
| 127 |
|
o model/src: |
| 128 |
|
- define internal flag "uniformFreeSurfLev" (=T when free-surface level-index |
| 129 |
|
is everywhere k=1) to speed-up some calculations (by pass kSurf reference); |
| 130 |
|
temporary set to usingZCoords value (to be changed to F when using shelfice). |
| 131 |
|
- hydrostatic pressure: implement partial cell near the surface also for |
| 132 |
|
Z-coordinates (for the case where uniformFreeSurfLev=F); replace test on |
| 133 |
|
usingZCoords with test on uniformFreeSurfLev where needed (in NonHyd code). |
| 134 |
|
o pkg/ctrl/ctrl_init.F: |
| 135 |
|
- add a formal parameter to s/r ctrl_init_rec to print |
| 136 |
|
xx_$(ctrl_valiable)_file for easier debugging |
| 137 |
|
- move call of cal_* routines into the if-block where they are really |
| 138 |
|
needed so that xx_$(ctrl_variable)_period can be 0 or -12 without having |
| 139 |
|
to specify a startdate |
| 140 |
|
o pkg/seaice: |
| 141 |
|
- update comments in SEAICE_OPTIONS.h (+ similar update of lab_sea/code and |
| 142 |
|
global_ocean.cs32x15/code versions). |
| 143 |
|
- turn on seaice Multi-category (in data.seaice, SEAICE_multDim=7) for |
| 144 |
|
verification experiment global_ocean.cs32x15.seaice and update results. |
| 145 |
|
o pkg/exf: |
| 146 |
|
- new S/R EXF_GETFIELD_START to return starting-time of a forcing field |
| 147 |
|
(according to useExfYearlyFields). |
| 148 |
|
o pkg/cal: |
| 149 |
|
- move calendar setting & summary (previously in cal_readparams.F) into new |
| 150 |
|
S/R CAL_INIT_FIXED to be called later in the initialisation process, once |
| 151 |
|
length of the simulation is set (after SET_PARMS call); This is an attempt |
| 152 |
|
to fix Pb when, e.g.: pkg/longstep, modifies the length of the run. |
| 153 |
|
- For safety: keep a record of pkg/cal parameter settings status |
| 154 |
|
to allow to STOP if one pkg/cal S/R is called too early. |
| 155 |
|
- add a stop in all basic pkg/cal subroutines if called before setting is done. |
| 156 |
|
- modify cal_getdate.F: stop when nothing was returned; |
| 157 |
|
if called with myIter=-1, simply retrun modelBaseDate (1,2 = startDate_1,_2); |
| 158 |
|
this allows to remove "cal.h" from seaice_read_parms.F & init_mnc_vars.F. |
| 159 |
|
|
| 160 |
|
checkpoint63l (2012/04/04) |
| 161 |
|
o pkg/bbl: |
| 162 |
|
- fix call to FIND_RHO_2D in bbl_calc_rho.F (case multi-tiles per proc); |
| 163 |
|
update global_with_exf output.yearly.txt file. |
| 164 |
|
o pkg/cal: |
| 165 |
|
- start to clean-up cal_set.F ; fix for the case where baseTime is non-zero. |
| 166 |
|
- remove 2 unsused and very short (1 line only) functions ; |
| 167 |
|
- remove unsused subroutines from AD-Diff list. |
| 168 |
|
o model/src: |
| 169 |
|
- start to implement clean wait to stop in initialization S/R (config_check.F, |
| 170 |
|
packages_check.F, ini_parms.F and pkg/obcs/obcs_check.F + pkg/mypackage) |
| 171 |
|
o folowing PRINT_MESSAGE changes: |
| 172 |
|
- remove reminder multiple print to the same file (STDOUT) when multi-threads. |
| 173 |
|
o pkg/bbl: |
| 174 |
|
- modification so it can work with z* |
| 175 |
|
- replace deltaT with dTtracerLev(k) for asynchronous time stepping |
| 176 |
|
- global_with_exf/input.yearly verification; changes output.yearly.txt |
| 177 |
|
o eesupp/src: |
| 178 |
|
- PRINT_MESSAGE: use OpenMP critical instruction (similar to PRINT_ERROR); |
| 179 |
|
writing to unit 0 (if unit=errorMessageUnit) extended to the case myThid=1 |
| 180 |
|
- CHECK_THREADS: specific code if using OpenMP threading: improve checking of |
| 181 |
|
# of threads across processors (e.g., different env OMP_NUM_THREADS setting) |
| 182 |
|
- improve special preliminary call to EESET_PARMS (if eedata is read 2 times). |
| 183 |
|
- when using MPI, always set usingMPI=T ; set default accordingly and stop |
| 184 |
|
if not consistent. |
| 185 |
|
- ALL_PROC_DIE: add a BARRIER after MPI_FINALIZE call ; skip this when thread |
| 186 |
|
argument is zero (in case called within single-thread section). |
| 187 |
|
o pkg/fizhi: |
| 188 |
|
clean-up turbulence cold-start switch: decided in fizhi_init_vars.F, stored |
| 189 |
|
in common bloc (fizhi_coms.h) and then passed as argument up to S/R TURBIO. |
| 190 |
|
o pkg/exch2: |
| 191 |
|
- add argument "updateCorners" to S/R exch2_get_uv_bounds (enable to |
| 192 |
|
activate EXCH_IGNORE_CORNERS in vector EXCH S/R). |
| 193 |
|
- switch 1rst EXCH call to IGNORE_CORNERS (instead of UPDATE_CORNERS) |
| 194 |
|
but keep the 2nd with UPDATE_CORNERS: this prevents overwriting good data |
| 195 |
|
with bad ones coming from a not-yet-updated halo. |
| 196 |
|
This should fix the problem when sNx <> sNy and GCD(sNx,sNy) < 2*Olx. |
| 197 |
|
o pkg/frazil: |
| 198 |
|
move FrzForcT diag from frazil_diagnostics_state.F to frazil_calc_rhs.F |
| 199 |
|
http://mitgcm.org/pipermail/mitgcm-devel/2012-March/005384.html |
| 200 |
|
o model/src/config_check.F: add a check for OLx>sNx/OLy>sNy |
| 201 |
|
o verification/lab_sea |
| 202 |
|
* #undef SEAICE_EXCLUDE_FOR_EXACT_AD_TESTING |
| 203 |
|
* #define ALLOW_SST_CONTROL & ALLOW_SST_CONTROL |
| 204 |
|
* nTimeSteps 3 -> 4 |
| 205 |
|
* changed reference output accordingly |
| 206 |
|
o pkg/ctrl: |
| 207 |
|
* fix ctrl_get_gen for xx_sst, xx_sss |
| 208 |
|
o pkg/fizhi: |
| 209 |
|
- fix 2 calls to S/R GRD2MSC (argument igrd) in update_earth_exports.F |
| 210 |
|
update results of exp. fizhi-cs-32x32x40 (fail with only 6 digits for cg2d) |
| 211 |
|
- by default, use new version of S/R GETPWHERE (#define TRY_NEW_GETPWHERE); |
| 212 |
|
fix few out-of-bounds indices in referencing vars (initialise igrd to 1); |
| 213 |
|
clean-up fizhi_turb.F (remove unused & uninitialised qdiaglsm array). |
| 214 |
|
o pkg/gmredi: |
| 215 |
|
- add diagnostic of Redi (main diagnonal) vertical diffusive flux of Temp |
| 216 |
|
o pkg/fizhi: |
| 217 |
|
- fix few initialisation and multi-tile Pb: aqualev & gridalt now pass the |
| 218 |
|
(old)aces open64 tuv test. |
| 219 |
|
o pkg/icefront: fix bug and reorder loop in icefront_init_varia.F |
| 220 |
|
o pkg/mom_vecinv: |
| 221 |
|
- fix 1 loop range in mom_vecinv.F to be like all the other loops which |
| 222 |
|
set/update gu,gv/Diss (fix Pb of uninitialised fVerUkp in admom_vecinv |
| 223 |
|
with non-lin free-surf). |
| 224 |
|
o model/src: move k-loop outside of i/j-loops when (re-)computing R_low |
| 225 |
|
in ini/update_masks_etc.F in order to avoid -O3 optimization problems with |
| 226 |
|
some compilers (gfortran 4.6.0 on MacOS) |
| 227 |
|
o momentum: |
| 228 |
|
- separate fVer?(:,:,kUp) & fVer?(:,:,kDown) in argument list of MOM_FLUXFORM |
| 229 |
|
& MOM_VECINV subroutines (to help TAF). This fixes Pb of uninitialised |
| 230 |
|
fVer?(kUp) in admom_fluxform with non-lin free-surf or bottom control. |
| 231 |
|
|
| 232 |
|
checkpoint63k (2012/03/17) |
| 233 |
|
o model/src: |
| 234 |
|
- selectAddFluid (ALLOW_ADDFLUID): add missing term in horizontal momentum |
| 235 |
|
flux-form (mom_fluxform.F) and vertical momentum (calc_gw.F) equations. |
| 236 |
|
- SOLVE_DIAGONAL: switch default to use adjoinable & vectorized version |
| 237 |
|
(#undef SOLVE_DIAGONAL_LOWMEMORY); update output of exp. front_relax.bvp. |
| 238 |
|
o model/src: solve_*diagonal.F |
| 239 |
|
- replace division with multiplication by inverse (similar to original version) |
| 240 |
|
update output of AD exp. global_ocean.90x40x15. |
| 241 |
|
- fix pkg/gmredi/gmredi_calc_psi_bvp.F & pkg/ggl90/ggl90_calc.F |
| 242 |
|
to enable the use of the vectorized & differentiable solve_tridiagonal.F |
| 243 |
|
- test exp. global_ocean.90x40x15/input_ad: switch to advection scheme 20 for |
| 244 |
|
vertical advection (to get solve_tridiagonal tested in AD mode). |
| 245 |
|
o pkg/seaice: ECCO-CLIVAR merge: |
| 246 |
|
1. seaice AREA relaxation |
| 247 |
* CPP option: SEAICE_ALLOW_AREA_RELAXATION |
* CPP option: SEAICE_ALLOW_AREA_RELAXATION |
| 248 |
* runtime flag: SEAICE_tauAreaObsRelax |
* runtime flag: SEAICE_tauAreaObsRelax |
| 249 |
2. modified global mean cost imbalance code: CPP option ALLOW_SSH_GLOBMEAN_COST_CONTRIBUTION |
2. modified global mean cost imbalance code: CPP option |
| 250 |
|
ALLOW_SSH_GLOBMEAN_COST_CONTRIBUTION |
| 251 |
o another iteration on solve_*diagonal.F |
o another iteration on solve_*diagonal.F |
| 252 |
Make choices independent of AUTODIFF |
Make choices independent of AUTODIFF |
| 253 |
Default is now the adjoinable vectorized version |
Default will be the adjoinable vectorized version. |
| 254 |
Alternative choices are: |
Alternative choices are: |
| 255 |
* still adjoinable but lower memory (k as inner-most loop) |
* non-adjoinable original version with low memory footprint: |
| 256 |
SOLVE_DIAGONAL_KINNER |
#define SOLVE_DIAGONAL_LOWMEMORY |
| 257 |
* non-adjoinable legacy version with low memory footprint |
* still adjoinable but does not vectorize (k as inner-most loop): |
| 258 |
SOLVE_DIAGONAL_LOWMEMORY |
#define SOLVE_DIAGONAL_KINNER (keeping SOLVE_DIAGONAL_LOWMEMORY undef) |
| 259 |
o testreport |
o testreport |
| 260 |
- add option to use a home made wrapper around "make" |
- add option to use a home made wrapper around "make" |
| 261 |
- remove MITGCM_IEEE env. variable (no longer active) |
- remove MITGCM_IEEE env. variable (no longer active) |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch