| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o pkg/cost: |
| 8 |
|
- define a "final cost function" for each tile, and use GLOBAL_SUM_TILE_RL |
| 9 |
|
(instead of _GLOBAL_SUM_RL) to get the final cost from all procs & threads: |
| 10 |
|
o this fixes global-sum value if multi-threads (since GLOBAL_SUM_TILE_RL |
| 11 |
|
cannot be applied to var in common block). |
| 12 |
|
o allows to get results independent of tile-per-proc setting (if using |
| 13 |
|
slower GLOBAL_SUM_SEND_RECV option). |
| 14 |
|
- update accordingly: seaice_cost_final.F, shelfice_cost_final.F, |
| 15 |
|
thsice_cost_final.F and ecco_cost_final.F (this one affects lab_sea |
| 16 |
|
forward gradients: update output). |
| 17 |
|
- update */code_oad/externalDummies.F by adding dummy GLOBAL_SUM_TILE_RL |
| 18 |
|
S/R and, for most exp., also comment out GLOBAL_SUM_R8. |
| 19 |
|
o pkg/mom_common: |
| 20 |
|
- move to pkg/mom_common and model/src (previously in tutorial_deep_convection |
| 21 |
|
code) 2nd version of isotropic 3-D Smagorinsky code interface: strain and |
| 22 |
|
viscosity are locally declared in dynmics.F and pass as argument to CALC_GW; |
| 23 |
|
ensure that all field value that are used are set. |
| 24 |
|
o tutorial_deep_convection (smag3d): |
| 25 |
|
- add missing scaling in mom_calc_smag_3d.F; |
| 26 |
|
- set smag3D_coeff in input.smag3d such that results do not change. |
| 27 |
|
o pkg/dic: |
| 28 |
|
- change dic_cost.F to compute cost on each tile first and then apply |
| 29 |
|
GLOBAL_SUM_TILE_RL: this affects forward gradient of AD experiments |
| 30 |
|
tutorial_dic_adjoffline & tutorial_global_oce_biogeo; update output. |
| 31 |
|
- remove kLev argument and use GLOBAL_SUM_TILE_RL in tracer_meanarea. |
| 32 |
|
o eesupp/src & pkg/mdsio: |
| 33 |
|
- remove gather/scatter_vector.F (from eesupp) and mdsio_read/writevector.F |
| 34 |
|
(from pkg/mdsio) since they are no longer used for tape IO. |
| 35 |
|
- move simple C subroutine "setdir.c" from pkg/compon_communic to eesupp/src |
| 36 |
|
(+ keep a copy in pkg/atm_ocn_coupler). |
| 37 |
|
o verification/isomip: |
| 38 |
|
- split domain in 8 squared tiles (instead of just 2) for FWD, AD & OAD tests. |
| 39 |
|
o tools/build_options: |
| 40 |
|
- optfile linux_amd64_ifort11, fast (no ieee) case: |
| 41 |
|
to get a working restart for all experiments, add "-fp-model source" |
| 42 |
|
to FOPTIM (also remove "-opt-streaming-stores auto" <- this is the default). |
| 43 |
|
|
| 44 |
|
checkpoint64p (2013/10/20) |
| 45 |
|
o pkg/autodiff: |
| 46 |
|
- adread_adwrite.F: when type and length do not match ALLOW_AUTODIFF_WHTAPEIO |
| 47 |
|
criteria, return to simple (default) tape IO format. |
| 48 |
|
o pkg/mdsio (+ eesupp/src): |
| 49 |
|
- rewrite/simplify tape-IO S/R (new S/R MDS_READ/WRITE_TAPE replace |
| 50 |
|
previous MDSREAD/WRITEVECTOR) with 2 array argument from each type (R4/R8); |
| 51 |
|
- fix globalFile and singleCpuIO options using simpler global mapping. |
| 52 |
|
- new pairs of scatter/gather vector S/R (one for each type, generated from |
| 53 |
|
template) corresponding to simpler global mapping. |
| 54 |
|
- adread_write.F: switch to new S/R (but leave old calls commented) and write |
| 55 |
|
to globalFile when parameter globalFiles is set to True. |
| 56 |
|
- switch on useSingleCpuIO in AD experiments: global_ocean.90x40x15 & isomip. |
| 57 |
|
o pkg/autodiff: |
| 58 |
|
- adread_adwrite.F: add a hack for RS arrays (when RS=real*4); |
| 59 |
|
with ALLOW_AUTODIFF_WHTAPEIO defined, since corresponding (real*4) is |
| 60 |
|
missing, just add a STOP if array type do not match. |
| 61 |
|
o pkg/exf: |
| 62 |
|
- to allow to use seaice-fraction from pkg/exf in other pkgs than pkg/seaice: |
| 63 |
|
rename CPP option EXF_ALLOW_SEAICE_RELAX to EXF_SEAICE_FRACTION |
| 64 |
|
and rename array obsSIce to exf_iceFraction. |
| 65 |
|
o tutorial_deep_convection: |
| 66 |
|
- add 1rst version of isotropic 3-D Smagorinsky code (from Louis-Philippe), |
| 67 |
|
for now all in this exp. code dir. |
| 68 |
|
- add secondary test using Smag-3D viscosity (keep very low backgroud visc for |
| 69 |
|
testing) and flux-limter advection scheme for Temp. (no explicit diffusion). |
| 70 |
|
o eesupp/src: |
| 71 |
|
- added SINGLE_DISK_IO option because reading of parameter files on |
| 72 |
|
very large number of processors did not work with existing set-up |
| 73 |
|
|
| 74 |
|
checkpoint64o (2013/09/25) |
| 75 |
|
o pkg/atm_phys & atm_compon_interf: |
| 76 |
|
- allows for coupling of gray-atmosphere (atm_phys) with ocean. |
| 77 |
|
o tools/build_options: |
| 78 |
|
- switch from "-fPIC" to "-mcmodel=medium" in optfiles: linux_amd64_gfortran |
| 79 |
|
& linux_amd64_ifort11 and in linux_amd64_g77 & linux_amd64_open64 also |
| 80 |
|
remove CPP setting (was identical to genmake2 default). |
| 81 |
|
o pkg/gmredi & model/src: Some updating and work done on the residual model. |
| 82 |
|
- created new file model/src/calc_eddy_stress.F to calculate the eddy stress. |
| 83 |
|
- fixed up the calculation of the eddy stress and how it is added to the |
| 84 |
|
external forcing via taueddy_external_forcing.F |
| 85 |
|
- at the moment, it is very experimental and only works with the variable eddy |
| 86 |
|
diffusivity calculated in gmredi_k3d.F |
| 87 |
|
o pkg/seaice/seaice_check.F: add a check to catch the case of specifying |
| 88 |
|
lw/swfluxfile instead of lw/swdownfile in data.exf together with pkg/seaice |
| 89 |
|
o testreport (& tools/do_tst_2+2): |
| 90 |
|
- for MPI test, set default COMMAND to 'mpirun -v -np TR_NPROC ./{executable}' |
| 91 |
|
o tools/build_options: linux_amd64_gfortran & linux_amd64_ifort11: |
| 92 |
|
- remove CPP='cpp -traditional -P' (this is genmake2 default) |
| 93 |
|
- add switch (for big setups) to use 'medium' memory-model (but leave it |
| 94 |
|
commented out for now). Note: needed with FC 19 to compile big setups. |
| 95 |
|
o pkg/atm_compon_interf: |
| 96 |
|
- regroup per field origin (model & pkgs) the coupling storage routines |
| 97 |
|
(which accumalate in time each coupling field) |
| 98 |
|
- move bi,bj loops inside atm_store_my_data.F and store also wind-stress; |
| 99 |
|
update accordingly + simplify aim_do_physics.F |
| 100 |
|
|
| 101 |
|
checkpoint64n (2013/08/26) |
| 102 |
|
o pkg/bbl: |
| 103 |
|
- update formulation to reduce dispersion of tracer properties in bbl |
| 104 |
|
- replaced horizontal and vertical transport parameters with velocities |
| 105 |
|
- changes results for verification/global_with_exf.yearly |
| 106 |
|
o verification/halfpipe_streamice: |
| 107 |
|
- change to matrix construction; update both Fwd & AD output. |
| 108 |
|
o verification/tutorial_tracer_adjsens: |
| 109 |
|
- OpenAD test: switch to vector-invariant momentum since fluxform with |
| 110 |
|
rStar causes problems with OpenAD; update output. |
| 111 |
|
o tools: |
| 112 |
|
- genmake2: define a common prefix for all temp-file names ; move them |
| 113 |
|
to local /tmp disk (but would be easy to move them back); |
| 114 |
|
- xmakedepend: apply the same switch for temp-files. |
| 115 |
|
o pkg/offline: |
| 116 |
|
- changes for use with rStar (NLFS): assume that we continue to provide |
| 117 |
|
UVELMASS/hFacW_ini & VVELMASS/hFacS_ini (hFacWi,S_ini as written in grid |
| 118 |
|
output files) and apply r* scaling at each time-step (function of the time |
| 119 |
|
varying Eta field); use normal pickup files (needed for restart with NLFS). |
| 120 |
|
o adjoint with OpenAD: |
| 121 |
|
- move OpenAD-specific files from verification/OpenAD/code_oad_all to a new |
| 122 |
|
package openad. This package has to be compiled for all runs with OpenAD |
| 123 |
|
instead of specifying code_oad_all as a modification directory. |
| 124 |
|
- link the fortran templates from tools/OAD_support to the build directory |
| 125 |
|
for preprocessing to avoid conflicts between different setups compiled |
| 126 |
|
simultaneously and simplify cleanup. |
| 127 |
|
o pkg/exf: |
| 128 |
|
- in exf_getforcing.F: comment out most of the EXCH calls (fields used by |
| 129 |
|
main model or other pkgs are exchanged in EXF_MAPFIELDS). |
| 130 |
|
o model/src: |
| 131 |
|
- do_stagger_fields_exchanges.F: skip EXCH calls when using pkg/offline |
| 132 |
|
(similar to do_fields_blocking_exchanges.F). |
| 133 |
|
o tools/genmake2: |
| 134 |
|
- remove TAF output src code just before calling staf (when not doing a fresh |
| 135 |
|
built, will stop if things go wrong with staf). |
| 136 |
|
- fix target "adobj" (left from TAF version-2 transition) |
| 137 |
|
o pkg/diagnostics: |
| 138 |
|
- add a parameter to track the status of pkg/diagnostics activation; |
| 139 |
|
this replace/extend the use of logical param "settingDiags" (to check |
| 140 |
|
when adding diag to the list) to also check any DIAGNOSTICS_[]FILL* call. |
| 141 |
|
New S/R to print error message and stop when pkgStatus is not right. |
| 142 |
|
- to try: add a warning when "data.diagnostics" exist but useDiagnostics=F. |
| 143 |
|
|
| 144 |
|
checkpoint64m (2013/08/13) |
| 145 |
|
o model/inc: |
| 146 |
|
- move FCORI_MACROS.h from PARAMS(_MACROS).h to GRID_MACROS.h (left from |
| 147 |
|
8 yrs ago modif when Coriolis arrays were moved from PARAMS.h to GRID.h) |
| 148 |
|
o pkg/kpp & aim_v23: |
| 149 |
|
- only call DIAGNOSTICS_FILL when useDiagnostics=TRUE. |
| 150 |
|
o pkg/seaice: |
| 151 |
|
- restore (last time on 2011/12/24) ability to compile pkg/seaice with |
| 152 |
|
undefined pkg/exf ALLOW_ATM_TEMP or ALLOW_RUNOFF CPP-options. |
| 153 |
|
- only call DIAGNOSTICS_FILL when useDiagnostics=TRUE. |
| 154 |
|
o eesupp/src: |
| 155 |
|
- fix check for overlap-size argument (myOLn.GT.MAX_OLY_EXCH) |
| 156 |
|
in exch1_rx[_ad].template (thanks to Francois Lefeuvre report). |
| 157 |
|
o switch to TAF version-2: |
| 158 |
|
- add flow directives to "freeze" naming convention for ad-routines that |
| 159 |
|
are used in I/O routines of adjoint state and associated common blocks |
| 160 |
|
that are used in adcommon.h (backward compatible with version-1) |
| 161 |
|
- update all ad-optfiles (in tools/adjoint_options + few local to some |
| 162 |
|
verification exp) to switch to latest TAF version (currently 2.5.5); |
| 163 |
|
merge "*_mpi" adoptfiles into the corresponding non-mpi (but commented |
| 164 |
|
out except for diva). |
| 165 |
|
- the major change with TAF version-2 is that both the differentiated |
| 166 |
|
routines and the forward routines (new) are returned as part of the TAF |
| 167 |
|
output source code. Update genmake2 accordingly. |
| 168 |
|
- new output for AD experiment tutorial_global_oce_biogeo (better AD gradient |
| 169 |
|
with version-2 when compared to TLM or FWD gradient); other experiments |
| 170 |
|
output show minor changes (machine truncation level); update the few AD & |
| 171 |
|
TLM output with 12 matching digits only (some prior to version-2 switch). |
| 172 |
|
|
| 173 |
|
checkpoint64l (2013/08/05) |
| 174 |
|
o model/src: |
| 175 |
|
- in dynamics.F, skip the call to CALC_VISCOSITY if momViscosity=F |
| 176 |
|
- main_do_loop.F: no longer includes additional (AUTODIFF) header files |
| 177 |
|
when using TAF but only if using OpenAD. |
| 178 |
|
o pkg/mom_common+fluxform+vecinv: |
| 179 |
|
- always set horiz. viscosity arrays to background value before calling |
| 180 |
|
MOM_CALC_VISC (in MOM_FLUXFORM & MOM_VECINV) and call S/R MOM_CALC_VISC |
| 181 |
|
only when using variable horiz. viscosity (useVariableVisc=T); |
| 182 |
|
- simplify mom_vecinv.F (only 1 block for momViscosity). |
| 183 |
|
o pkg/mom_common: variable horiz. viscosity in NH code |
| 184 |
|
- move horiz-viscosity 3-D arrays viscAh_W & viscA4_W (used in calc_gw.F) |
| 185 |
|
from model/inc/NH_VARS.h to pkg/mom_common/MOM_VISC.h |
| 186 |
|
- store in common block (in MOM_VISC.h) logical parameters: useHarmonicVisc, |
| 187 |
|
useBiharmonicVisc & useVariableVisc. |
| 188 |
|
- fix old bug (already in 1rst implementation of variable horizontal viscosity |
| 189 |
|
in calc_gw.F, version 1.27, 2006/06/20): |
| 190 |
|
horizontal bi-harmonic viscosity in gW was missing when wiscA4W=0 |
| 191 |
|
even if variable bi-harmonic viscosity (e.g., viscA4Grid) was non-zero. |
| 192 |
|
- change background horiz viscosity in vertical momentum (gW) to always be |
| 193 |
|
viscAhW,viscA4W (instead of switching to viscAhD, viscA4D when using |
| 194 |
|
any type of variable viscosity, e.g., viscAh/4Grid) |
| 195 |
|
- change default value of viscAhW,viscA4W to viscAhD,viscA4D (instead of |
| 196 |
|
viscAh,viscA4) which default to viscAh,viscA4 : this avoid missing |
| 197 |
|
horiz. viscosity in W eq when just setting viscAhD/Z,viscA4D/Z (since |
| 198 |
|
default viscAh,viscA4 is zero). |
| 199 |
|
o model/src: |
| 200 |
|
- extend length of drC & recip_drC from Nr to Nr+1. |
| 201 |
o Divided Adjoint (DIVA) with MPI: |
o Divided Adjoint (DIVA) with MPI: |
| 202 |
- genmake2: do not create local ./mpi_headers dir when no file to copy into. |
- genmake2: do not create local ./mpi_headers dir when no file to copy into. |
| 203 |
- remove option ALLOW_DIVIDED_ADJOINT_MPI (replaced by ALLOW_DIVIDED_ADJOINT |
- remove option ALLOW_DIVIDED_ADJOINT_MPI (replaced by ALLOW_DIVIDED_ADJOINT |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch