4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
5 |
============================== |
============================== |
6 |
|
|
7 |
o useSingleCpuInput separates single-CPU input from single-CPU output |
o pkg/thsice |
8 |
o pkg/seaice/seaice_lsr.F: fix an initialisation bug for the |
- corrected store directive fixes thsice adjoint gradient checks |
9 |
SEAICE_VECTORIZE_LSR_ZEBRA option that lead to NaNs on a NEC-SX8 |
o genmake2 & testreport: |
10 |
|
- add genmake2 option "-oad" to generate a Makefile for OpenAD built and |
11 |
|
in this case, use tools/adjoint_options/adjoint_oad as default AD optfile. |
12 |
|
Update testreport accordingly. No need to set "export OPENAD=true" |
13 |
|
in adjoint_options/adjoint_oad anymore (will be removed later). |
14 |
|
o pkg/shelfice: |
15 |
|
- put T & S shelfice forcing into model surfaceForcing_T,_S (for now, |
16 |
|
only if SHELFICEboundaryLayer=F); Change model/src/external_forcing.F |
17 |
|
so that surface forcing applies to k=kSurf if using pkg/shelfice. |
18 |
|
o pkg/ctrl, pkg/grdchk, pkg/ecco: |
19 |
|
- Changes to make GENARR more modular (i.e. invisible if not used) |
20 |
|
o model/src/temp_integrate.F and salt_integrate.F |
21 |
|
- re-include store directives that were removed on Dec 27th. |
22 |
|
o model/src/external_forcing_surf.F |
23 |
|
- add storage directives for EmPmR and PmEpR to avoid recomputation of |
24 |
|
seaice_model in the adjoint that is problematic when seaice_model is |
25 |
|
turned off in the adjoint. The resolved issue dated back to July when |
26 |
|
EXTERNAL_FORCING_SURF call was moved. |
27 |
|
Update global_ocean.cs32x15/*/output_adm.seaice_dynmix.txt accordingly. |
28 |
|
o pkg/exf/exf_check.F |
29 |
|
- fix a small bug again (originally found by Wentao Liu, now found again) |
30 |
|
- streamline error output |
31 |
|
o pkg/streamice and aim_v23: |
32 |
|
- replace a few "_EXCH_XY_RL (" with "_EXCH_XY_RL(", so that |
33 |
|
Apple's /usr/bin/cpp knows what to do: |
34 |
|
Apple LLVM version 5.0 (clang-500.2.79) (based on LLVM 3.3svn) |
35 |
|
Target: x86_64-apple-darwin13.0.0 |
36 |
|
|
37 |
|
checkpoint64s (2014/01/05) |
38 |
|
o tools/adjoint_options: |
39 |
|
- switch to fastopt.net TAF server (instead of default fastopt.de) for |
40 |
|
AD optfiles adjoint_diva & adjoint_f95 (similar to adjoint_default). |
41 |
|
o model/src: |
42 |
|
- remove some unnecessary TAF storage directives in forward_step.F and |
43 |
|
in dynamics.F (note: not always removed by TAF, e.g., double storage |
44 |
|
of kappaRU,kappaRV in former version of dynamics.F). |
45 |
|
o model/src & pkg/ptracers: |
46 |
|
- call directly GAD_IMPLICIT_R & IMPLDIFF + DWNSLP_APPLY from temp_integrate.F |
47 |
|
salt_integrate.F and ptracers_integrate.F (instead of from thermodynamics.F) |
48 |
|
and remove ptracers_dwnslp_apply.F & ptracers_implicit.F |
49 |
|
- call MATRIX_STORE_TENDENCY_EXP for each tracer (in ptracers_integrate.F) |
50 |
|
- add fZon & fMer as output argument of S/R GAD_CALC_RHS. |
51 |
|
- update TAF AD-output of experiments global_ocean.90x40x15 (13 for AD-grad). |
52 |
|
o useSingleCpuInput separates single-CPU input from single-CPU output; |
53 |
|
for now only implemented for in exf_interp_read.F |
54 |
|
o verification/advect_xy: |
55 |
|
- fix missing EXCH call in local version of ini_salt.F and update output. |
56 |
|
o momentum diagnostics: |
57 |
|
- fix which deltaT is used in Total-Tendency diagnostics. |
58 |
|
- fix typo in diagnostics for momentum tendency from AB. |
59 |
|
o pkg/seaice/seaice_lsr.F: fix an initialisation bug for the |
60 |
|
SEAICE_VECTORIZE_LSR_ZEBRA option that lead to NaNs on a NEC-SX8 |
61 |
|
|
62 |
checkpoint64r (2013/12/08) |
checkpoint64r (2013/12/08) |
63 |
o verification/*/results: |
o verification/*/results: |
163 |
o pkg/cost: |
o pkg/cost: |
164 |
- define a "final cost function" for each tile, and use GLOBAL_SUM_TILE_RL |
- define a "final cost function" for each tile, and use GLOBAL_SUM_TILE_RL |
165 |
(instead of _GLOBAL_SUM_RL) to get the final cost from all procs & threads: |
(instead of _GLOBAL_SUM_RL) to get the final cost from all procs & threads: |
166 |
o this fixes global-sum value if multi-threads (since GLOBAL_SUM_TILE_RL |
o this fixes global-sum value if multi-threads (since GLOBAL_SUM_RL |
167 |
cannot be applied to var in common block). |
cannot be applied to var in common block). |
168 |
o allows to get results independent of tile-per-proc setting (if using |
o allows to get results independent of tile-per-proc setting (if using |
169 |
slower GLOBAL_SUM_SEND_RECV option). |
slower GLOBAL_SUM_SEND_RECV option). |