| 3 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 4 |
============================== |
============================== |
| 5 |
|
|
| 6 |
|
checkpoint50a_post |
| 7 |
|
o Bug fix for merging between c50 and KPP. |
| 8 |
|
ikey was passed from thermodynamics to kpp_calc via |
| 9 |
|
common block rather than being recomputed in kpp_calc, |
| 10 |
|
in contradiction with new key itdkey. |
| 11 |
|
New key ikppkey created, and tamc.h headers updated. |
| 12 |
|
|
| 13 |
|
checkpoint50 |
| 14 |
|
- 2nd step of merge with ecco-branch: |
| 15 |
|
merge changes in ecco-branch back into main trunk |
| 16 |
|
- bug fix in find_alpha |
| 17 |
|
- more storing in gmredi_calc_tensor.F, kpp_routines.F |
| 18 |
|
- avoid iterative loop in ini_pressure for ALLOW_AUTODIFF_TAMC |
| 19 |
|
- merged some sliced AD support routines |
| 20 |
|
|
| 21 |
checkpoint49 |
checkpoint49 |
| 22 |
|
merged from ecco-branch: |
| 23 |
|
o exf: |
| 24 |
|
- Enable initialisation of forcing fields to constant |
| 25 |
|
(runtime) values. |
| 26 |
|
- in exf_getffields.F |
| 27 |
|
Reduce i-/j-loop to interior domain, discarding overlaps. |
| 28 |
|
That also fixes wrong TAF-key computations for key_1, key_2 |
| 29 |
|
with bulf formulae. |
| 30 |
|
- exf_init.F modify #ifdef for exf_init_evap |
| 31 |
|
- exf_getffieldrec.F, ctrl_getrec.F |
| 32 |
|
The following INT-usages are not safe: |
| 33 |
|
fldsecs = int(fldsecs/fldperiod)*fldperiod |
| 34 |
|
fldcount = int(fldsecs/fldperiod) + 1 |
| 35 |
|
and were modified. |
| 36 |
|
o autodiff: checkpoint_lev... |
| 37 |
|
- updated storing for bulk formulae (some arrays were missing) |
| 38 |
|
- added state for seaice package |
| 39 |
|
- synchronized with main branch |
| 40 |
|
o DIVA: |
| 41 |
|
Modif.s in |
| 42 |
|
- the_main_loop.F |
| 43 |
|
- the_model_main.F |
| 44 |
|
and added cost_final_restore.F cost_final_store.F |
| 45 |
|
o With these changes, it is possible to fully separate |
| 46 |
|
the first forward run to compute the cost function (mdthe_main_loop) |
| 47 |
|
and adjoint runs over individual DIVA intervals (adthe_main_loop) |
| 48 |
|
into separate model executions. |
| 49 |
|
This enables better tuning of DIVA to given queue sizes. |
| 50 |
|
o Loss of cost function value in consecutive DIVA runs is |
| 51 |
|
fixed by storing to 'costfinal' file. |
| 52 |
|
o Appropriate call of ctrl_unpack/_pack,grdchk_main at |
| 53 |
|
beginning/end of adjoint integration is accounted for. |
| 54 |
|
The current changes apply to TAF-1.4.26. |
| 55 |
|
TAF has one remaining bug, requiring one manual change |
| 56 |
|
(see MITgcm manual). |
| 57 |
|
o cal: synchronized between MAIN and ecco-branch |
| 58 |
|
|
| 59 |
|
checkpoint48i_post |
| 60 |
o Changes to restore differentiability of code w.r.t. previous tag |
o Changes to restore differentiability of code w.r.t. previous tag |
| 61 |
(mostly adding new routines to make list and replacing |
(mostly adding new routines to make list and replacing |
| 62 |
pressure by totPhiHyd). |
pressure by totPhiHyd). |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch