| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o model/src: |
| 8 |
|
- collect momentum forcing tendency into local 2-D array (passed as argument |
| 9 |
|
to apply_forcing S/R); this simplifies mom. Forcing diagnostics and CD_CODE |
| 10 |
|
parts in timestep.F (without CD_CODE_NO_AB_MOMENTUM defined = the default); |
| 11 |
|
affects results at machine truncation level (e.g., if using OBCS_SPONGE). |
| 12 |
|
- collect T & S forcing tendency into local 2-D array (passed as argument to |
| 13 |
|
apply_forcing S/R); add 3-D diagnostics for temp & salt forcing tendency. |
| 14 |
|
affects results at machine truncation level (e.g., ocean: if using |
| 15 |
|
SHORTWAVE_HEATING and atmos:, if using FrictionHeating or (atm_Rq<>0) |
| 16 |
|
moist correction due to vertical advection). |
| 17 |
|
- update output of experiments: fizhi-cs-32x32x40 (13 digits), |
| 18 |
|
global_ocean.cs32x15 (fwd: icedyn, seaice, thsice, all 3 @ 11d; adm & tml: |
| 19 |
|
seaice_dynmix & thsice, fwd grd @ 3 & 8), hs94.1x64x5 (ad grd @ 13), |
| 20 |
|
obcs_ctrl (ad & fwd grd @ 12 & 7). |
| 21 |
|
o pkg/ptracers & diagnostics: |
| 22 |
|
- use ptracers (short) name (instead of long one) to make ptracers diagnostics |
| 23 |
|
title (to avoid some unexpected truncation); sort out ptracers diags units. |
| 24 |
|
- collect ptracer forcing tendency into local 2-D array (passed as argument |
| 25 |
|
to apply_forcing S/R); add 3-D diagnostics for ptracers forcing tendency. |
| 26 |
|
|
| 27 |
|
checkpoint65 (2014/07/18) |
| 28 |
|
o pkg/fizhi: |
| 29 |
|
- add parenthesis around fizhi tendencies in fizhi_tendency_apply.F: this |
| 30 |
|
affects results at machine truncation level. Update output of experiments |
| 31 |
|
fizhi-cs-32x32x40 (12 digits for cg2d) & fizhi-cs-aqualev20 (13 digits). |
| 32 |
|
o pkg/seaice: more clean up |
| 33 |
|
- rename TAF common block tapes to be more consistent and remove |
| 34 |
|
unnecessary duplicates |
| 35 |
|
- add store directives to remove more recomputations, but use smaller |
| 36 |
|
tapes (and adjust keys) when SEAICE_LSR_ADJOINT_ITER is undefined. |
| 37 |
|
The idea behind this flag is to reduce the memory footprint when the |
| 38 |
|
adjoint of the LSR is not required. As of now, defining the flag |
| 39 |
|
SEAICE_LSR_ADJOINT_ITER increases 4 common block tapes for 2D field by |
| 40 |
|
a factor of SOLV_MAX_FIXED (=500 by default in SEAICE_SIZE.h). |
| 41 |
|
For experiment offline_exf_seaice, this means on a linux box with gfortran |
| 42 |
|
nearly a factor of 10 (but keep in mind there is mostly seaice code in |
| 43 |
|
this experiment): |
| 44 |
|
> size mitgcmuv_ad |
| 45 |
|
text data bss dec hex |
| 46 |
|
5100659 1216 220458952 225560827 d71c8fb |
| 47 |
|
5339425 1200 1868783272 1874123897 6fb4dc79 (SEAICE_LSR_ADJOINT_ITER) |
| 48 |
|
but only a moderate increase of 36% for global_ocean.cs32x15: |
| 49 |
|
text data bss dec hex |
| 50 |
|
6695027 1648 1014578664 1021275339 3cdf6ccb |
| 51 |
|
6695203 1648 1382505320 1389202171 52cd8afb (SEAICE_LSR_ADJOINT_ITER) |
| 52 |
|
- note: there are still recomputations in global_ocean.cs32x15 and |
| 53 |
|
offline_exf_seaice and the gradient agreement changes (gets a |
| 54 |
|
little worse) for global_ocean.cs32x15 (-> update output_adm.seaice.txt) |
| 55 |
|
- in general, the AD gradients still are wrong |
| 56 |
o OpenAD |
o OpenAD |
| 57 |
- clean up code_oad/ directories somewhat by moving files |
- clean up code_oad/ directories somewhat by moving files |
| 58 |
dontCompile, dontTransform, keepOriginal to OAD_support/ |
dontCompile, dontTransform, keepOriginal to OAD_support/ |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch