4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
5 |
============================== |
============================== |
6 |
|
|
7 |
|
o model/src/the_main_loop.F: |
8 |
|
adjust store directives for seaice evp and multicategory according to |
9 |
|
pkg/ecco/the_main_loop.F |
10 |
|
o adjoint: |
11 |
|
global_sum_ad was incomplete (noted many times by many people ;) |
12 |
|
nice description by Jean U. as follows: |
13 |
|
* allreduce(var) in global_sum.F can be interpreted as |
14 |
|
reduce(sum) followed by broadcast of sum. |
15 |
|
* The adjoint of that is |
16 |
|
sum(adjoint) followed by broadcast(summed adjoint value) |
17 |
|
or in other words, allreduce(sum, adjoint vars) |
18 |
|
We were missing the sum(adjoint)! |
19 |
|
Benign, except(!) for "global" scalars appearing in tile loops (bi,bj) |
20 |
|
(noted by J.M. Campin, e.g., in cost_ssh_mean, variable offset). |
21 |
|
The fix should produce 16-digit accuracy between |
22 |
|
single-CPU multi-tile (e.g. nSx=2, nPx=1) and |
23 |
|
multi-CPU single-tile (e.g. nSx=1, nPx-2) |
24 |
|
We'll probably change this yet again shortly |
25 |
|
o eesupp/src: |
26 |
|
- fix global_sum_r4 (wrong from previous modif in Jun 2009; ever used ?) |
27 |
|
o pkg/icefront |
28 |
|
- ALLOW_SUBGLACIAL_RUNOFF option adds capability for time-evolving |
29 |
|
specification of addmass array based on a 2D field, which is added |
30 |
|
at bottommost wet level at each horizontal location. |
31 |
|
o model/src: |
32 |
|
horizontal grid initialisation (other than curvilinear): add code for Exch2 IO |
33 |
|
layout; (not always compatible with delX,delY setting; commented out for now). |
34 |
|
o pkg/exch2: |
35 |
|
change tile to processor setting (case nSy*nPy > 1): allows default model mapping |
36 |
|
(myX and myYGlobalLow) to be used for simple (1 facet, no blank tile) config. |
37 |
|
|
38 |
checkpoint62e (2010/04/15) |
checkpoint62e (2010/04/15) |
39 |
o pkg/obcs: |
o pkg/obcs: |
40 |
- fix wrong index in OBCS_SPONGE_V for Eastern-OB (reported by David Hebert). |
- fix wrong index in OBCS_SPONGE_V for Eastern-OB (reported by David Hebert). |