4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
5 |
============================== |
============================== |
6 |
|
|
7 |
|
o pkg/autodiff: |
8 |
|
- include PACKAGES_CONFIG.h in exch*ad.flow & exch*ad.F |
9 |
|
(not supposed to be in CPP_OPTIONS.h) to get ALLOW_EXCH2 defined correctly. |
10 |
|
+ clean-up comments (_RS is not always real*4) |
11 |
|
- fix bug in ADEXCH_UV_3D_RX (without EXCH2, was calling the forward exch S/R) |
12 |
|
- add a stop in ADEXCH_UV_AGRID_3D_RX if trying to use it with EXCH2: |
13 |
|
was calling a forward exch2, which is wrong ; used in SEAICE & EXF pkgs. |
14 |
|
o testreport (AD-tests): |
15 |
|
no longer link files from dir "../input" (1rst link files from |
16 |
|
"../input_ad.[xx]" if secondary test "xx" and then from "../input_ad") |
17 |
|
|
18 |
|
checkpoint62f (2010/04/29) |
19 |
o model/src/the_main_loop.F: |
o model/src/the_main_loop.F: |
20 |
adjust store directives for seaice evp and multicategory according to |
adjust store directives for seaice evp and multicategory according to |
21 |
pkg/ecco/the_main_loop.F |
pkg/ecco/the_main_loop.F |
22 |
o adjoint: |
o adjoint: |
23 |
global_sum_ad was incomplete (noted many times by many people ;) |
global_sum_ad was incomplete (noted many times by many people ;) |
24 |
nice description by Jean U. as follows: |
nice description by Jean U. as follows: |
25 |
* allreduce(var) in global_sum.F can be interpreted as |
* allreduce(var) in global_sum.F can be interpreted as |
26 |
reduce(sum) followed by broadcast of sum. |
reduce(sum) followed by broadcast of sum. |
27 |
* The adjoint of that is |
* The adjoint of that is |
28 |
sum(adjoint) followed by broadcast(summed adjoint value) |
sum(adjoint) followed by broadcast(summed adjoint value) |
29 |
or in other words, allreduce(sum, adjoint vars) |
or in other words, allreduce(sum, adjoint vars) |
30 |
We were missing the sum(adjoint)! |
We were missing the sum(adjoint)! |
31 |
Benign, except(!) for "global" scalars appearing in tile loops (bi,bj) |
Benign, except(!) for "global" scalars appearing in tile loops (bi,bj) |