| 4 |
Notes on tags used in MITgcmUV |
Notes on tags used in MITgcmUV |
| 5 |
============================== |
============================== |
| 6 |
|
|
| 7 |
|
o pkg/diagnostics/diagstats_calc.F: add special code for TARGET_NEC_SX |
| 8 |
|
that vectorises. This code uses 3 extra 2D fields, F90 array assignments |
| 9 |
|
and F90 intrinsic routines SUM, MINVAL, MAXVAL, but since TARGET_NEC_SX |
| 10 |
|
does not even come with an F77 compiler anymore, this should not be |
| 11 |
|
a limitation |
| 12 |
|
|
| 13 |
|
checkpoint63e (2011/11/06) |
| 14 |
|
o pkg/seaice adjoint of LSR: |
| 15 |
|
- substantially rework STORE dirs. |
| 16 |
|
- add new flag SEAICE_LSR_ADJOINT_ITER and parameter SOLV_MAX_FIXED |
| 17 |
|
this flag will do SOLV_MAX_ITERS but only store SOLV_MAX_FIXED fields |
| 18 |
|
and only perform SOLV_MAX_FIXED adjoint iters. |
| 19 |
|
If flag SEAICE_LSR_ADJOINT_ITER is undef-ined, the store in the iteration |
| 20 |
|
always goes to same record, meaning that the last forward field will be |
| 21 |
|
the one used in the adjoint - an approximation. |
| 22 |
|
- now produces non-zero LSR-related gradients, and acceptable gradients for |
| 23 |
|
An's zonal (x-z) test case (to be checked in) |
| 24 |
|
- lab_sea: produces acceptable gradients for lab_sea with forward/backward LSR |
| 25 |
|
(but these are screwed up by the thermodynamics in the first place); |
| 26 |
|
reduce from 4 to 3 timesteps to avoid t.d. issue as temporary solution; |
| 27 |
|
output_adm.txt also reflects changes to seaice_lsr |
| 28 |
|
- global_ocean.cs32x15: |
| 29 |
|
* the "default" input_ad.seaice has been re-defined to always exclude dynamics |
| 30 |
|
(output_adm.seaice.txt was updated) |
| 31 |
|
* new input_ad.seaice_dynmix tests a forward LSR / backward free-drift |
| 32 |
|
(output_adm.seaice_dynmix.txt) |
| 33 |
|
o test-exp. exp4+stevens: |
| 34 |
|
- switch off pTracers and nonHydrostatic (restart problems with NH); update output |
| 35 |
|
o pkg/seaice (lsr solver): |
| 36 |
|
- put coriolis term into FORCEX,Y (as in seaice_evp.F; previously in rhsU,V only); |
| 37 |
|
update output (change truncation) of FWD test exp lab_sea & global_ocean.cs32x15 |
| 38 |
|
- fix SEAICEuseFREEDRIFT for new OBCS version (ndef OBCS_UVICE_OLD) |
| 39 |
|
- new parameter "LSR_mixIniGuess" allows (if=2,4) to mix free-drift solution |
| 40 |
|
into initial guess for linearised solver; use weight ~ (1/err)^2,4. |
| 41 |
|
- compute (with new S/R) and print initial (+ free-driff sol. if available) |
| 42 |
|
and final residual of linearised system. |
| 43 |
|
o pkg/pp81: add diagnostics for vertical eddy viscosity and diffusivity |
| 44 |
|
(requires new subroutines pp81_init_fixed.F/pp81_diagnostics_init.F) |
| 45 |
|
o pkg/obcs: Stevens BCs (still only for east and west) |
| 46 |
|
- fix Stevens BCs by using intermediate velocities of previous |
| 47 |
|
time step as "baroclinic" contribution rather velocity of one grid cell |
| 48 |
|
inside the boundaries; more consistent but makes the entire treatment |
| 49 |
|
more complicated, requires new pickup files and two new S/Rs: |
| 50 |
|
obcs_copy_uv_n and obcs_save_uv_n |
| 51 |
|
keep old version with flag OBCS_STEVENS_USE_INTERIOR_VELOCITY |
| 52 |
|
(undefined by default) |
| 53 |
|
- silently move computation of some grid factors into separate loops to |
| 54 |
|
save some recomputations |
| 55 |
|
- add a hack for non-linear free surface |
| 56 |
|
- add a verification experiment to exp4 that tests this code |
| 57 |
|
o make gluemnc/xplodemnc work with (moron) non-GNU shell tools |
| 58 |
|
|
| 59 |
|
checkpoint63d (2011/10/25) |
| 60 |
o pkg/obcs + seaice-dynamics: |
o pkg/obcs + seaice-dynamics: |
| 61 |
- new implementation calls OBCS_APPLY_UVICE (in seaice_dynsolver.F) before |
- new implementation calls OBCS_APPLY_UVICE (in seaice_dynsolver.F) before |
| 62 |
solving for uIce,vIce and prevent SEAICE_LSR or SEAICE_EVP to modify OB |
solving for uIce,vIce and prevent SEAICE_LSR or SEAICE_EVP to modify OB |
| 77 |
o pkg/dic: |
o pkg/dic: |
| 78 |
- Remove subroutine CALC_PCO2_APPROX_CO3 from carbon_chem.F and add |
- Remove subroutine CALC_PCO2_APPROX_CO3 from carbon_chem.F and add |
| 79 |
carbonate computation/output to CALC_PCO2_APPROX |
carbonate computation/output to CALC_PCO2_APPROX |
| 80 |
|
o pkg/generic_advdiff |
| 81 |
|
- make TAF generate code that vectorizes: |
| 82 |
|
gad_dst3fl_adv_*.F, gad_u3c4_impl_r.F |
| 83 |
o model/src: |
o model/src: |
| 84 |
- improve vectorization also for adjoint code of |
- improve vectorization also for adjoint code of |
| 85 |
solve_tridiagonal and solve_pentadiagonal. As this requires up to |
solve_tridiagonal and solve_pentadiagonal. As this requires up to |
| 86 |
5 new 3D-fields, this code is only turned on for the combination of |
5 new 3D-fields, this code is only turned on for the combination of |
| 87 |
ALLOW_AUTODIFF_TAMC and TARGET_NEC_SX. |
ALLOW_AUTODIFF_TAMC and TARGET_NEC_SX. Also remove error message |
| 88 |
|
for singular matrix check (only for TARGET_NEC_SX). |
| 89 |
- remove hard-wiring of CPP-flag ALLOW_SOLVERS_KLOOPINSIDE to |
- remove hard-wiring of CPP-flag ALLOW_SOLVERS_KLOOPINSIDE to |
| 90 |
ALLOW_AUTODIFF. This flag now needs to be set outside of |
ALLOW_AUTODIFF. This flag now needs to be set outside of |
| 91 |
solve_tri/pentadiagonal.F |
solve_tri/pentadiagonal.F |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch