#1398 closed Bug (fixed)
LIM3 is not riproducibile
| Reported by: | flavoni | Owned by: | nemo |
|---|---|---|---|
| Priority: | normal | Milestone: | |
| Component: | LIM3 | Version: | v3.6 |
| Severity: | Keywords: | LIM* v3.6 | |
| Cc: |
Description
LIM3 run on cluster IBM, proc Intel ES 4650
2 tests :
jpi=2 jpj= 8, and jpi=4 jpj=4
after 12 timesteps solver.stat are differents.
( options:
- -O0 (compiler option)
- key mpp_rep
- jperio = 4: cyclic east-west and north fold with T-point pivot)
- ln_limdyn = .true.
)
Commit History (1)
| Changeset | Author | Time | ChangeLog |
|---|---|---|---|
| 4833 | flavoni | 2014-11-06T15:58:37+01:00 | fix LIM3 reproducibility problems, see ticket #1398 |
Change History (6)
comment:1 Changed 10 years ago by flavoni
comment:2 Changed 10 years ago by flavoni
- Resolution set to fixed
- Status changed from new to closed
comment:3 Changed 7 years ago by nemo
- Keywords LIM* added
comment:4 Changed 7 years ago by nemo
- Keywords release-3.6* added
comment:5 Changed 7 years ago by nemo
- Keywords release-3.6* removed
comment:6 Changed 2 years ago by nemo
- Keywords v3.6 added
Note: See
TracTickets for help on using
tickets.
Problem of LIM3 reproducibility was du to this IF in limthd_lac.F90 routine :
IF( zv_newfra > 0._wp ) THEN
(because different domain decomposition give sometimes zv_newfra >0 and sometimes not.Revision 4833 concerne this bug fix,and I add also initialization, add do from 2 in do loop in IF( fraz_swi == 1 ) and lbc_lnk,and do initialization from 1 to nbpac instead of all jpi*jpj when needed.