[MITgcm-devel] tiling with seaice

[Menemenlis, Dimitris (3248)]

JM, this is what I would expect if there is sea ice crossing the tile boundary.
The LSR solver (as coded) will not give the exact same solution if the
tiling changes.  This difference can be decreased by decreasing
LSR_ERROR but this increases computational cost, i.e., more solver iterations.
Dimitris

On Mar 19, 2010, at 7:44 AM, Jean-Michel Campin wrote:

> Hi,
> 
> Just to confirm:
> I run testreport (forward, with gfortran on my laptop) on lab_sea 
> but with both GLOBAL_SUM_SEND_RECV & CG2D_SINGLECPU_SUM
> defined in CPP_EEOPTIONS.h
> And I compared the 4-tiles (as set-up in code/SIZE.h) results
> with a single tile (20x16) set-up:
> The 2 evp tests (standard & hb87) give the same resuts
> (zero diff for pickup files) between 4-tiles & 1-tile.
> And the 2 lsr tests (lsr & salt_plume) give different
> results (difference in cg2d_ini_res are similar to what
> I got when changing the compiler).
> 
> Cheers,
> Jean-Michel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel