[MITgcm-devel] Default option regaring solve_*diagonal.F

[Jean-Michel Campin]

Hi patrick,

An update:
I made little changes in solve_tridiagonal.F (+ similar one
in solve_pentadiagonal.F), replacing division with multiplication 
by inverse (more similar to original version)
so that now, with undef SOLVE_DIAGONAL_LOWMEMORY, and since ggl90_calc.F
 & gmredi_calc_psi_bvp.F have been fixed,
 vermix.ggl90 : identical results (previously ony 2 digits for Salt StdDev)
 front_relax.bvp -> gain 1 matching digits (compare to previously)
but will still have to update the output (10 cg2d digits match).
I will let it go for 1 night testing and change the default
SOLVE_DIAGONAL_LOWMEMORY in model/inc/CPP_OPTIONS.h tomorrow.

Plan to do a new tag, checkpoint63k, after that.
Ian and Dimitris, what is your time-table for the seaice+frazil
interface ? Can this wait 2 days (then it will be next week) ?

I tested the Adjoint with the 2 experiments global_ocean.90x40x15
and global_ocean.cs32x15, results change a little bit (truncation)
but still pass, so I don't plan to update the output.

Patrick, I have a suggestion regarding these 2 tests: can we switch one
of the 2 (e.g., global_ocean.90x40x15) to use a simpler advection scheme
in the vertical (e.g., 20 ?) to also test solve_tridiagonal.F ?
Right now both use 30 => pentadiagonal. What is your view on this ?

Cheers,
Jean-Michel

On Wed, Mar 14, 2012 at 09:06:39PM -0400, Patrick Heimbach wrote:
> Yes, fine with me.
> p.
> 
> On Mar 14, 2012, at 8:38 PM, Jean-Michel Campin wrote:
> 
> > Hi Patrick,
> > 
> > I will probably, for a short period of time, change the default
> > in model/inc/CPP_OPTIONS.h (define SOLVE_DIAGONAL_LOWMEMORY) 
> > so that I don't have to update few output.txt that change. 
> > The 2 AD experiments should be fine (since they have their own 
> > CPP_OPTIONS.h).
> > 
> > And when I am done with the few changes that are needed to use
> > the new default (i.e., undef SOLVE_DIAGONAL_LOWMEMORY), in:
> > pkg/gmredi/gmredi_calc_psi_bvp.F and pkg/ggl90/ggl90_calc.F 
> > I will put back the default CPP_OPTIONS.h to what you set.
> > 
> > I think there might be some way to reduce the truncation differences
> > between the 2 versions, and might be better to make these changes
> > now (before updating the output.txt).
> > 
> > Cheers,
> > Jean-Michel
> > 
> > On Wed, Mar 14, 2012 at 03:55:25PM -0400, Patrick Heimbach wrote:
> >> Update of /u/gcmpack/MITgcm/model/inc
> >> In directory forge:/tmp/cvs-serv5155
> >> 
> >> Modified Files:
> >> 	CPP_OPTIONS.h 
> >> Log Message:
> >> Add default CPP options for solve_*diagonal.F
> >> 
> >> 
> >> _______________________________________________
> >> MITgcm-cvs mailing list
> >> MITgcm-cvs at mitgcm.org
> >> http://mitgcm.org/mailman/listinfo/mitgcm-cvs
> > 
> > _______________________________________________
> > MITgcm-devel mailing list
> > MITgcm-devel at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> ---
> Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
> MIT | EAPS 54-1420 | 77 Massachusetts Ave | Cambridge MA 02139 USA
> FON +1-617-253-5259 | FAX +1-617-253-4464 | SKYPE patrick.heimbach
> 
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel