[MITgcm-devel] testing

[Jean-Michel Campin]

Hi Martin,

Things are sligtly more complicated:
> >There also many experiments, that are not byte-exact (<16),
Since I made those changes: (between checkpoint57o_post & 57p_post):
in tag-index:
o Dissipation & phiHyd gradient are always added to gU,gV in timestep.F
  (was already the case for dissipation with mom_vecinv,
   and also the case for grad.PhiHyd when staggered-timeStep)
  This will allow to put dissipation out-off the AB time-stepping.
 Unfortunately, affects truncation error. Update output of exp. that "fails"
  (cg2d matching number) and leave the others for later: front_relax(10),
  global_ocean_pressure(10),ideal_2D_oce(10),internal_wave(9),lab_sea(12),
  matrix_example(7!),natl_box(12),plume_on_slope(11),vermix(12x3,11x2).
we lost few "full matching". 
But since I still have few little modifications to check-in that will 
affect some outputs at the truncation level, will do a full update
after.

Regarding fizhi, g77 is not "safe", and the reference output is
generated on faulks with pgf77.

And finally, It's the changes I made last Friday: 
from tag-index:
o add brackets in multi-terms sum to get same truncation, independent of
  the face number (manage to get exactly same value of gU,gV at the cs-edge)
  affects several test. exp. Update global_ocean.cs32x15 (+.thsice).
that are responsible for those 3 "fails":
> >MLDAdjust.0.leith
> >MLDAdjust.0.leithD
> >MLDAdjust.1.leith
Problem is that those exp. are quiet changing a lot those days,
and I wanted to see with Baylor what's the best thing to do.

And for the last one (but only this one):
> >MLDAdjust
It's exactly what Baylor described:
> Hi Martin,
>   The fails on MLAdjust are my fault.  I have made changes to the 
> horizontal viscosities and haven't updated the output.txt files yet.
>   Cheers,
>     -Baylor

Cheers,

Jean-Michel

> On Sep 26, 2005, at 3:36 AM, Martin Losch wrote:
> 
> >Hi there,
> >
> >when I run testreport on our Suns I get a lot of "FAIL"s (which I am 
> >not too concerned about), but I noticed that many experiments even 
> >fail on faulks with g77 (which I thought was the reference machine, 
> >but maybe I am wrong about this?). It would probably be helpful if the 
> >verification experiments all passed on at least on platform, wouldn't 
> >it?
> >These fail on faulks with linux_ia32_g77:
> >MLDAdjust
> >MLDAdjust.0.leith
> >MLDAdjust.0.leithD
> >MLDAdjust.1.leith
> >fizhi-cs-32x32x10
> >fizhi-cs-aqualev10
> >There also many experiments, that are not byte-exact (<16), which 
> >shouldn't happen on the reference platform, should it? E.g.,
> >the first 3 aim.5l experiments
> >cfc_example
> >exp0-5
> >rotating_tank
> >solid-body.cs-32x32x1
> >and a few others,
> >Maybe I am a bit too picky about this, but it helps (me) a lot to know 
> >that on a specific platform the experiments work exactly, before I 
> >mess up the code (o:
> >
> >Martin
> >
> >_______________________________________________
> >MITgcm-devel mailing list
> >MITgcm-devel at mitgcm.org
> >http://mitgcm.org/mailman/listinfo/mitgcm-devel
>