[MITgcm-support] running the adjoint on openmpi

[Sarah Zedler]

Hi,

I'm trying to compile and run the mit and its' adjoint using openmpi.  I am
using an old version of the MIT code (we can't move to the new one, because
we have extensively modified the old code to work for hourly assimilation of
data).

First, I am wondering if anyone has successfully run the mitgcm and its
adjoint on openmpi.  I was able to get the code to compile and run, but I
get different derivatives between my mpich test case and the openmpi test
case.  The fields are similar; the largest differences seem to be on the
seams of the grid (where information is passed). The output to the tbar
files (from the forward code) is identical between the runs, as is the
costfunction, so I think it is something with the adjoint.

I would be very appreciative if anyone has anything useful to say about
this, as I am not sure what to try next.

Best regards,
Sarah
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20100729/87518d51/attachment.htm>