[Mitgcm-support] latest on DIVA with 1.4.26

[mitgcm-support at dev.mitgcm.org]

Hi Ralf,

I've checked in a few changes into the ecco-branch
which refer to DIVA.
They solve the issues of
- correctly dealing with TAF's '-pure' option
- correctly dealing with TAF's '-mpi' option
- store the cost function value which is lost in
   consecutive DIVA intervals
- calling ctrl_unpack/pack,grdchk_main only at the
   appropriate times.

I've also made a modification in the ecco code to modify
the current strategy of
1. computing fc followed by first adjoint DIVA interval
    in 1st model execution
2. computation of individual DIVA intervals only
    in consecutive executions
to the following
1. computing fc only in first execution
2. computation of individual DIVA intervals only
    in consecutive executions

The reason is to be able to better tune queue sizes
(the first execution would otherwise always be
much longer than the following ones).
To make this work I changed the
CADJ INIT onetape = common, 1
CADJ INIT onetape = user
i.e. I rather store mytime to file, otherwise the
values get lost.
And I avoid the storing of myiter (integer).
I've tested it and it works fine.

So now the only hand fix that remains is the bracketing
of the adcost_final ... adcost_averagesfields block by
       if (idivbeg .ge. nchklev_3) then
       ...
       endif

Cheers
-Patrick

PS:
The relevant modif's are in
pkg/ecco/ecco_the_main_loop.F
pkg/ecco/ecco_the_model_main.F
pkg/cost/cost_averagesfields.F
pkg/cost/cost_init.F
pkg/cost/cost_final_store.F
pkg/cost/cost_final_restore.F

-- 
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
Patrick Heimbach ... Massachusetts Institute of Technology
FON: +1/617/253-5259 .................. EAPS, Room 54-1518
FAX: +1/617/253-4464 ............. 77 Massachusetts Avenue
mailto:heimbach at mit.edu ............... Cambridge MA 02139
http://www.mit.edu/~heimbach/ ..................... U.S.A.