[MITgcm-devel] main_do_loop and DIVA

[Matthew Mazloff]

Hi Patrick

Thanks for working on this!

So I haven't run the executable, but it looks OK with one exception. There is an odd and hidden recomputation. In the part "C skip adjoint before divided loop" it is calling main_do_loop. I assume that call should not be there (and I could easily comment it out). I am not sure if that recomp is specific to my setup, which is definitely possible, or if you have that too…let me know

Matt






On May 30, 2013, at 2:41 PM, Patrick Heimbach <heimbach at MIT.EDU> wrote:

> Hi Matt (and Jean),
> 
> I've made another change to the_main_loop and main_do_loop.
> In part, I've reverted the time-stepping which was mainly meant to enable revolve loop for OpenAD.
> What's done now is that we keep the time-stepping loop in the_main_loop when using TAF
> and have it in main_do_loop when using OpenAD (changes are actually rather minor, 
> and perhaps even a bit less convoluted).
> 
> @ Matt: please take a look whether the fix allows you to build your config with latest
> 
> @ Jean: I tested the change with OpenAD on one of the verifications, seems to work fine, but might be worth double-checking
> 
> @ All: we should keep a verification experiment that tests DIVA.
> Right now I've tested two: lab_sea (which was the closest to old natl_box_adjoint)
> and global_ocean.cs32x15 (which is closest to new ECCO v4).
> We'll need to make a decision of which config to use, and either modify an existing verif
> or add a new one (since other compile/build options are needed).
> 
> Cheers
> -Patrick
> 
> On May 17, 2013, at 12:35 PM, Matthew Mazloff <mmazloff at ucsd.edu> wrote:
> 
>> OK -- one hint is that it is opening level 4 tape files in the_main_loop -- so that is a problem if we want to start the divided loop in main_do_loop
>> 
>> the ordering is definitely off...
>> 
>> I will get back to this later -- have to run now -- but it looks like we may have to move some tape calls…
>> 
>> if you have any suggestions I would greatly appreciate it!
>> 
>> Matt
>> 
>> 
>> 
>> On May 17, 2013, at 9:07 AM, Matthew Mazloff <mmazloff at ucsd.edu> wrote:
>> 
>>> Hi Patrick
>>> 
>>> For some reason TAF is ignoring the 
>>> 
>>> CADJ loop = divided
>>> 
>>> statement in main_do_loop.F
>>> 
>>> Has anyone else seen this issue - does it have to do with my staf options. (It was working fine before in the_main_loop.F.)
>>> 
>>> Or is it just a TAF bug and I need to email them
>>> 
>>> Thanks for your help
>>> Matt
>>> 
>>> 
>>> 
>>> _______________________________________________
>>> MITgcm-devel mailing list
>>> MITgcm-devel at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>> 
>> 
>> _______________________________________________
>> 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