[MITgcm-devel] autodiff_store/restore/ALLOW_ECCO_EVOLUTION

Mon Mar 10 09:37:03 EDT 2014

Hi Jean-Michel,

not a good idea, since Gael and I (if I am not mistaken) wanted this specific experiment to be close to an ECCO setup. Can you please be patient for a few days until some things have cleared up, and we can take a closer look.

Thanks
-p.

On Mar 10, 2014, at 9:34 AM, Jean-Michel Campin <jmc at ocean.mit.edu> wrote:

> Hi Martin,
> 
> I will try to get (at least) one AD experiment (global_ocean.cs32x15)
> to work without ECCO_CPPOPTIONS.h. The next one on the list would be 
> lab_sea. It might take some time since we have many CPP options in 
> these pkgs that are not very well documented.
> 
> I am considering also to replace most of ALLOW_AUTODIFF_TAMC
> with ALLOW_AUTODIFF (except for TAF storage directives),
> and replace also most of ALLOW_AUTODIFF_OPENAD with ALLOW_OPENAD.
> These changes are all related, and will allow to add few less
> #include AUTODIFF_OPTIONS.h in case ECCO_CPPOPTIONS.h is not used.
> 
> Cheers,
> Jean-Michel
> 
> On Fri, Mar 07, 2014 at 05:06:51PM +0100, Martin Losch wrote:
>> Hi Jean-Michel,
>> 
>> thanks for the clarification. I just like to use the same options for my adjoint model as for my forward integration. I was OK with having the autodiff/ctrl/cost options in one file, because these packages are really very much connected (at least I could not think of a situation where you would use only one of them), but I do agree with you on the points you made.
>> 
>> The simplest fix appears to replace the #include “ECCO_CPPOPTIONS.h” by this:
>> #ifdef PACKAGES_CONFIG_H
>> C# include "ECCO_CPPOPTIONS.h"
>> # ifdef ALLOW_AUTODIFF
>> #  include "AUTODIFF_OPTIONS.h"
>> # endif 
>> # ifdef ALLOW_CTRL
>> #  include "CTRL_OPTIONS.h"
>> # endif 
>> # ifdef ALLOW_COST
>> #  include "COST_OPTIONS.h"
>> # endif 
>> # ifdef ALLOW_ECCO
>> #  include “ECCO_OPTIONS.h"
>> # endif 
>> #endif
>> 
>> At least I was able to compile my configuration with this. It looks more involved to actually put the individual *_OPTIONS.h in the right places in the code. My first attempt failed and it’s annoying to debug, because you need TAF to do the error analysis (a lot of problems in the CADJ directives that the normal compiler does not see).
>> 
>> Martin
>> 
>> On Mar 7, 2014, at 4:58 PM, Jean-Michel Campin <jmc at ocean.mit.edu> wrote:
>> 
>>> Hi Martin,
>>> 
>>> I think that with some work, we should be able to put all the 
>>> CPP options from ECCO_CPPOPTIONS.h in their associated pkg
>>> option files (AUTODIFF_OPTIONS.h COST_OPTIONS.h CTRL_OPTIONS.h
>>> ECCO_OPTIONS.h and EXF_OPTIONS.h). This was the purpose of
>>> adding an option file "ECCO_OPTIONS.h" (in Aug 2012) to make a clear 
>>> distinction between pkg/ecco option file and the "special" multi-pkg
>>> option file ECCO_CPPOPTIONS.h
>>> 
>>> In the mean time, I feel like it could become confusing to 
>>> do half the work, i.e., to just move pkg/exf options
>>> in EXF_OPTIONS.h but to keep ECCO_CPPOPTIONS.h for all others pkg
>>> (autodiff, cost, ctrl, ecco) options:
>>> 
>>> 1) it introduce a 3rd way a specifying CPP options that
>>> is in contradiction with the current ECCO_CPPOPTIONS.h documentation.
>>> For instance, in each ECCO_CPPOPTIONS.h we have these comments at the top:
>>> 
>>> C-- Collect here, in a single option-file, options to control which optional
>>> C   features to compile in packages AUTODIFF, COST, CTRL, ECCO, CAL and EXF.
>>> C   If used, this option-file needs to be directly included in CPP_OPTIONS.h
>>> C   Although this method, inherited from ECCO setup, has been traditionally
>>> C   used for all adjoint built, work is in progess to allow to use the
>>> C   standard metod (each of the above pkg get its own options from its
>>> C   specific option-file) also for adjoint built.
>>> 
>>> but this would need to be changed (in some places ? but not every-where ?)
>>> 
>>> 2) when we multiply the number of different ways to make these setting,
>>> it makes more complicated to do the right check/stop.
>>> This is a more general problem (in MITgcm development) regarding user
>>> setting changes (going through multiple little changes which all 
>>> affect user settings instead of a clear, 1 step, update from 
>>> old setting to new one).
>>> 
>>> 3) a cleaner way to introduce this new method (the 3rd way) would be to 
>>> use a different file name rather than to change the meaning of
>>> the file ECCO_CPPOPTIONS.h 
>>> 
>>> Cheers,
>>> Jean-Michel
>>> 
>>> On Fri, Mar 07, 2014 at 03:24:50PM +0100, Martin Losch wrote:
>>>> Hi again,
>>>> 
>>>> I don’t know about the utility of ALLOW_ECCO_EVOLTION, but I think I solved my problem, in case anyone is interested:
>>>> 
>>>> I prefer that have the cpp-flags for the exf-pkg in the EXF_OPTIONS.h file. This somehow clashes with the convention that for adjoint models these cpp-flags should now be in ECCO_CPPOPTIONS.h. This convention, which I’d like to challenge (again), has been cemented a little more by these lines 
>>>> #ifdef ECCO_CPPOPTIONS_H
>>>> dont do anythin
>>>> #else
>>>> here are the exf flags
>>>> #endif
>>>> that I found in my EXF_OPTIONS.h (which I grabbed from MITgcm_contrib/arctic/cs_36km/code, but it’s also in pkg/exf/EXF_OPTIONS.h).
>>>> With this if-construct, EXF_OPTIONS.h are never loaded after ECCO_CPPOPTIONS.h has been loaded. In this sense the 
>>>> #ifdef ALLOW_ECCO_EVOLUTION
>>>> #ifdef ALLOW_EXF
>>>> #include “EXF_OPTIONS.h”
>>>> #endif
>>>> #endif
>>>> has no effect, because ALLOW_ECCO_EVOLUTION is typically defined in ECCO_CPPOPTIONS.h
>>>> 
>>>> My solution is to remove the if statement from the EXF_OPTIONS.h or should I rather remove ECCO_CPPOPTIONS.h and distribute the flags in there to AUTODIFF_OPTIONS.h/CTRL_OPTIONS.h/COST_OPTIONS.h/EXF_OPTIONS.h
>>>> What’s the preferred method now?
>>>> 
>>>> Martin
>>>> 
>>>> On Mar 5, 2014, at 12:52 AM, Patrick Heimbach <heimbach at mit.edu> wrote:
>>>> 
>>>>> Hi Martin,
>>>>> 
>>>>> I can take a look after Thursday (currently on travel).
>>>>> The flags seem a little messed up.
>>>>> 
>>>>> p.
>>>>> 
>>>>> On Mar 4, 2014, at 12:29 PM, Martin Losch <Martin.Losch at awi.de> wrote:
>>>>> 
>>>>>> Hi there,
>>>>>> 
>>>>>> I have a setup for which I do not want to use the ecco package, but seaice, exf, etc.
>>>>>> 
>>>>>> I think that autodiff_store/restore do not compile without ECCO, at least not in my case because the
>>>>>> #ifdef ALLOW_ECCO_EVOLUTION 
>>>>>> prevents loading the EXF_OPTIONS.h, that in turn prevent defining ALLOW_RUNOFF and the variable runoff is not defined (line 548).
>>>>>> 
>>>>>> How do I fix that so that it doesn’t break things for ECCO or anyone else? Remove the #ifdef ALLOW_ECCO_EVOLUTION? so that EXF_OPTIONS.h are always included as long as ALLOW_EXF is defined?
>>>>>> 
>>>>>> Martin
>>>>>> 
>>>>>> 
>>>>>> _______________________________________________
>>>>>> 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
>>>> 
>>>> 
>>>> _______________________________________________
>>>> 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
>> 
>> 
>> _______________________________________________
>> 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

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 1588 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20140310/28643b98/attachment.p7s>