[Carpet] Saving memory

[Ian Hinder]

Erik Schnetter wrote:
> On Mar 6, 2008, at 20:10:18, Ian Hinder wrote:
> 
>> Erik Schnetter wrote:
>>> On Apr 21, 2006, at 05:14:03, Ian Hawke wrote:
>>>> It is already possible to choose the prolongation method at runtime by
>>>> choice of parameter. In order to do this you set, e.g.,
>>>>
>>>> tags='ProlongationParameter="Whisky::Whisky_Prolongation"'
>>>>
>>>> in the interface.ccl for the appropriate groups, and then have the
>>>> appropriate parameter
>>>>
>>>> # Prolongation by Carpet etc.
>>>>
>>>> STRING Whisky_Prolongation "The prolongation operator used by Carpet"
>>>> {
>>>>  "ENO"  :: "Second order ENO operators"
>>>>  "TVD"  :: "Second order TVD operator - does not work well"
>>>>  "WENO" :: "Fifth order WENO operators"
>>>>  ".*"   :: "Anything else"
>>>> } "ENO"
>>>>
>>>> in the parameter.ccl. I believe it has to be a string, not a keyword,
>>>> but am not sure.
>>>>
>>>> This modification would be the simplest way to work; Bela would then
>>>> specify prolongation="none" for the ADMBase variables in the parameter
>>>> file.
>>>
>>> Béla,
>>>
>>> the enclosed patch for ADMBase should do just this.  After applying it,
>>> you have to set
>>>
>>> ADMBase::prolongation_type = "none"
>>>
>>> in your parameter file.  Can you try this and tell me whether it works?
>>
>> Can this patch be applied to CVS?  It will be very useful as the new
>> default in Carpet for enable_all_storage is "no".
> 
> 
> Ian,
> 
> preparing patches is a bit of work, since one has to make sure that
> things don't break.  Do you want to help out?  In particular, you would
> need to apply the patch to CactusEinstein, update the documentation,
> potentially add a test case, ensure that it compiles and does the right
> thing for you, that the default settings still work with old parameter
> files, and then send e.g. a diff e.g. to <developers at cactuscode.org>. 
> This may seem like overkill, but apart from using cvs+patch I don't see
> how this could be simplified.  This would basically put the details and
> the burden of testing onto you.

Indeed.  I will investigate the memory saving and see if it is worth the
effort.

> I was just looking through the archive, and it seems that my proposed
> patch was incomplete, as remarked by Thomas.  Anyway, since this patch
> is quite old by now, we shouldn't just apply it.

I cannot find such a message from Thomas - maybe it was to you only?

http://lists.carpetcode.org/archives/developers/2006-April/thread.html

The patch you sent adds a tensor metric to the variables, which is not
required for this patch, as far as I can tell?  Other than that, what is
incomplete about the patch?  I have applied it to my own tree and it
appears to work.

-- 
Ian Hinder
hinder at gravity.psu.edu
http://www.gravity.psu.edu/~hinder