[Carpet] sync vs sync+restrict

[Bela Szilagyi]

Erik,

indeed I meant prolongation rather than restriction.

Turning off the prolongation is worth the effort in an excision thorn of mine 
that goes through the excision points layer by layer, doing up to 7 extra 
syncronizations within a single MoL subiteration.  Having prolongation turned 
off (by means of the aliased function EnableProlongation) is a significant 
gain for me.

While looking at this I realized that MoL uses the same function to define 
when should prolongation occur.  And why thinking of this came the idea that 
one could construct fully Xmas tree algorithms that would not need constant 
time-steps.

This is all.



On Thursday 04 May 2006 17:55, Erik Schnetter wrote:
> On Apr 27, 2006, at 10:07:51, Bela Szilagyi wrote:
> > Can one steer the synchronization + restriction proccess within
> > carpet by
> > means of adjusting the appropriate tags table entries at runtime?
> >
> > To be more precise, I would want that some of my synchronization
> > requests
> > would be honored without the additional expense of restriction.
> > Can this be
> > done from outside Carpet by tags table entry manipulation?  Or, is
> > it known
> > that the most expensive part of combined sync+restriction is mostly
> > the
> > syncing?
>
> To be clear, restricting means copying fine grid values to coarse
> grid values.  Synchronisation happens e.g. whenever there is a SYNC
> statement in a schedule item.  Restriction happens at a defined
> point, namely between the EVOL and POSTSTEP bins.  Both are
> completely independent.  I don't know which is more expensive --
> synchronisation is called more often, but restriction has to move
> more data.
>
> You may be talking about prolongation instead.  Prolongation is
> indeed more expensive than synchronisation, and it makes certainly
> sense for certain analysis quantities to not be prolongated.  There
> is a tags table entry for this, but its value is examined only at
> startup.  You cannot change that during evolution.
>
> Are you sure that this is worth the effort?  Are you trying to see
> how much this would improve things, or do you have timing numbers
> showing that this prolongation is indeed worth switching off?
>
> -erik

-- 
Bela Szilagyi
----------------------------------------------------
Max-Planck-Institut für Gravitationsphysik
Albert-Einstein-Institut
Tel: +49 331 567 7632
Fax: +49 331 567 7649
----------------------------------------------------