[Users] [Einstein Toolkit] #446: 1D HDF5 output nonfunctional with multipatch - std::out_of_range

[Erik Schnetter]

Roland

Not, it isn't! The dh class stores

  light_mboxes light_boxes;     // grid hierarchy [ml][rl][c]
  local_mboxes local_boxes;     // grid hierarchy [ml][rl][lc]
  fast_mboxes fast_boxes;       // grid hierarchy [ml][rl][p]

where the light boxes (light, little content) contain information
about every component (even remote ones), the local boxes contain
(much more) information about the process-local components, and the
fast boxes (misnomer) contain the (process-local relevant part of the)
communication schedule.

We could store also the per-level information. I assume it is much
smaller, since it contains information only about levels (and not
components):

  level_mboxes level_boxes; // grid hierarchy [ml][rl]

and then clear the data structure instead of declaring it in
dh::regrid, i.e. treat it similar to fast_mboxes instead of similar to
full_mboxes.

There is another data structure, full_mboxes, which contains all the
information you ever wanted to know about everything in the grid
hierarchy. However, this costs too much memory, and is therefore only
allocated temporarily for a single level, and is freed again right
after it has been used. This distinction between these different kinds
of information and not storing all of it is what makes hg Carpet
faster and use less memory.

-erik

On Mon, Jun 6, 2011 at 6:44 AM, Roland Haas
<roland.haas at physics.gatech.edu> wrote:
> Hello Ian, Erik,
>
>>  Could the routine GetAllActive be replaced by just accessing
>>  level_dboxes::active?
> It could yes. GetAllActive computes level_dboxes::active (it dpulicates
> code from dh.cc, expanding the code in one of the bboxset constructors
> taking a vector and a pointer-to-member as arguments). I am not sure
> though if this information is actually stored anywhere outside of
> dh::regrid (in level_boxes and the various references to parts of it).
> Erik: is the "active" information stored permanently anywhere?
>
> Yours,
> Roland
>
>
> _______________________________________________
> Users mailing list
> Users at einsteintoolkit.org
> http://lists.einsteintoolkit.org/mailman/listinfo/users
>
>



-- 
Erik Schnetter <schnetter at cct.lsu.edu>   http://www.cct.lsu.edu/~eschnett/