Roland Balanced domain decomposition is not used by default. It should, in principle, lead to better decompositions than Carpet's current algorithm. You probably found a bug in unused code.
-erik On Tue, Sep 12, 2017 at 7:39 PM, Roland Haas <[email protected]> wrote: > Hello all, > > trying to hack Carpet's regrid scheme to be more multi-patch friendly, > I had a look at the funtion region_t::split which has (in line 104ff of > region.cc): > --8<-- > outer_boundaries[idir][1] = false; > newreg.outer_boundaries[idir][0] = false; > --8<-- > where idir is computed above as > --8<-- > // Choose a direction (prefer the z direction) > int const idir = maxloc1(extent.shape()); > --8<-- > ie idir runs 0..2 for x..z. However outer_boundaries is a b2vect so the > order of indices is [face][dim] where face is upper/lower face and dim is > the dimension xyz. > > The split function is used in CarpetLib/balance.cc:39 > > return item.split(ratio_new_over_old); > > Now I am wondering if I found a bug that affects the balanced splitting > code or just a bug in dead code or no bug at all but am confused about > the code. > > Yours, > Roland > > -- > My email is as private as my paper mail. I therefore support encrypting > and signing email messages. Get my PGP key from http://pgp.mit.edu . > > _______________________________________________ > Developers mailing list > [email protected] > http://www.cactuscode.org/mailman/listinfo/developers > > -- Erik Schnetter <[email protected]> http://www.perimeterinstitute.ca/personal/eschnetter/
_______________________________________________ Developers mailing list [email protected] http://www.cactuscode.org/mailman/listinfo/developers
