[petsc-dev] [petsc-maint #89809] 3D DMDAs in petsc-3.2

[Barry Smith]

  I have #if defined() out all these changes since they break the "normal case". Once there is a fix for the issue I can put them back in.

   Barry

On Oct 9, 2011, at 11:01 PM, Jed Brown wrote:

> On Sun, Oct 9, 2011 at 22:54, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>   Jed,
> 
>       It appears your  some of your many
> 
> if (s_t >= x*y*z) {s_t = bases[n26] + (i-1)*x_t;} /* 2d case */
> 
> things are breaking the communication with a stencil width of 2
> 
> da3.c:980
> M-x xhg-ann<RET>
> 
> Looks like that line comes from
> 
> changeset:   18790:4310a5aad280
> user:        Ethan Coon <ecoon at lanl.gov>
> date:        Wed Apr 20 00:57:49 2011 +0200
> summary:     cruft to handle the special case of solving a 2D problem in a 3D algorithm with stencil size > 1.  This only works in the 3D case, z-dimension
> 
> http://petsc.cs.iit.edu/petsc/petsc-dev/rev/4310a5a
>  
> the second layer of ghost points uses the first layer instead of getting the second layer.
> 
> Since I don't understand your /* 2d case */ business I leave it to you to fix
> 
> I don't understand it either and haven't tried. Maybe Ethan can take a look.
>