dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Johan Hake wrote:
> On Wednesday 19 August 2009 18:50:13 Garth N. Wells wrote:
> > Johan Hake wrote:
> > > On Wednesday 19 August 2009 17:29:52 Garth N. Wells wrote:
> > > > What's going on behind the scene when I copy one vector to another in
> > > > PyDOLFIN using
> > > >
> > > >      u0.vector()[:] = u.vector()[:]
> > > When I add:
> > >
> > > u2 = Function(V)
> > > u2.vector()[:] = u.vector()[:]
> > >
> > > # Plot solution
> > > plot(u2)
> > >
> > > In a the Poisson demo it works fine.
> > In parallel?
>
> Yupp!
>
> > > This type of slice only uses the assignment operator.
> > The GenericVector assignment operator?
>
> Yes. This is done in __setslice__/__getslice__ in the extended python class.

I don't see then why the we end up inside

  _compare_vector_with_vector

for the Cahn-Hilliard demo when running in parallel?

> > I delved into the problem and the program ends up in the function
> > _compare_vector_with_vector from dolfin_la_get_set_items.i. It uses
> > GenericVector::get and GenericVector::set. These need to be used with
> > caution in parallel.
>
> Yes, when other type of slices are used we need to find a way to do that in 
> parallel. I have just started digging in the parallel code, and need to look 
> into all the changes to get familiar with it before I try to solve this.

This shouldn't be too hard.

   set(const double* block, uint m, const uint* rows)

works in parallel but

   get(double* block, uint m, const uint* rows)

doesn't yet (not too hard to fix though). The really evil functions are

  set(const double*)

and

  get(double*)

Garth


> The best solution would be to have a full fledge VectotView in place ;)
>
> Johan
>
> > Garth
> >
> > Other slices might not
> >
> > > work. Have not checked.
> > >
> > > Johan
> > >
> > > > ? The issue is that it doesn't work in parallel.
> > > >
> > > > Garth
> > > > _______________________________________________
> > > > DOLFIN-dev mailing list
> > > > DOLFIN-dev@xxxxxxxxxx
> > > > http://www.fenics.org/mailman/listinfo/dolfin-dev