[[Imported from SVN: r3672]]

[[Imported from SVN: r3671]]

[[Imported from SVN: r3670]]

[[Imported from SVN: r3669]]

All tests should run through now again.

[[Imported from SVN: r3668]]

[[Imported from SVN: r3667]]

contains only some boundary segments with parametrizations, only
those have to be given explicitly.  The remaining ones are inserted
automatically as straight-line segments.

[[Imported from SVN: r3666]]

[[Imported from SVN: r3665]]

[[Imported from SVN: r3664]]

two files. So, sequential applications do not try to include MPI even if HAVE_MPI is true

[[Imported from SVN: r3663]]

Only comm() should be used

[[Imported from SVN: r3662]]

- fixed Task #67, P1 functions and operators should work level and leafwise now also in parallel

[[Imported from SVN: r3661]]

[[Imported from SVN: r3660]]

[[Imported from SVN: r3659]]

[[Imported from SVN: r3658]]

[[Imported from SVN: r3657]]

for complex<T> now ...

[[Imported from SVN: r3656]]

[[Imported from SVN: r3655]]

[[Imported from SVN: r3654]]

1) Generation of sparse matrix
2) Assembling of global stiffness matrix from local stiffness matrices
3) Interface for providing a local stiffness matrix and boundary conditions

All this is only for P1, but it works with hanging nodes, in parallel (with
algebraic extension of nonoverlapping grids) and for systems.
(Systems are yet untested).

[[Imported from SVN: r3653]]

[[Imported from SVN: r3652]]

that seperates it in a local (elemenwise) part and a global part

[[Imported from SVN: r3651]]

matrix row.

Compiles again and works for the sequential case.

[[Imported from SVN: r3650]]

[[Imported from SVN: r3649]]

[[Imported from SVN: r3648]]

[[Imported from SVN: r3647]]

[[Imported from SVN: r3646]]

method insertLinearSegment() is now private, as it is not supposed to be used from the outside any more

[[Imported from SVN: r3645]]

[[Imported from SVN: r3644]]

[[Imported from SVN: r3643]]

[[Imported from SVN: r3642]]

Cannot anybody test this definitely on a machine that has no MPI. I
simulated it by setting HAVE_MPI to false.

[[Imported from SVN: r3641]]

[[Imported from SVN: r3640]]

[[Imported from SVN: r3639]]

... well now my lib builds again and the test works aswell

... hope that's all ...

[[Imported from SVN: r3638]]

[[Imported from SVN: r3637]]

[[Imported from SVN: r3636]]

[[Imported from SVN: r3635]]

[[Imported from SVN: r3634]]

This is needed as one wants to setup only those indices involved in
the  communication.

First step in making AMG work with those smaller index sets.

[[Imported from SVN: r3633]]