[[Imported from SVN: r1201]]

[[Imported from SVN: r1200]]

[[Imported from SVN: r1199]]

[[Imported from SVN: r1198]]

[[Imported from SVN: r1197]]

[[Imported from SVN: r1195]]

[[Imported from SVN: r1194]]

This makes the memory layout contiguous and removes one level of pointer indirection.
Since the Matrix class is used by some as the element stiffness matrix we may expect
some general speedup.  I didn't measure, though.  Further speedup may be possible
by using the fact that all rows have the same length. This will need more programming,
though.

This fixes FS 623.



[[Imported from SVN: r1193]]

[[Imported from SVN: r1192]]

[[Imported from SVN: r1191]]

[[Imported from SVN: r1190]]

[[Imported from SVN: r1189]]

[[Imported from SVN: r1187]]

[[Imported from SVN: r1186]]

[[Imported from SVN: r1185]]

[[Imported from SVN: r1183]]

[[Imported from SVN: r1182]]

This prepares for changes in superlu.m4 that make usage of SuperLU optional,
even if SuperLU was found by configure.

[[Imported from SVN: r1180]]

Distribute matrixmarket.hh
AMG: make x consistent even if there is only one level
galerkin.hh: Initialize Matrixblock correctly
splvers.hh: Always call pre mthod of the preconditioner _before_ using
the operator. Thus x is consistent and the computed defect will be,
too.
matrixmarket.hh: use int instead of size_t as template parameter as
this is the case for the FieldMatrix, too.

[[Imported from SVN: r1179]]

  const_iterator didn't work

merge to release

[[Imported from SVN: r1177]]

[[Imported from SVN: r1176]]

merge to release

[[Imported from SVN: r1175]]

merge to release

[[Imported from SVN: r1174]]

merge to release

[[Imported from SVN: r1173]]

merge to release

[[Imported from SVN: r1172]]

[[Imported from SVN: r1170]]

[[Imported from SVN: r1168]]

[[Imported from SVN: r1167]]

Use MPIHelper instead of MPI directly. Should make make headercheck
succeed even if MPI is not installed. 

Merge to release to fix the autobuilds.

[[Imported from SVN: r1166]]

[[Imported from SVN: r1165]]

[[Imported from SVN: r1160]]

[[Imported from SVN: r1159]]

the Thomas algorithm to solve block-tridiagonal systems of equations.

NOTE:  Yes, I know that this is not the time to add new features.
However, tridiagonal matrices and the Thomas algorithm are important
components in line smoothers for anisotropic problems.  We will
need that soon for the AdaptHydroMod project, and therefore I would
like to have this matrix in the 2.0 release.

Also, this patch is not invasive at all: it just adds another matrix
class.

Please merge this to the release branch.


[[Imported from SVN: r1158]]

[[Imported from SVN: r1157]]

[[Imported from SVN: r1156]]

Now the hierarchy building is a factor 2 faster than previously. (And
as fast as in the beginning of October 2009).

[[Imported from SVN: r1153]]

This does not introduce any semantic change but allows to
save memory. If a matrix is copied the sparsity pattern is
reused. If the pattern of a copy is changed it gets newly
allocated and does not influence the original matrix.

Patch by Martin Weiser from ZIB

[[Imported from SVN: r1152]]

[[Imported from SVN: r1151]]

[[Imported from SVN: r1150]]

[[Imported from SVN: r1149]]