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]]

int as type for the field of scaledidmatrix causes ambiguities because there are no std::sqrt(int) and std::fabs(int)

[[Imported from SVN: r1148]]

icc does not seem to no associated type value_type of an iterator. Now we use the iterator_traits to extract it.

[[Imported from SVN: r1147]]

[[Imported from SVN: r1146]]

[[Imported from SVN: r1145]]

[[Imported from SVN: r1144]]

Fixed some more bugs.
Patch was provided by Atgeirr Rasmussen.

[[Imported from SVN: r1143]]

[[Imported from SVN: r1142]]

should run make check _before_ committing.

[[Imported from SVN: r1141]]

[[Imported from SVN: r1140]]

[[Imported from SVN: r1139]]

[[Imported from SVN: r1138]]

[[Imported from SVN: r1137]]

[[Imported from SVN: r1136]]

Sync with my version.

[[Imported from SVN: r1135]]

Support for domains consisting of pairs connected by an edge of the
matrix graph.

[[Imported from SVN: r1134]]

removed warnings.

[[Imported from SVN: r1133]]

[[Imported from SVN: r1132]]

[[Imported from SVN: r1131]]

[[Imported from SVN: r1130]]

*add calls to mv() and mtv()
*also test ScaledIdentityMatrix and DiagonalMatrix

[[Imported from SVN: r1129]]

[[Imported from SVN: r1128]]

* Removed ugly ReferenceStorageIterator since
  * it stored either a reference or a fake reference
  * you could not use iterator->method
  * you could not assign iterators
* Introduce ContainerWrapperIterator
  * always stores an assignable ContainerWrapper object
  * allows iterator->method()
  * can be used for any other sparse row type

[[Imported from SVN: r1127]]

[[Imported from SVN: r1126]]

This method should really be removed in favor of mtv().

[[Imported from SVN: r1125]]

accumulate to one process on the finest level.

[[Imported from SVN: r1124]]

of all non-template and un-inlined functions. As a quick hack we qualified them all with
the inline keyword.

The better way would probably be to skip inline where not needed and create an istl lib. But that is bigger effort.

[[Imported from SVN: r1123]]

[[Imported from SVN: r1122]]

[[Imported from SVN: r1121]]

[[Imported from SVN: r1120]]

[[Imported from SVN: r1119]]

[[Imported from SVN: r1118]]

[[Imported from SVN: r1117]]

[[Imported from SVN: r1116]]

[[Imported from SVN: r1115]]

[[Imported from SVN: r1114]]

* doxument the _unused_ paramter tryHard

[[Imported from SVN: r1113]]