Patch by Andreas Lauser

[[Imported from SVN: r1627]]

[[Imported from SVN: r1626]]

[[Imported from SVN: r1623]]

[[Imported from SVN: r1622]]

[[Imported from SVN: r1620]]

[[Imported from SVN: r1618]]

[[Imported from SVN: r1609]]

[[Imported from SVN: r1608]]

[[Imported from SVN: r1601]]

[[Imported from SVN: r1600]]

[[Imported from SVN: r1599]]

Thanks to Andreas Lauser for pointing this out.

[[Imported from SVN: r1591]]

MPI_Finalize. Now this is a valid MPI program.

[[Imported from SVN: r1585]]

During redistribution of the data objects were created with new and
deallocated using the allocator. This works for the standard allocator
but might break with others. This fix prevents this behaviour by using
the allocator for the creation and allocation, too.
Additionally, it prevents false positive during memory leak detection
of valgrind. 

Should fix flyspray #1047!

I'll leave it to the relase manager to decide whether to merge this to
the release.

[[Imported from SVN: r1581]]

[[Imported from SVN: r1579]]

Since FieldVector<T,1> should be usable like T this should work.
Currently this fails due to an ambigous overload (see fs#1024).

[[Imported from SVN: r1572]]

[[Imported from SVN: r1570]]

[[Imported from SVN: r1569]]

[[Imported from SVN: r1568]]

Sort included headers and remove some empty lines from superlutest.cc and matrixmarkettest.cc.

[[Imported from SVN: r1567]]

MM_SEQUENTIAL were not correctly replaced by #if HAVE_MPI.


[[Imported from SVN: r1566]]

[[Imported from SVN: r1565]]

Christoph's issues.

[[Imported from SVN: r1562]]

[[Imported from SVN: r1561]]

sequential or not.

[[Imported from SVN: r1560]]

[[Imported from SVN: r1554]]

Should fix flyspray #1093

[[Imported from SVN: r1553]]

[[Imported from SVN: r1552]]

Unfortunately DUNE_DEPRECATED does not work before definitions of
template member functions inside the class.

[[Imported from SVN: r1551]]

[[Imported from SVN: r1550]]

[[Imported from SVN: r1549]]

wrapper of SuperLU solver.

[[Imported from SVN: r1548]]

direct solver.
Added a method to AMG to check whether it uses a direct solver on the
coarse level.

[[Imported from SVN: r1547]]

communicate result of coarse solver to other processes, otherwise catching the exception leads to a deadlock

[[Imported from SVN: r1546]]

[[Imported from SVN: r1545]]

Thanks to Uli Sack for the Patch.

[[Imported from SVN: r1544]]

[[Imported from SVN: r1542]]

* use std::abs and std::sqrt instead of fabs and sqrt
* use real_type instead of double to (hopefully) fix probelms with std::complex

[[Imported from SVN: r1540]]

remove tabs from solvers.hh and fix indentation

[[Imported from SVN: r1539]]

[[Imported from SVN: r1538]]