Use correct type for second template argument of `std::array`

See merge request !90

Using `unsigned long` instead of `size_t` only works on architectures
where the two types are identical, but not on, for example, i386.  On
architectures where the types differ, the generic version would be
used instead of the specialization for `std::array`.

Bug-Debian: https://bugs.debian.org/860666

[bugfix] add missing header

See merge request !89

Feature/dynamicpolymorphism

See merge request !84

the two methods weren't needed up to now. After switching to dynamic
construction of the scalar product, they have to be defined, but still
they should not be called (currently). Thus we implement them with an
assert(false).

getSolverCategory() is used by the AMG communication layer. For convinience
we add this new specialization of category(...)

the enum on the Communication object was never used,
thus we drop it immediately.

depending on a preprocessor define we switch the base classes to be
pure virtual, or implement a dummy category(). If the classes are pure
virtual the old interface with an enum is not possible anymore.

To enable/disable the backwards compatibility check we provide a cmake
flag DUNE_ISTL_SUPPORT_OLD_CATEGORY_INTERFACE, which defaults to 1
(e.g. backwards compatibility enabled).

- we add helper code to extract the solver category from the enum
  or the virtual function
- we update the base classes to actually use this code to perform
  consistency checks

We introduce a new method category to check for the same SolverCategory of different
components at run time.

This is a drawback compared to compile time checks, but it adds a lot of flexibility as we will
be able to  write factory classes for solvers and compose/exchange solvers at runtime.

Besides this we added the following changes:
- an additional intermediate class IterativeSolver encapsulates the data of almost all solvers
- the solvers now inherit from IterativeSolver and can usually reuse its constructor
- the solvers now only have to implement _one_ apply method, the other is implemented
  generically

State:
- currently the example compiles
- none of the AMG headers has been updated yet

[test] We removed the deprecated constructors of RestartedGMResSolver. Don't test the old constructor.

Feature/improve simd handling

See merge request !86

[cleanup] replace deprecated type trait

See merge request !82

Felix Gruber authored 8 years ago and Felix Gruber committed 8 years ago

The enable_if implementation defined in dune-common is deprecated.

Change module version number to 2.6-git

See merge request !81

Feature/simd for multi rhs

Update the solvers to solve for multiple rhs vectors simultaniously. This makes use of SIMD abstractions in dune-common (see core/dune-common!16) to use vectors of SIMD vectors as `FieldType` of the right-hand-side. By this we can make immediate use of SIMD units in modern processors. Using SSE for example we get a speedup of 2 basically for free. This extension is useful whenever the same operator has to be solved for many rhs vectors, e.g. in the case of MS-FEM.

AMG Ist working as long as the coarse grid solver is an iterative solver. I therefore added an additional preprocessor macro to disable the direct coarse solver explicitly.

What is missing currently is multi-rhs support for the direct solvers. I'm preparing a separate merge request for this feature, which will add multi-rhs direct-coarse-grid support for the AMG.

See also: core/dune-common!81, core/dune-geometry!13.


See merge request !17