It is a memory reinterpretation

... instead of reimplementing it all over the place

It otherwise is a pain to debug problems arising in cgen.

Merge branch 'feature/sumfactorization' of ssh://parcomp-git.iwr.uni-heidelberg.de:20022/dominic/dune-perftool into feature/sumfactorization

Conflicts:
	python/dune/perftool/sumfact/amatrix.py

Conflicts:
	python/dune/perftool/pdelab/quadrature.py
	python/dune/perftool/sumfact/amatrix.py

Get endless temporaries who circularly switch around their base storage.
Any temporary may have different shape or dim_tags, reinterpreting the
data at will.

Feature/eliminate dimension variable

Fixes #42

See merge request !46

Feature/update loopy

Some more upstream changes.

Test suite passes.

See merge request !44

Feature/fix stokes stress

This fixes #44.

See merge request !43

Implement test function splitting

As a step towards sum factorization, this implements a splitting of test function modified arguments from the rest of the form.

While at it, I introduced a data structure `ModifiedArgument`, which uses `pytools.Record` and is much better for debugging.

See merge request !41

Use UFL for guessing quadrature order

1. ~~In order to test this branch you need to apply the patch in patch/ufl/ufl.patch to ufl. At the moment estimating degree in UFL doesn't work if there is a FunctionView followed by a ReferenceValue. This of cousre happens in stokes stress numdiff (where else?). That leads directly to the question: Should we keep that patch and introduce new version madness or should we do something else. This is the reason for WIP.~~  
 With the estimated polynomial degree from the metadata method the patch is not necessary anymore.

2. ~~The other interesting part is the behaviour of estimate_total_polynomial_degree from UFL. If you estimate the polynomial degree of Laplace without going to the reference elements for linear basis you get 0 (makes sense). If you go to the reference element you get 1. For poisson going to the reference values also adds 1 to the estimation. It is not obvious what you want to have but this seems a bit strange. Maybe we should print (or make it possible to print) the estimated degree?~~  
 This also simply gives the quadrature order you would expect from the weak formulation.

3. We have to look at our expressions at some point. Right now we give them order 0 if cellwise constant or order 1 if not. This kind of depends on the whole how to describe everything in the DSL. So it probably makes sense to postpone ;)...

This closes issue #38.

See merge request !40