[Dune-devel] [dune-functions] hierarchicVectorWrapper and hierarchicMatrixWrapper

Simon Praetorius simon.praetorius at tu-dresden.de
Wed Feb 21 13:17:23 CET 2018


Hi folks,

during my development of a discretization module based directly on 
dune-functions, I've implemented wrappers for hierarchically nested 
vectors and matrices. The former was already available, but I don't see 
how it can be extended from vectors to matrices. Thus, I've made a new 
approach and came up with the idea of coupling the hierarchically nested 
vector/matrix with the corresponding nesting structure. When declaring 
the vector/matrix type this must already be known and is naturally 
related to the index-merging strategy of the GlobalBasis.

This coupling of data-structure and nesting-structure works quite fine 
and (nearly) all the former tests for hierarchicVectorWrapper can be 
passed. Additionally, some more data-structures are supported, e.g. for 
a taylor-hood basis

- TupleVector<std::array<std::vector<bool>,dim>, std::vector<bool>>
- TupleVector<BitFieldVector<dim>, std::vector<bool>>

usable as a bit-field in interpolation methods, for example.

With "nesting-structure" I mean a light-weight representation of the 
index-merging strategies, e.g. for the examples above:

- Blocked<Blocked<Flat>, Flat>
- Blocked<LeafBlocked<dim>, Flat>

where `Blocked` is related to `BlockedLexicographic`, `Flat` is related 
to (`FlatLexicographic`, `FlatInterleaved`) and `LeafBlocked<dim>` is 
related to `LeafBlockedInterleaved`.

This development is now ported back from our discretization module to 
dune-functions and is uploaded as an experimental branch, see MR !90. 
Currently, a lot of utility code is necessary. But maybe this can be 
eliminated or replaced with other code.

What do you think? Is this a way to go or should we look for a another 
way to write wrappers purely on the data-structure without knowledge of 
the nesting-structure?

Best wishes,
Simon




More information about the Dune-devel mailing list