search

nutofem / nuto

NuTo - yet another finite element library
https://nuto.readthedocs.io
Boost Software License 1.0
17 stars 5 forks source link

Simplify BlockVector/matrix #2 #57

Open TTitscher opened 7 years ago

TTitscher commented 7 years ago

The Block types implement arithmetic operators without any state of the art expression template lazy evaluation. If we derive the BlockVector directly from Eigen::VectorXd, we only need to provide operators to access a certain .segment for a specific dof type. This may involve some index magic but sounds feasible.

Maybe we can even benefit from the lazy evaluations provided by Eigen::SparseMatrix. But the block operations are limited. Read access is no problem though.

joergfunger commented 7 years ago

I'm not sure if these lazy evaluation optimization really improves the speed, due to our usage of the vectors/matrix. However, I agree that writing our own interface for matrix/vector multiplications is sub optimal. However, I don't see a nice solution in particular for the sparse matrices. Even when we switch to Eigen::SparseMatrix, I would guess that accessing this via block routines is suboptimal (after having checked the documentation, I'm not sure if there are block methods available for sparse matrices anyway)