Boost logo

Ublas :

From: daniel l elliott (danelliottster_at_[hidden])
Date: 2006-12-17 06:45:57


Hello,

I am trying to create an initial version for a transformation matrix class. I
have already run into a couple difficult design decisions.

Given that his is a broad class of matrices, this class must be able to handle a
large variety of storage mechanisms. Therefore, this class should be able to
handle and (eventually) take advantage of storage techniques ranging from the
super-sparse (permutations like those used in lu.hpp) and the more dense (like
eigenvectors).

Here is my first, most important decision I am confronted with: should we follow
the hermitian matrix model and make transformation matrices on the same level as
sparse and dense matrices or can we take advantage of the work that has already
gone into optimizing the various matrix formats and make the matrix type a
template parameter or data member?

I strongly prefer the latter suggestion, but I fear that this could make
subclassing from the matrix_expression and matrix_container classes troublesome.
 Therefore, I am turning to those who know this code, design, and future of
ublas many times better than I. I would not be surprised if the crack group of
C++ gurus who work on ublas will be able to contribute more options.

I appreciate any direction I might receive.

- dan elliott