Re: [boost] ublas & lapack

From: Matthias Troyer (troyer_at_[hidden])
Date: 2002-01-31 02:17:38

On Wednesday, January 30, 2002, at 11:58 PM, Peter Schmitteckert (boost) 
wrote:

> Dear all,
>
> I would volunteer to write an interface for ublas to at least some
> Lapack functions. Before implementing them I'd like to have a few
> comments on my appended example, especially:
>
> -> should I use boost::scoped_array for the workspace pointers?
> -> should I avoid dynamic memory allocation within the routines? i.e.
>    one has to provide the workspace as arguments.
> -> naming: should I try to match Lapack names ? e.g. I'd prefer
>    to implement the various diagonalization routines by overloading
>    a 'diagonalize(···)' function instead of implementing
>    'hpevd', 'hevd', ... . Same applies to the various SVD routines.
> -> Am I correct that this makes only sense for the dense
>    numerics::unbounded_array memory shemes.
> -> What's the "canonical" ublas way to specify the start of the memory 
> block ?
>    Is &E[0] fine ?
> -> How do I get the leading dimension for  matrix_ranges ?

I actually like the current MTL interface and style (in MTL-2).
Are there any plans to merge MTL-3 and ublas?

Matthias