-------- Forwarded Message --------
Subject: Re: [OPM] Making OPM modular: support for more linear solvers
Date: Tue, 28 Oct 2014 13:31:23 +0100
From: Jørgen Kvalsvik <[email protected]>
To: Arne Morten Kvarving <[email protected]>
On 10/28/2014 01:24 PM, Arne Morten Kvarving wrote:
On 28/10/14 13:19, Atgeirr Rasmussen wrote:
hi,
my .02c;
However, I am afraid that it
may end up as "Yet Another Sparse Matrix" and only add to any
complexity or confusion there already is. Abandoning the Eigen
SparseMatrix for opm-autodiff will require a substantial effort, that
I do not think we'll make (at this point at least), so I think that
will remain.
this. no matter which is chosen, conversions will be the consequence and
these are nontrivial for large systems. having the flexibility to use
the native type of whatever underlying solver library used should be the
goal. that is, a thin wrapper on top of native matrix types. istl got
this right with its LinearOperators (sadly, but obviously, only as long
as no algebraic preconditioners are used).
arnem
The templated solution I'm currently writing can easily support this.
Some details and dependencies will have to be worked out in the case of
separated builder and representation, and it might even hurt run-time
selection of solvers, but I'll keep it in mind and see what I can figure
out.
_______________________________________________
Opm mailing list
[email protected]
http://www.opm-project.org/mailman/listinfo/opm
