dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Martin Sandve Alnæs wrote:
> Would it be ok now to add arithmetic operations in the Generic*
> interfaces? They can be limited to objects of the same type, using a
> type assertion like init already must do for the sparsity pattern. A
> fairly small set of operations will allow us to write many linear
> algebra algorithms in PyDolfin for any backend, great for
> application-specialized algorithms and rapid prototyping.
>
> Vectors:
> v+u
> a*v
>
> Matrices:
> A+B
> a*A
>
> Matrix-vector:
> A*v

I can't comment on integration pyDOLFIN, but we went to uBlas to get 
this type of functionality for free. The problem we had with adding a 
lot of operators to Generic* is that the code for each back-end really 
grows, especially if the back-end is not well suited to the operation 
(this was often the case for PETSc).

Garth


> If there are any good reasons to avoid this (f.ex. virtual operators
> in C++ doesn't exist), I suggest either defining separate interfaces
> for arithmetic operations (f.ex. a GenericOperator could be something
> with a matrix-vector product) or at the very least defining
> conventions for these operations, which would allow PyDolfin apps to
> use this by "duck-typing" (since Python is dynamically typed).
>
> --
> Martin
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev

--
Dr Garth N Wells
Department of Engineering
University of Cambridge
Trumpington Street
Cambridge CB2 1PZ
United Kingdom

tel.   +44 1223 3 32743
fax.   +44 1223 3 32662
e-mail gnw20@xxxxxxxxx