[PPL-devel] Strange thing in Polyhedron.cc

[Elisa Ricci]

Enea Zaffanella wrote:

> > Hi there,
> >
> > this is the contents of lines 1411-1414 of Polyhedron.cc.
> >
> > #if 0
> >   //#ifdef BE_LAZY
> >   // FIXME: this has to be checked carefully
> >   // for correctness and/or efficiency.
> >
> > Can someone explain what is happening here?
>
> It _was_ happening that I (maybe together with Elisa) was looking
> for places where we could have been lazier than we are now.
> Unfortunately, then I got busy with other stuff, so that the
> "#ifdef BE_LAZY" is still in its place, commented out and wrong.
>
> In that particular case, I noted that we were using the
> procedure Matrix::merge_rows_assign
> which takes as argument a reference to a _const_ Matrix.
> Therefore, this procedure _copies_ rows from its argument.
>
> However, when we call this procedure in add_constraints,
> our actual parameter is not required to be const, so that
> we could have avoided this copy and swap constraints instead.
> I then realized that swap of constraints could not be done
> if the space dimensions were different.
> Still, instead of copying, we could have swapped the coefficient
> of the constraints, therefore achieving some improvement when
> dealing with big numbers.
> Now, don't ask me why in the commented code we were still
> using assignment instead of swaps ... as I said, I had to stop
> working on it, commented it out and then plainly forgot
> about this improvement.
>
> Maybe Elisa remembers something more ...
>
> Enea.
>
> _______________________________________________
> PPL-devel mailing list
> PPL-devel at cs.unipr.it
> http://www.cs.unipr.it/mailman/listinfo/ppl-devel

I only remember that Enea added this part of code to prove adding rows
without using
Matrix::merge_rows_assign. This was done before the function
Matrix::merge_... resizes the system that
we want to add.

Ciao,
    Elisa