Re: Bug 89387 - improve performance for some matrix operations

On Wed, 2016-12-14 at 13:04 +0100, Eike Rathke wrote:
> Hi Filippo,
>
> On Wednesday, 2016-12-14 09:59:28 +0000, Filippo giacchè wrote:
>
> > -why there is no need to modify that function?
> Because it is called only in one place and that place in turn is
> called
> very rarely. There's almost no gain over all if only MatCopy() alone
> was
> optimized, despite that it proabbly can't, unless the underlying
> matrix
> structures/algorithms are touched from which all matrix functions
> could
> benefit, which would involve getting acquainted with
> mdds::multi_type_matrix that is about the hardest stuff you can get
> into.

On that front, I might just add that I've made the following change to
mdds::multi_type_matrix::copy (which MatCopy() uses internally)

https://gitlab.com/mdds/mdds/commit/6bd53e22dc21b6b88511d7e03e349021c37
56c47

which *should* automatically improve the performance of MatCopy() quite
a bit, depending on the use case.

The version of mdds we use on the master branch (and possibly some of
the older stable branches) already has this change in.

Kohei