Re: GSL 2.0 roadmap (one man's view)

[Gerard Jungman <jungman@lanl.gov>]

On Mon, 2009-09-07 at 16:09 +0100, Brian Gough wrote:
> 
> As I mentioned in previous emails, there isn't a feasible general
> solution for LAPACK with the column-major matrix restriction, due to
> the combinatorial explosion of cases that must be handled.


I don't understand "combinatorial explosion". LAPACK only handles
column-major, so it can only be fed column-major. If you want to
use it, all your data structures are born column-major, or you
convert them as needed.

We need to simply allow for the existence of the column-major
world. It's only fair. And it's a simple variation to encode
in any of our data structures.

Every serious C++ vector/matrix library allows for both column-major
and row-major layout, because they know people need it. Templates, or
other gymnastics, are not required. You just have to decide that you
want it.


> We could adopt FLAME, which is a much more general framework, C-based
> and faster than LAPACK.  I really think this is a better way to go
> than LAPACK.  Unfortunately it doesn't have so many routines at the
> moment.

For me, the string "LAPACK" stands for any existing code base that
people will want/need to use. FLAME is just another "LAPACK" for these
purposes. There is nothing for us to "adopt". It's the end user who
"adopts" a package to use. We just have to make it easy for them to
adopt whatever they like.

And if FLAME, or any other library, uses data structures which make it
hard for people to talk to or translate to, then they have made a
mistake.


On Mon, 2009-09-07 at 14:21 -0400, Robert G. Brown wrote:
> I've never been that fond of LAPACK anyway -- it's not like it is easy
> to use and perfectly intuitive anyway.

I know. But it works. And people want it. It's crazy to ignore it.

--
G. Jungman