[Gsll-devel] announcement: affi (affine indexing)

[Tamas K Papp]

Hi Liam,

On Tue, May 20, 2008 at 06:07:14PM -0400, Liam Healy wrote:

> Very interesting.  Does this map onto the GSL notion of views in any way,
> even conceptually?  I know we discussed this before, I'm just wondering
> if a restricted set of affine indices could be used with GSL.

Yes, I think it does allow the handling of GSL's views, including

- matrix views (GSL has i*tda+j, which is an affine mapping, so we can use
submatrix views, etc)

- vector views on a matrix (row, column, or contiguous subviews of
those)

But affi allows much more, including views of multidimensional arrays,
or implementing a transpose or index permutation with "views".  You
can even do an affi that traces out the diagonal of a matrix...  The
possibilities are endless.

I have to admit that I don't have a very good understanding on how and
where GSL uses views, so at this point I can't offer my thoughts on
how to best integrate this with the semantincs GSLL if you want to,
especially if you want to do it in a Lispy manner and not merely
mirroring GSL's views.  But I am happy to discuss things, or add
anything you need to affi.

Best,

Tamas