Re: [tree-ssa]: Implementing unimodular and non-singular loop transforms

[Daniel Berlin <dberlin at dberlin dot org>]

On Jan 6, 2004, at 2:08 PM, Richard Henderson wrote:

> On Tue, Jan 06, 2004 at 12:54:19PM -0500, Daniel Berlin wrote:
> > I didn't mention option #3, which is
> >
> > Write our own matrix math package, and rewrite the lambda algorithms
> > because i really have a hard time believing there is no matrix math
> > package we can use.
>
> Not sure what kind of matrix math you want, but surely the fallback
> would be BLAS.

Very simple stuff, of course:
Basic matrix and vector and lattice operations (add in various forms, 
multiply in various forms, inversion, etc)
A routine to compute hermite form

Nothing highly complex.
But still a bit tedious to code, and if we can get it as part of 
something other parts of gcc will use more of, so much the better.


> I expect some of that would start to get used by
> gfortran anyway, for various high-level constructs.
>
>
> r~