unflattening fortran matrices

[Jack Howarth]

Janne,
    It certainly looks like most of the interesting optimizations
in gcc 4.3 will be unavailable to us due to the flat matrices.
Specifically I see that is supposed to be the case for
predictive commoning feature. I guess we will have to wait for
the replacement for the haifa scheduler in gcc 4.4 before we
can expect major gains in gfortran performance.
              Jack

On Tue, Sep 18, 2007 at 04:25:53PM +0300, Janne Blomqvist wrote:
> Not only optimization issues, but debug information for arrays as
> well. If we would pass multidimensional arrays to the middle-end as
> is, debug info would be relatively simple.
>