unflattening fortran matrices

[Jack Howarth]

Steve,
    I thought at this point the gfortran compiler was pretty much
feature complete for f95. In any case, it still merits trying
to find out if gfortran is abnormal in using flat representations
of matrices compared to other fortran compilers that support
aggressive vectorization. Specifically, I think we should find
some numerical benchmarks that exist as both c and fortran versions
with similar levels of source code optimization. This way we could
get a sense of how badly we are being hurt compared to c in not
supporting as much vectorization. Remember that Dorit will be
switching vectorization on at -O3 very shortly. 
              Jack

On Mon, Sep 17, 2007 at 02:04:36PM -0700, Steve Kargl wrote:
> On Mon, Sep 17, 2007 at 04:40:37PM -0400, Jack Howarth wrote:
> >    It appears that gfortran is going to be poorly vectorized
> > due to the use of a flat representation of multidimensional
> > matrices. Has any thought been given to creating a branch to
> > begin work on conversion to a non-flat matrix representation?
> > Also how common is it among fortran compilers to use a
> > flat representation for matrices?
> 
> I think that it is more important to produce a working
> Fortran 95 compiler before one rips the scalarizer out
> of the compiler.
> 
> Of course, you're more than welcome to supply said patch.
> 
> -- 
> Steve