This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Netlib's lapack
On Tue, Apr 13, 2004 at 09:43:57PM +0800, Feng Wang wrote:
> --- Steve Kargl <sgk@troutmask.apl.washington.edu> wrote??> Gang,
> > The bad news is the test programs
> > for the COMPLEX and DOUBLE COMPLEX types are showing many
> > failures. The LAPACK test programs also show a similar
> > pattern with the exception that a few REAL and DOUBLE
> > PRECISION programs are failing.
>
> Yes. gfortran still has a long way to go. In fact, I worry about
> its performance a little now. Many optimizations implemented in
> gcc backend are not used by gfortran, or there are still some
> "gaps" between the frontend and backend. For example, some loops
> orgnized by gfortran cannot be identified by backend and the alias
> analyzing also has problem.
Hi Feng,
In some informal benchmarks, gfortran has performed as well as g77
and NAG's Fortran compiler. These benchmarks use a reduced set of
Fortran 77, but it shows decent performance. I would prefer to
make gfortran first work for any missing or broken pieces, then
we can worry about performance.
Can you give us some idea of the areas that you and your CCRG
colleagues are currently investigating?
--
Steve