This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Inline DOT_PRODUCT revisited
- From: Steven Bosscher <stevenb dot gcc at gmail dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: Paul Thomas <paulthomas2 at wanadoo dot fr>, "'fortran at gcc dot gnu dot org'" <fortran at gcc dot gnu dot org>, Dominique Dhumieres <dominiq at lps dot ens dot fr>
- Date: Sun, 26 Feb 2006 01:39:50 +0100
- Subject: Re: Inline DOT_PRODUCT revisited
- Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:to:subject:date:user-agent:cc:references:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:message-id:from; b=T5EKx50zP/gNsjcoT3+IsaCQ4Qr85ThOTfvUZ4i0icqwEX8oAD7xtZKavOSn4rbBYL8RcJwdww0T/9zF3YL3IQJKvMMtsYl2+ukfFIMR/81qTSa6PTkiviCaCRyl/s/juET4aMCgff53RYB5uBaDg3INL7SNlgeyuG1HPOfuOiE=
- References: <43730120.7030808@wanadoo.fr> <200602252337.39067.steven@gcc.gnu.org> <200602252347.57483.steven@gcc.gnu.org>
On Saturday 25 February 2006 23:47, Steven Bosscher wrote:
> On Saturday 25 February 2006 23:37, Steven Bosscher wrote:
> > I would _very_ much like to have this patch in GCC 4.1.1
>
> ...the motivation being, of course, that the almost industry-standard
> Fortran benchmark (Polyhedron) benefits from this patch tremendeously,
...like so:
Before patch:
induct Run # 1 281.69211 281.69211 - Error=100.0000%
After patch:
induct Run # 1 45.12120 45.12120 - Error=100.0000%
I had hoped that the reduction support in the vectorizer would help
improve things further, but unfortunately with vectorization enabled
and the patch applied, we are slower (at least, on my AMD64 box) than
with vectorization disabled:
induct Run # 1 54.64700 54.64700 - Error=100.0000%
Gr.
Steven