This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Polyhedron performance regression
- From: dominiq at lps dot ens dot fr (Dominique Dhumieres)
- To: burnus at net-b dot de, dominiq at lps dot ens dot fr
- Cc: fortran at gcc dot gnu dot org
- Date: Mon, 20 Nov 2006 00:18:19 +0100 (CET)
- Subject: Re: Polyhedron performance regression
> I could imagine that on a PowerPC
Yes (I think Intel requires 10.4).
My understanding of the ac test is that a significant fraction of
cpu time may be eaten by the random number generator based on mod.
I got the impression that the gain done by inlining the proc is
to avoid a non necessary repetition of a division at each call
and have a new random number with only a muladd. I have a variant
doing this kind of optimization and is about twice as fast as
the old ac version (25s compared to 40s on OSX, no gain on Linux/AMD).
Dominique