This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
graphite compile time improvements
- From: Jack Howarth <howarth at bromo dot med dot uc dot edu>
- To: Sebastian Pop <spop at gcc dot gnu dot org>
- Cc: gcc at gcc dot gnu dot org
- Date: Fri, 28 Aug 2009 22:39:47 -0400
- Subject: graphite compile time improvements
- References: <20090828173902.GA14640@bromo.med.uc.edu> <cb9d34b20908281125q4f83ae85r73fd74327fad88c@mail.gmail.com> <cb9d34b20908281345l39c0ceafm9f126bb11ac013ca@mail.gmail.com>
On Fri, Aug 28, 2009 at 03:45:38PM -0500, Sebastian Pop wrote:
> On Fri, Aug 28, 2009 at 13:25, Sebastian Pop<spop@gcc.gnu.org> wrote:
> > I am doing this right now. ?I will first merge trunk into graphite
> > branch and then commit all the changes from graphite to trunk (modulo
> > some changes that should remain in the graphite branch, like enabling
> > -floop-* in -O2 and some debug utils).
> >
>
> Done.
>
> > Sebastian
> >
FYI, here is the compile time improvements that I am seeing under x86_64-apple-darwin10
for -ffast-math -funroll-loops -msse3 -O3 -fgraphite-identity -floop-interchange -floop-strip-mine
in the Polyhedron 2005 benchmarks...
Benchmark Compile Time (sec)
gcc trunk 8/15/09 gcc trunk 8/15/09 gcc trunk 8/28/09
(no graphite) (graphite) (graphite)
ac 1.60 6.07 3.44
aermod 57.99 88.54 59.48
air 4.16 7.79 5.72
capacita 2.67 3.20 2.83
channel 0.99 2.54 1.95
doduc 8.18 8.92 8.63
fatigue 3.08 5.42 3.25
gas_dyn 4.97 5.65 5.31
induct 6.74 26.53 10.81
linpk 0.99 15.77 1.11
mdbx 2.39 2.83 2.66
nf 2.65 4.27 2.80
protein 7.85 8.24 8.93
rnflow 8.30 11.29 9.36
test_fpu 6.80 9.60 7.05
tfft 0.73 2.31 0.81
Nice.
Jack