This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Polyhedron Benchmark
- From: Tobias Burnus <burnus at net-b dot de>
- To: "'fortran at gcc dot gnu dot org'" <fortran at gcc dot gnu dot org>
- Date: Mon, 16 Jul 2007 19:10:48 +0200
- Subject: Polyhedron Benchmark
Hello,
there is a new Polyhedron benchmark available; this time only on Pentium
D820 dual core and not on AMD. The gfortran compiler is still "4.2.0"
(what ever version that is).
See:
http://www.polyhedron.com/
http://www.polyhedron.com/pb05/linux/f90bench_p4.html
* * *
As a reminder, I run a comparison on AMD Athlon64 4800+ using all kind
of gfortran options, but fewer commercial compilers. The result is at:
http://physik.fu-berlin.de/~tburnus/gcc-trunk/benchmark/#rt
Nutshell: gfortran is 7% slower than ifort and 16% slower than Pathscale.
SUSE has also daily runs (more reliable gfortran runs, fewer compilers):
http://www.suse.de/~gcctest/c++bench/polyhedron/
* * *
I did a quick comparison on an Intel Core2 Duo T7300 @2.00GHz using
(-m64) using:
gfortran -march=native -ffast-math -funroll-loops -ftree-vectorize
-ftree-loop-linear -O3
and with ifort 10 using:
ifort -O3 -xT -ipo
Result:
benchmark gfortran ifort
------------ ----- -----
ac 12.95 17.36 ++
aermod 36.32 25.30 --
air 9.41 8.37 -
capacita 51.85 0.09 (ifort crashed)
channel 2.87 3.56 ++
doduc 47.41 47.69 0
fatigue 12.86 10.96 -
gas_dyn 8.57 3.97 ----
induct 92.82 63.08 ---
linpk 22.67 22.47 0
mdbx 19.03 17.66 -
nf 26.10 0.23 (ifort crashed)
protein 60.30 55.11 -
rnflow 35.11 42.29 ++
test_fpu 12.68 10.78 --
tfft 2.97 2.88 0
------------ ----- -----
Geometr.mean 18.99 N/A
gfortran (as usual; with -ffast-math):
linpk FAILED 1 fails and 4 passes
> Value= 25.114499300 Target= 23.100000000 Tolerance= 2.0000000000
ifort 10:
I don't see ad hoc why ifort 10 has problems with "capacita" and "nf",
but the a.out crashes even with -O0 on this PC;
maybe it does not like my glibc 2.6 or my gcc 4.2 system compiler?
Tobias