Polyhedron benchmark on Opteron

["François-Xavier Coudert" <fxcoudert at gmail dot com>]

Hi all,

[spoiler: to all people wanting to lend a beginner's hand to gfortran
development, there is something for you to do at paragraph 4 of this
mail]

I wanted to report some results for the Polyhedron benchmark** on
Opteron (hardware details at the bottom of this mail). I used gfortran
mainline (4.2.0 on 2006-09-28) and Intel 9.1.037 for comparison.
Options used are :
 * gfortran -march=k8 -ffast-math -funroll-loops -static -O3
 * ifort -O3 -xW -ipo -static -V

The geometric mean execution times are 31.8 (gfortran) and 28.3
(Intel), which is a 12% difference (down from 21% in May 2006).

For most individual mail, the execution times are within 10% of each
other, with gfortran winning for 7 tests (total number of tests: 16).
Unfortunately, there are also tests for which Intel is a clear winner:
 -- aermod, by 44%
 -- air, by 30%
 -- fatigue, by 22%
 -- gas_dyn, by 115%
 -- nf, by 29%

So, there's still some progress possible here. Some people asked about
beginner projects for helping us making gfortran a better compiler: I
think profiling these cases, with gfortran and some other compilers to
see where we end up spending our time, and filing PRs so that we can
keep track of things that need to be done.

Thanks,
FX



Hardware details: I used one processor from a biprocessor Dual Core
AMD Opteron(tm) Processor 265, 1.8 GHz, with 2 Go RAM, running
Scientific Linux SL release 4.3 (Beryllium). The results reported here
are for a Standard run. Complete results can be found in the
attachements of this mail.

** Polyhedron 2005 Fortran Benchmarks, http://www.polyhedron.co.uk/

Attachment: gfortran_bench.txt
Description: Text document

Attachment: intel_bench.txt
Description: Text document