This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: benchmarking.



On 15 Jun 2008, at 15:24, Tobias Burnus wrote:


Tobias Schlüter wrote:
IainS wrote:
Is there a gfortran preferred or standard performance benchmark suite?
gfortran is regularly tested with the Polyhedron benchmark suite <http://polyhedron.com/polyhedron_benchmark_suite0html> and with the SPEC fp benchmark suites <http://www.spec.org/>.
See also: http://gcc.gnu.org/benchmarks/

There you also find PowerPC 64 SPEC 2000 runs (third link).


Regarding: The polyhedron test suite, http://www.polyhedron.co.uk/MFL6VW74649

You can simply create a .par file yourself; it simply contains the command incuding a "%n.f90 -o %n" for the file name.

I currently use (see http://www.physik.fu-berlin.de/~tburnus/gcc- trunk/benchmark/):

gfortran -march=opteron -ffast-math -funroll-loops -ftree-loop- linear -ftree-vectorize -msse3 -O3 %n.f90 -o %n

You can then run it using:

../pbharness  gfortran.par  quick.par
../pbvalid gfortran.run  ../../f90valid.in

I'll send you a modified version of the pb*f90 files, which I was able to compile with gfortran on MacOS (the default version requires some extensions gfortran does not have).

Thanks very much, hopefully I'll try these overnight tonight.


I assume that on a multi-core or multi-processor machine it's conventional to force all but one core/processor off?

Iain


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]