On Sun, Mar 07, 2010 at 08:17:54PM +0100, Toon Moene wrote: > Jack Howarth wrote: > >> Toon, >> I would suspect this boost is from the series of changes >> to "Update default arch for x86" from HJ Lu. This should now >> have -msse2 in use as the default (except on darwin where >> we went to -msse3 since all of our processors support that). > > That could well be. I now configure as follows: > > --with-arch-64=native --with-tune-64=native > > as opposed to: > > --with-arch-64=core2 --with-tune-64=core2 > > previously. > > -- > Toon Moene - e-mail: t...@moene.org - phone: +31 346 214290 > Saturnushof 14, 3738 XG Maartensdijk, The Netherlands > At home: http://moene.org/~toon/ > Progress of GNU Fortran: http://gcc.gnu.org/gcc-4.5/changes.html
Toon, My benchmarks for the polyhedron 2005 benchmarks on x86_64-apple-darwin10 confirmed HJ Lu's observations on darwin that the core2 tuning actually perform worse than generic tuning... http://gcc.gnu.org/ml/gcc-patches/2010-02/msg01272.html Hopefully FSF gcc will eventually fix this so that processor specific costs actually improve the benchmarks. Jack
