On Tue, Jun 12, 2007 at 09:41:58PM -0500, Callum Lerwick wrote: > On Wed, 2007-06-13 at 04:28 +1000, Mike Kearey wrote: > > I'd like to see repeatable, measurable tests not subjective 'I think > > it's faster' observations. I am no criticizing here, it's just that > > humans are actually bad at this sort of thing. Measurements are better. > > I've been doing rather extensive performance profiling on OpenJPEG, > using Fedora 6/7's gcc 4.1, and I've discovered a few things: > > Compiling for pentium3 rather than "generic" measurably improves > performance on both my i386 test platforms, mobile Celeron 1.3 (PIII > based), Celeron 2.1 (P4 based). This is at least partly due to > optimizing signed integer math with cmovs. No surprise really. generic isn't targetting either of the processors you mention. The idea behind generic is 'optimise for _todays_ CPUs', which right now typically means Intel Core, and AMD Opteron/Athlon64. On these platforms, cmov isn't a win (and is even a loss in some cases). For stuff that's really performance sensitive, runtime selection of optimised routines is a far better solution than n different packages for each flavour of CPU. Dave -- http://www.codemonkey.org.uk -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
