I have a prescott (model 3, flag pni). Between gcc benchmarks and glibc build/tests I'm finding '-march=i686 -mtune=prescott' works best together.
My gcc-3.4.3 benchmarks like march=pentium3 best, but that makes one glibc test fail. march=pentium4 also makes that test fail. Performance of march=pentium3 and march=pentium4 are almost identical, but maybe my gcc benchmarks aren't trying to take advantage of the new cpu flags. Using march=prescott produces slower programs than march=i386, for whatever reason. Google doesn't agree with my findings, but I think a lot of people are shooting in the dark and assuming march=prescott is ideal. I'm also using -O3 and --enable-omitfp with good results btw. robert On March 1, 2005 03:19 pm, Robert Connolly wrote: > Hi. I have a pentium-4 3ghz 1mb cache. Should I be expecting better results > with march=pentium3 with gcc-3.4.3? because for some reason it's performing > better than march=pentium4. Google has some references to march=pentium4 > being not good with gcc-3.2, but no word on gcc-3.4. > > robert -- http://linuxfromscratch.org/mailman/listinfo/lfs-chat FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
