I recently upgraded to GCC3 (3.2.3), and rebuilt libvorbis in the process, going from ebuild 1.0-r2 to 1.0-r3. The latter does not filter out "-march=*" flags, and this caused a problem with rock music, which has some jarring higher frequency sounds like cymbal crashes and electric guitar harmonics. So I changed my CFLAGS from "-march=pentium4 -Os -pipe" to "-mcpu=i686 -Os -pipe" and now my OGG tracks sound nominal. My recommendation is to retain the "-march=*" filter in the libvorbis ebuilds. Reproducible: Always Steps to Reproduce: 1. (see "Details")
Whoever does libvorbis please.
Can you retest with -r4. If it is still an issue we can filter pentium4 out.
There is some improvement, but still not acceptable. Going from "-march=pentium4" to "-march=pentium3" doesn't help; "-mcpu=*" is works fine, as before. This Forums thread suggests that "-march=*" is untenable for AMD architectures as well: http://forums.gentoo.org/viewtopic.php?t=73292&highlight=libvorbis
Understood. I'll add something to filter march in r4. Thanks.
Now filtering -march=pentium?. Let me know if this works. I'm not convinced that the athlon-zp problem is the same. Reading into the forum post i'd guess that -sse +3dnow results in the simd patch not being applied and consequently everything being okay. If this is the case then perhaps the patch could be applied if (sse || 3dnow), but i have no idea if that will result in any improvement on althon-xp. Anyway, that bug will wait for another day; i don't have an athlon box to experiment with.
Where are you filtering? I'm not seeing any changes after "emerge rsync" nor through the CVS browser on Gentoo.org. Anyway, I'm hitting the sack soon -- let me know when you commit the changes, and I'll try out the ebuild to make sure it works.
Revised -r4 has propagated, and works fine.