Ok, I've added an --enable-mmx parameter. I don't think it's a good
idea to compile for i386, I think it would be better to optimize for
the pentium-mmx, and let those with pre-mmx processors compile their
own. I had a dual pentium-pro machine once, top of the line, it ran
at a 200Mhz. There is no way it could run a frontend, and as far as
a backend goes the slots were probably PCI v1.0... Also even if you
make a pentium-mmx version, you should at least make a pentium2
version, the cmov instruction alone results in a 10% speedup for
ffmpeg.

Yes, I'm aware that it's not a great idea, but redhat has it "on good authority" that i386 tuned for ppro or p4 is actually faster in most cases than i586 turned for ppro or p4. can't move to i686 because of missing registers (or something like that, I'm not a cpu guy) in the via procs (which is apparently the only difference between 586 and 686 to gcc).


I do plan to make a couple of different versions of the rpm (even though the fedora guys don't really want to deal with it), but want to make sure that the handful that I make will work for as many people as possible (and then I'll try to make it easy for people to build their own from the src.rpm and use the new auto-detection stuff for best optimization) Unfortunately, right now, I can't seem to get things to accept the make parameters passed in by rpmbuild, so I have to recreate all of the arch stuff by hand. (Axel's rpm's manually add them to QMAKE_CXXFLAGS_RELEASE and QMAKE_CFLAGS_RELEASE, after doing a "-= pentiumpro" but that doesn't seem to do anything for me, even if I hard-code pentiumpro into the configure phase)

Incidentally, is pentium2 better/worse than pentiumpro?

And is there any particular reason why this sort of detection/optimization isn't done in runtime like it is with programs like mplayer?

It looks like the libmythmpeg2 library that was added a few
weeks ago did not have "#ifdef MMX" protection around a couple
MMX parts. It should be ok now.

Thanks.

-Chris
_______________________________________________
mythtv-dev mailing list
[email protected]
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-dev

Reply via email to