3.3 vs 3.1 on powerpc

[Brad Lucier]

I've been doing some tests of 3.3 versus (Apple's) 3.1 on PowerPC Darwin.

With -fnew-ra and the new scheduler, I'm expecting 3.3 to kick serious butt
compared to 3.1 on some of my floating-point codes---studying the scheduler
output indicates a 50%+ improvement when scheduled for -mcpu=7400.  (For
some reason, the scheduler has a more difficult time with -mcpu=7450.)
Some of this speedup results from fewer spills.

Unfortunately, I can't figure out how to get mainline 3.3 to accept some
of Apple's headers, so I can't actually run anything yet ... ;-).  

Brad