GCC trunk SPEC2000 performance

[David Edelsohn]

	The 6% performance improvement spike of SPEC CPU2000 172.mgrid in
January appears to correspond to Daniel Berlin's commit and reversion of
his LCM changes to use iterative dataflow framework.  The patch diminished
the performance of 179.art and 301.apsi slightly.  The average performance
stayed about the same.

	Is it possible to resurrect that patch so that it is more
selective? 

Thanks, David