rs6000 build breakage

[David Edelsohn]

	Something started to break the rs6000 build beginning Tuesday,
August 25.  Originally it caused compiler internal errors and crashes, but
then was "fixed" to silently produce wrong code.

	Significant changes affecting the common backend around this time
include Mark Mitchell's loop.c varrays change and CC_MODE tweak as well as
Richard Henderson's haifa-sched.c changes.  Both of these patches had
additional fixes applied during the time I observe the failure mode
changing from internal errors to crashes to silently wrong code.

	These bugs may have been completely fixed in later updates and the
current problem cause by a different patch, but I cannot find a period
when I can perform a successful bootstrap beginning 25 Aug 1998.  That's
about as far as I have been able to track this problem with repeated
bootstrap trials for different CVS timestamps.

David