my bct loop hacks

[Michael Hayes]

David Edelsohn writes:
 > 	I am testing out the re-integration of the BCT code with the
 > current version of EGCS in CVS.  With my version of loop.c prior to the
 > check_dbra_loop() changes in August, loop_iterations() discovered the
 > constant number of iterations can be discovered anymore.  BCT insertion
 > then fails.

Ah, I've been having problems with this as well.  It looks like a
problem with recent changes to loop.c:get_condition not allowing a
VOIDmode comparison operator following a CCmode COMPARE operation.

(set (reg:CC ccreg) (compare:CC (...)))
(set (pc) (if_then_else (lt (reg:CC ccreg) ...)