[PATCH] Fix PR middle-end/6100

[Kaveh R. Ghazi]

 > From: Jakub Jelinek <jakub at redhat dot com> 
 > 
 > Hi!
 > 
 > We were failing to compile gcc.c-torture/compile/20020320-1.c -O0 with rtl
 > checking, since REG_BR_PRED's argument was CONCAT not CONST_INT.
 > David, do you remember why you used REG_BR_PRED and not REG_BR_PROB like
 > all other targets do?
 > Jan, current REG_BR_PRED handling is inconsistant at least. It is documented
 > to be CONCAT (in rtl.h), predict* use CONCAT but reorg.c creates REG_BR_PRED
 > notes with CONST_INT and quite different meaning (this is what sparc backend
 > used).
 > 
 > 2002-03-30  Jakub Jelinek  <jakub@redhat.com>
 > 
 > 	PR middle-end/6100
 > 	* config/sparc/sparc.c (output_cbranch): Use REG_BR_PROB, not
 > 	REG_BR_PRED.
 > 	(output_v9branch): Likewise.

Jakub,

Thanks for addressing some of the RTL checking failures, last night's
run looks better.

BTW, is it possible that the above patch might also have fixed PR6101?
PR6101 no longer appears, and it's symptom looked really similar to
PR6100.  Would you take a quick look at the PR and confirm that your
patch was responsible so I can close that issue too?

		Thanks,
		--Kaveh
--
Kaveh R. Ghazi			Director of Systems Architecture
ghazi@caip.rutgers.edu		Qwest Global Services