[cond-optab][patch RFA] Fix bootstrap failure on SH

[Kaz Kojima]

Hi,

Bootstrap has failed for sh4-unknown-linux-gnu on cond-optab
revision 145778 at stage2 with

../../INTEST/cond-optab/gcc/tree-ssa-structalias.c: In function 'get_constraint_for_ptr_offset':
../../INTEST/cond-optab/gcc/tree-ssa-structalias.c:2938: error: could not split insn
(jump_insn 174 749 175 ../../INTEST/cond-optab/gcc/tree-ssa-structalias.c:2884 (parallel [
            (set (pc)
                (if_then_else (ne (reg:DI 5 r5)
                        (const_int -9223372036854775808 [0x8000000000000000]))
                    (label_ref 260)
                    (pc)))
            (clobber (reg:SI 7 r7))
            (clobber (reg:SI 147 t))
        ]) 6 {cbranchdi4_i} (expr_list:REG_BR_PRED (const_int 49 [0x31])
        (expr_list:REG_UNUSED (reg:SI 147 t)
            (expr_list:REG_UNUSED (reg:SI 7 r7)
                (expr_list:REG_BR_PROB (const_int 0 [0x0])
                    (nil))))))
../../INTEST/cond-optab/gcc/tree-ssa-structalias.c:2938: internal compiler error: in final_scan_insn, at final.c:2608

It seems that this is a target problem which is revealed
on that revision of cond-optab branch.  sh.md:cbranchdi4_i
define_insn_and_split looks like

(define_insn_and_split "cbranchdi4_i"
  [(set (pc)
	(if_then_else (match_operator 0 "comparison_operator"
			[(match_operand:DI 1 "arith_operand" "r,r")
			 (match_operand:DI 2 "arith_operand" "rN,i")])
		      (label_ref (match_operand 3 "" ""))
		      (pc)))
   (clobber (match_scratch:SI 4 "=X,&r"))
   (clobber (reg:SI T_REG))]
  "TARGET_CBRANCHDI4"
  "#"
  ...

The failed insn matches the second constraint "i" for
the second operand 0x8000000000000000, but only 8-bit
constants are permitted for non SHmedia targets with
arith_operand which is defined at sh/predicates.md.
I've confirmed that the attached patch fixes the failure
and it doesn't cause any regressions on the cross compiler
for sh4-unknown-linux-gnu.  The native bootstrap isn't
completed yet but it's now in the stage3.
OK for cond-optab branch?

Regards,
	kaz
--
2009-04-09  Kaz Kojima  <kkojima@gcc.gnu.org>

	* config/sh/sh.md (cbranchdi4_i): Use an "I08" constraint
	instead of an "i" constraint.

--- cond-optab/gcc/config/sh/sh.md~	2009-04-07 19:37:50.000000000 +0900
+++ cond-optab/gcc/config/sh/sh.md	2009-04-09 13:23:51.000000000 +0900
@@ -736,7 +736,7 @@
   [(set (pc)
 	(if_then_else (match_operator 0 "comparison_operator"
 			[(match_operand:DI 1 "arith_operand" "r,r")
-			 (match_operand:DI 2 "arith_operand" "rN,i")])
+			 (match_operand:DI 2 "arith_operand" "rN,I08")])
 		      (label_ref (match_operand 3 "" ""))
 		      (pc)))
    (clobber (match_scratch:SI 4 "=X,&r"))