This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[m32c] fix overzealous peephole
- From: DJ Delorie <dj at redhat dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Wed, 12 Nov 2008 11:35:06 -0500
- Subject: [m32c] fix overzealous peephole
Committed.
2008-11-12 DJ Delorie <dj@redhat.com>
* config/m32c/cond.md (cond_to_int peephole2): Don't eliminate the
insns if the intermediate value will be used later.
Index: config/m32c/cond.md
===================================================================
--- config/m32c/cond.md (revision 141786)
+++ config/m32c/cond.md (working copy)
@@ -297,22 +297,25 @@
(const_int -1))))]
"TARGET_A24"
"sceq\t%0\n\tbmgt\t1,%h0\n\tdec.w\t%0"
[(set_attr "flags" "x")]
)
-;; A cond_to_int followed by a compare against zero is essentially a no-op.
+;; A cond_to_int followed by a compare against zero is essentially a
+;; no-op. However, the result of the cond_to_int may be used by later
+;; insns, so make sure it's dead before deleting its set.
(define_peephole2
[(set (match_operand:HI 0 "mra_qi_operand" "")
(if_then_else:HI (lt (reg:CC FLG_REGNO) (const_int 0))
(const_int -1)
(if_then_else:HI (eq (reg:CC FLG_REGNO) (const_int 0))
(const_int 0)
(const_int -1))))
(set (reg:CC FLG_REGNO)
(compare (match_operand:HI 1 "mra_qi_operand" "")
(const_int 0)))
]
- "rtx_equal_p(operands[0], operands[1])"
+ "rtx_equal_p (operands[0], operands[1])
+ && dead_or_set_p (peep2_next_insn (1), operands[0])"
[(const_int 1)]
"")