On Mon, 3 Apr 2017, Richard Biener wrote:
--- gcc/match.pd (revision 246642)
+++ gcc/match.pd (working copy)
@@ -1153,7 +1153,14 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(if (tree_nop_conversion_p (type, TREE_TYPE (@0))
&& tree_nop_conversion_p (type, TREE_TYPE (@1))
&& !TYPE_OVERFLOW_SANITIZED (type))
- (minus (convert @0) (convert @1))))
+ (with
+ {
+ tree t1 = type;
+ if (INTEGRAL_TYPE_P (type)
+ && TYPE_OVERFLOW_WRAPS (type) != TYPE_OVERFLOW_WRAPS (TREE_TYPE (@1)))
+ t1 = TYPE_OVERFLOW_WRAPS (type) ? type : TREE_TYPE (@1);
+ }
+ (convert (minus (convert:t1 @0) (convert:t1 @1))))))
Why do we match
(plus:c (convert1? @0) (convert2? (negate @1)))
instead of the simpler
(plus:c @0 (convert? (negate @1)))
/* A - (-B) -> A + B */
Actually, I believe this transformation has issues as well, it seems
possible to transform X-INT_MIN into X+INT_MIN if INT_MIN comes as
(int)(-(unsigned)INT_MIN).
--
Marc Glisse
