------- Comment #11 from rguenth at gcc dot gnu dot org 2007-03-10 22:46
-------
Actually my patch is wrong as well. We just trigger it only with ians patch as
-[-INF,-1] is [1, +INF(OVF)] with overflowed +INF.
I'll re-test the following with my previous fix for negative shift counts.
Index: tree-vrp.c
===================================================================
*** tree-vrp.c (revision 122805)
--- tree-vrp.c (working copy)
*************** vrp_int_const_binop (enum tree_code code
*** 1554,1559 ****
--- 1554,1565 ----
&& (sgn1 >= 0
? !is_positive_overflow_infinity (val2)
: is_negative_overflow_infinity (val2)))
+ /* We only get in here with positive shift count, so the
+ overflow direction is the same as the sign of val1.
+ Actually rshift does not overflow at all, but we only
+ handle the case of shifting overflowed -INF and +INF. */
+ || (code == RSHIFT_EXPR
+ && sgn1 >= 0)
/* For division, the only case is -INF / -1 = +INF. */
|| code == TRUNC_DIV_EXPR
|| code == FLOOR_DIV_EXPR
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31115
