bit_value_binop initializes VAL regardless of the final mask. It even
has a comment to that effect:
/* Ensure that VAL is initialized (to any value). */
However, bit_value_unop, which in theory shares the same API, does not.
This causes range-ops to choke on uninitialized VALs for some inputs to
ABS.
Instead of fixing the callers, it's cleaner to make bit_value_unop and
bit_value_binop consistent.
OK for trunk?
gcc/ChangeLog:
* tree-ssa-ccp.cc (bit_value_unop): Initialize val when appropriate.
---
gcc/tree-ssa-ccp.cc | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/gcc/tree-ssa-ccp.cc b/gcc/tree-ssa-ccp.cc
index 73fb7c11c64..15e65f16008 100644
--- a/gcc/tree-ssa-ccp.cc
+++ b/gcc/tree-ssa-ccp.cc
@@ -1359,7 +1359,10 @@ bit_value_unop (enum tree_code code, signop type_sgn,
int type_precision,
case ABS_EXPR:
case ABSU_EXPR:
if (wi::sext (rmask, rtype_precision) == -1)
- *mask = -1;
+ {
+ *mask = -1;
+ *val = 0;
+ }
else if (wi::neg_p (rmask))
{
/* Result is either rval or -rval. */
@@ -1385,6 +1388,7 @@ bit_value_unop (enum tree_code code, signop type_sgn, int
type_precision,
default:
*mask = -1;
+ *val = 0;
break;
}
}
--
2.41.0
