https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98069
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
The following helps:
diff --git a/gcc/tree-data-ref.c b/gcc/tree-data-ref.c
index 3bf460cccfd..a7606b3db99 100644
--- a/gcc/tree-data-ref.c
+++ b/gcc/tree-data-ref.c
@@ -754,9 +759,10 @@ split_constant_offset_1 (tree type, tree op0, enum
tree_code code, tree op1,
&& TYPE_PRECISION (type) >= TYPE_PRECISION (itype)
&& (POINTER_TYPE_P (type) || INTEGRAL_TYPE_P (type)))
{
- if (INTEGRAL_TYPE_P (itype) && TYPE_OVERFLOW_WRAPS (itype)
+ if (INTEGRAL_TYPE_P (itype)
&& (TYPE_PRECISION (type) > TYPE_PRECISION (itype)
- || TYPE_UNSIGNED (itype) != TYPE_UNSIGNED (type)))
+ || (TYPE_UNSIGNED (itype) != TYPE_UNSIGNED (type)
+ && TYPE_OVERFLOW_WRAPS (itype))))
{
/* Split the unconverted operand and try to prove that
wrapping isn't a problem. */
but I fear this will have quite some negative impact on dependence analysis.
I'm still not sure I have the issue nailed down properly :/
