https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113630
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> Confirmed.
>
> I really think what PRE does is correct here since we have an aliasing set
> of 0 for both. Now what is incorrect is hoist_adjacent_loads which cannot do
> either of any of the aliasing sets are 0 ...
>
>
>
> I think even the function below is valid for non-strict aliasing:
> ```
> int __attribute__((noipa,noinline))
> f(struct S *p, int c, int d)
> {
> int r;
> if (c)
> {
> r = ((struct M*)p)->a;
> }
> else
> r = ((struct M*)p)->b;
> return r;
> }
> ```
>
> That is hoist_adjacent_loads is broken for non-strict-aliasing in general
> and has been since 4.8.0 when it was added (r0-117275-g372a6eb8d991eb).
It looks it relies on
/* The zeroth operand of the two component references must be
identical. It is not sufficient to compare get_base_address of
the two references, because this could allow for different
elements of the same array in the two trees. It is not safe to
assume that the existence of one array element implies the
existence of a different one. */
if (!operand_equal_p (TREE_OPERAND (ref1, 0), TREE_OPERAND (ref2, 0), 0))
continue;
for the correctness test. Note the MEM accesses are of size sizeof (struct M).
With -fno-strict-aliasing we're not wiping that detail so I think it _is_
a bug in PRE that it merges the two accesses.
I'll have a more detailed look.
