[Bug target/122583] [16 Regression] GCN 'gcc.dg/vect/vect-align-2.c' vs. 'vect: Remove type from misalignment hook'

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122583

--- Comment #4 from Robin Dapp <rdapp at gcc dot gnu.org> ---
Ok, thanks, then it's no gather/scatter ;)

If I'm not missing something obvious the target hook returns false where it
returned true before.  As said, the difference is that is_packed is now always
computed rather than just when misalignment = DR_MISALIGNMENT_UNKNOWN.

Before Andrew's change we had

-  /* If the misalignment is unknown, we should be able to handle the access
-     so long as it is not to a member of a packed data structure.  */
-  if (misalignment == -1)
-    return true;

(but is_packed wasn't computed for that case).

So semantics wise I'd suggest always returning true but ensuring that no scalar
subregs of vector modes are allowed.  Maybe via gcn_can_change_mode_class?