https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77550
--- Comment #16 from Bernd Edlinger <bernd.edlinger at hotmail dot de> --- (In reply to Richard Biener from comment #15) > The patch should at least work (loads are similarly affected). At some > point we want some infrastructure in the middle-end to chose a better > fallback (as Andrew says, if possible we can use the alias-set of the > containing aggregate for example). Good. Then I go ahead and look for a way how to fix all of the different load and store code path. If the whole group does not agree in the alias_set I would fall back to ptr_type_node, I think if I put that code in an extra function it can be changed later to the alias-set of the containing aggregate.
