https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101718
Iain Sandoe <iains at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |iains at gcc dot gnu.org
Status|NEW |ASSIGNED
--- Comment #4 from Iain Sandoe <iains at gcc dot gnu.org> ---
Created attachment 51310
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51310&action=edit
trial fix
So it seems that the logic was slightly off, but in a way that would not fire
unless a non-aggregate return-in-memory occurred, and those are very rare
(non-existent, with the clang ABI) for Darwin which is the only NeXT v2 user
with test coverage.
the change is:
if (ret_type && !VOID_TYPE_P (ret_type)
&& targetm.calls.return_in_memory (ret_type, 0)
&& !(targetm.calls.struct_value_rtx (0, 0)
&& (TREE_CODE (ret_type) == RECORD_TYPE
|| TREE_CODE (ret_type) == UNION_TYPE)))
So this says, "if there's a return value and it's returned in memory and the
target makes no special provision for returning aggregates then use the _stret
versions".
Probably some scope for factoring some of that code - but bug fixes first.
Does this work for your platform?