[Bug middle-end/45722] [4.6 Regression] FAIL: gcc.c-torture/execute/20040709-2.c execution at -O1 and -Os

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45722

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |uweigand at gcc dot gnu.org

--- Comment #29 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-12 12:19:15 UTC ---
We indeed do rely on type align info on the MEM_REF (which I think is not
the very best idea, but I ended up re-instantiating that because it's all
a mess).  So if SRA rewrites an access it has to transfer misaligned
info to the MEM_REF.  OTOH, given that we have

;; x$l_3 = MEM[(struct B *)&sB + 2B];

we know the alignment exactly because we can infer it from the DECL
alignment of sB.  So we might run into the

void
set_mem_attributes_minus_bitpos (rtx ref, tree t, int objectp,
                                 HOST_WIDE_INT bitpos)
{
  alias_set_type alias = MEM_ALIAS_SET (ref);
  tree expr = MEM_EXPR (ref);
  rtx offset = MEM_OFFSET (ref);
  rtx size = MEM_SIZE (ref);
  unsigned int align = MEM_ALIGN (ref);
                      ^^^^^^^^^^^^^

bug on strict-alignment targets again here.  Uli promised to fix that up
but hasn't delivered yet.