[Bug middle-end/59734] Simple strict-volatile-bitfields case not working

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59734

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-01-09
                 CC|                            |jason at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, the C++ memory model only disallows this because the padding might be
re-used in a derived class like

struct foo : str {
  short x;
};

but at least 'str' has to be non-POD for this to happen it seems.  That is, for

struct str {
    str() {}
    volatile unsigned f1: 8;
};
struct foo : str { short x; };

sizeof (foo) == sizeof (str) == 4 and thus accessing f1 with SImode would
introduce a data race with 'x' which is not allowed.

If the FE would provide a way to tell whether tail padding might be possibly
re-used we can avoid this for the simple case.  Jason?

That is, add langhook that tells us this (or a flag on the types).