[Bug middle-end/51994] [4.6/4.7 Regression] git-1.7.8.3 miscompiled due to negative bitpos from get_inner_reference

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51994

--- Comment #19 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-26 10:01:23 UTC ---
I agree, all callers of get_inner_reference need to cope with a negative
bitpos.  Those that forward it unchecked to functions that expect an
unsigned bitpos are broken.

Thus I think fixing the prototypes is correct.  If that exposes other
issues we have to fix them.  The issue in extract_split_bit_field
is obviously the same - unsigned prototype and unsigned offset in

  while (bitsdone < bitsize)
    {
      unsigned HOST_WIDE_INT thissize;
      rtx part, word;
      unsigned HOST_WIDE_INT thispos;
      unsigned HOST_WIDE_INT offset;

      offset = (bitpos + bitsdone) / unit;

also

      thispos = (bitpos + bitsdone) % unit;

might not be correct with a negative (bitpos + bitsdone).

extract_fixed_bit_field has the same prototype issue, so eventually we
want to simply account for them in the callers (if there are less).
Only memory operands may have a negative bitpos and those we should be
able to adjust via adjust_address (but by what amount?) to make bitpos
positive.

So you could say already the routines called from the get_inner_reference
callers should do that.

Eric, you should know this area the best - what do you recommend here?
[we could assert in the unsigned bitpos taking functions that the MSB
is not set on bitpos]