> On Thu, Nov 12, 2020 at 11:29:21AM +0100, Jan Hubicka wrote:
> > > If OEP_ADDRESS is used also on non-addressable stuff, just to compare
> > > that two COMPONENT_REFs access the same memory, then just comparing
> > > DECL_BIT_FIELD_REPRESENTATIVE is not sufficient, you could have:
> > > struct S { int c; int a : 7, b : 1; };
> > > struct T { int c; int a : 7, b : 1; };
> > > and compare s->a vs. t->b with OEP_ADDRESS and the offsets of their
> > > DECL_BIT_FIELD_REPRESENATIVE is the same, yet we don't want to say
> > > the two bit-fields are the same.
> >
> > You are right, I was just thinking of that. I suppose it indeed makes
> > more sense to assert that there are no bitfields here and in the AO
> > comparsion take care of stripping the last bitfield reference and
> > handling it specially?
>
> Or just compare DECL_FIELD_OFFSET and DECL_FIELD_BIT_OFFSET of the fields
> rather than their DECL_BIT_FIELD_REPRESENTATIVE?
I think I will need to compare bitfields specially at the ao_ref_compare
side anyway to distinguish
struct S { int c; int a : 5, b : 1; };
struct T { int c; int a : 5, b : 3; };
s->b and t->b. Those have same base address (bitwise) but still we do
not want to consider them equal.
So handling this on operand_equal_p is probably not that useful and
perhaps extra sanity check would be. I will fire boostrap to see if
there is any other place that trips bitfields with OEP_ADDRESS_OF.
Thanks,
Honza
