[dataflow] partial register handling

[Paolo Bonzini]

Richard Sandiford wrote:
> Paolo Bonzini <bonzini@gnu.org> writes:
>> My opinion:
>>>> (set (subreg:HI (reg:SI) 2) ...
>>>> DF_REF_READ_WRITE?
>> yep, only case without PARTIAL
> 
> Sorry if this is a daft question, but why would it be treated as a read?
> I thought that without strict_lowpart, this sort of subreg assignment
> was morally equivalent to a normal REG definition in which the upper
> bits are "don't care".

Yes, I said elsewhere to ignore this message...

Paolo