Is (set (subreg:SI (reg:HI) 0) (const_int)) valid rtx?

[Kazu Hirata]

Hi Richard,

> > I had "before register allocation" in mind and assumed a target
> > architecture allows (set (reg:SI) (const_int)).  Given these, would
> > you also say that the validity depends on WORD_REGISTER_OPERATIONS as
> > Joern suggested?
> 
> No, I wouldn't think so.  Seems like a "normal" case of
> a paradoxical subreg to me.  It will require less than
> desirable work-arounds from reload, if it makes it that
> far, but it should *work*.

Hmm.  So if local-alloc causes an ICE due to

  (set (subreg:SI (reg:HI x) 0)
       (...)),

then is it probably a bug in local-alloc, not somebody that creates
the above rtx!?

Kazu Hirata