Re: [RFA] A patch for PR 25115

[Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>]

> > +  if (!set)
> > +    {
> > +      rtx note = find_reg_equal_equiv_note (insn);
> > +      gcc_assert (note);
> > +      new_insn = gen_move_insn (reg, XEXP (note, 0));
> > +      new_insn = emit_insn_after (new_insn, insn);
> > +
> > +      /* Keep register set table up to date.  */
> > +      record_one_set (regno, new_insn);
> > +    }
> >    
> I don't understand how this is supposed to work in the general case.
>
> The expression in the note may very well be of a form that *cannot*
> be re-computed by a simple gen_move_insn -- there may not be any
> single pattern that does the computation, or the pattern may require
> additional CLOBBERs etc.
>  
The note's argument is entered in the hash table only if 
can_assign_to_reg_p returns true, and this checks whether it is a 
general_operand tries to recognize the instruction if it is not a 
general_operand (which covers most interesting cases, including 
SYMBOL_REF expressions), or if a pattern exists that matches the insn.

In particular, clobbers of pseudos are taken care of in emit_move_insn_1 
(which is called by gen_move_insn) when it expands a define_expand), and 
clobbers of hard registers are excluded by can_assign_to_reg_p, because 
it does

 return ((icode = recog (PATTERN (test_insn), test_insn, 
&num_clobbers)) >= 0
         && (num_clobbers == 0 || ! added_clobbers_hard_reg_p (icode)));

Paolo