reload fix

[Richard Henderson]

On Tue, May 14, 2002 at 11:57:41AM +0200, Jan Hubicka wrote:
> I am not quite sure if the attached patch is the proper fix.  I think idea of
> constraint_accepts_reg_p is simply broken,

I agree.  There's no way we can check one operand in isolation like this.

> 	* reload1.c (emit_input_reload_insns): Use constrain_operands
> 	instead of constraint_accepts_reg_p to verify optimization.
> 	(constraint_accepts_reg_p): Kill

Ok.


r~