Re: [PATCH] reorganize simplify_const_relational_operation

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

> Thanks.  Btw, what do you think about the extension code just above?
>
>       /* If WIDTH is nonzero and smaller than HOST_BITS_PER_WIDE_INT,
> 	 we have to sign or zero-extend the values.  */
>       if (width != 0 && width < HOST_BITS_PER_WIDE_INT)
> 	{
> 	  l0u &= ((HOST_WIDE_INT) 1 << width) - 1;
> 	  l1u &= ((HOST_WIDE_INT) 1 << width) - 1;
>
> 	  if (l0s & ((HOST_WIDE_INT) 1 << (width - 1)))
> 	    l0s |= ((HOST_WIDE_INT) (-1) << width);
>
> 	  if (l1s & ((HOST_WIDE_INT) 1 << (width - 1)))
> 	    l1s |= ((HOST_WIDE_INT) (-1) << width);
> 	}
>       if (width != 0 && width <= HOST_BITS_PER_WIDE_INT)
> 	h0u = h1u = 0, h0s = HWI_SIGN_EXTEND (l0s), h1s = HWI_SIGN_EXTEND (l1s);
>
> Why l0s and l1s are not (explicitly) sign-extended in the positive case?

Actually I think that the sign-extension is useless, only the l0u and 
l1u zero-extension does something.  A QImode (8-bit) pattern "10000000" 
should always be represented sign-extended, i.e. as (const_int -128), as 
far as trunc_int_for_mode tells me:

  /* Sign-extend for the requested mode.  */

  if (width < HOST_BITS_PER_WIDE_INT)
    {
      HOST_WIDE_INT sign = 1;
      sign <<= width - 1;
      c &= (sign << 1) - 1;
      c ^= sign;
      c -= sign;
    }

Paolo