[Bug middle-end/94715] Squared multiplies are incorrectly signextended

[wilco at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94715

Wilco <wilco at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-04-23
     Ever confirmed|0                           |1
             Status|RESOLVED                    |REOPENED
         Resolution|INVALID                     |---

--- Comment #2 from Wilco <wilco at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #1)
> I think GCC is correct in assuming that x * x is positive since overflow
> with signed arithmetic is undefined.  Thus on GIMPLE we elide
> 
>   _1 = x_2(D) * x_2(D);
>   t_3 = (unsigned int) _1;
>   _4 = (long long unsigned int) t_3;
> 
> to
> 
>   _1 = x_2(D) * x_2(D);
>   _4 = (long long unsigned int) _1;

If we assume x * x is always positive, using unsigned extension would make more
sense. It still adds an unnecessary extra instruction on most targets which
cannot be removed in RTL.