Re: types in GIMPLE IR

[Andrew Pinski <pinskia@gmail.com>]

On Thu, Jun 29, 2023 at 12:10 PM Krister Walfridsson via Gcc
<gcc@gcc.gnu.org> wrote:
>
> On Thu, 29 Jun 2023, Richard Biener wrote:
>
> > IIRC we have some simplification rules that turn bit operations into
> > arithmetics.  Arithmetic is allowed if it keeps the values inside
> > [-1,0] for signed bools or [0, 1] for unsigned bools.
> >
> >> I have now verified that all cases seems to be just one operation of this
> >> form (where _127 has the value 0 or 1), so it cannot construct values
> >> such as 42. But the wide signed Boolean can have the three different
> >> values 1, 0, and -1, which I still think is at least one too many. :)
> >
> > Yeah, I'd be interested in a testcase that shows this behavior.
>
> I created PR 110487 with one example.
>
>
> >> I'll update my tool to complain if the value is outside the range [-1, 1].
>
> It is likely that all issues I have seen so far are due to PR 110487, so
> I'll keep the current behavior that complains if the value is outside the
> range [-1, 0].

Yes there are many similar to this all over GCC's folding.
In this case checking TYPE_PRECISION as described in the match.pd is
not even the right approach.
The whole TYPE_PRECISION on boolean types is definitely a big can of worms.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102622 is related but
that was signed boolean:1.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106053 is another related case.

For this weekend, I am going to audit some of the match patterns for
these issues.


>
>     /Krister