https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113138
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #1)
> Actually the following list is correct:
> signed
> X s< ~X --> X s< 0
> X s<= ~X --> X s< ~X --> X s< 0
>
> X s> ~X --> X s> 0
> X s>= ~X --> X s> ~X --> X s> 0
>
> unsigned:
> X u< ~X --> X u> SIGNBIT_OF(X)
> X u<= ~X --> X u< ~X --> X u> SIGNBIT_OF(X)
>
> X u> ~X --> X u< SIGNBIT_OF(X)
> X u>= ~X --> ~X u> X --> X u< SIGNBIT_OF(X)
>
> So the correct idea is this:
> (for (cmp lt le gt ge)
> (rcmp gt gt lt lt)
> (simplify
> (cmp @0 @1)
> (with { bool wascmp; }
> (if (bitwise_inverted_equal_p (@0, @1, wascmp)
> && (!wascmp || element_precision (type) == 1))
> (with {
> tree inner_type = TREE_TYPE (@0);
> int precision = TYPE_PRECISION (inner_type);
> // the min value here is in the opposite signedness.
> auto ncst = wi::min_value (precision, TYPE_UNSIGNED (inner_type) ?
> SIGNED : UNSIGNED);
> }
> (rcmp @0 { wide_int_to_tree (inner_type, ncst); })
> )
> )
> )
Will something then fold it further if it is actually ~X s< X (because the
above would then fold it into ~X s< 0)?
