[Bug c++/78655] gcc doesn't exploit the fact that the result of pointer addition can not be nullptr

[vanyacpp at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78655

--- Comment #4 from Ivan Sorokin <vanyacpp at gmail dot com> ---
(In reply to Richard Biener from comment #1)

> +  if (flag_delete_null_pointer_checks
> +      && is_gimple_assign (stmt)
> +      && is_gimple_assign (stmt)

Duplicate conjunct.

> +      && gimple_assign_rhs_code (stmt) == POINTER_PLUS_EXPR
> +      && tree_expr_nonzero_p (gimple_assign_rhs2 (stmt)))
> +    return true;
> +
>    return false;
>  }

I thought a bit more about the issue. I think that not only the result of
pointer-integer addition, but also its argument are values that could be proven
non-zero. Consider the following example:

bool f(int* a)
{
  bool x = a == nullptr;
  a += 10;
  return x;
}

In this example x could be proven to be false. Because if it is true then
(a+=10) causes undefined behavior. At the moment clang doesn't optimize this. I
don't any real example when this is needed. Just the way this optimization can
be made more widely applicable.