[Bug middle-end/69526] ivopts candidate strangeness

["rdapp at linux dot vnet.ibm.com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69526

--- Comment #13 from rdapp at linux dot vnet.ibm.com ---
Created attachment 38535
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=38535&action=edit
VRP/match.pd patch

Found some time again and hacked together a fix for match.pd and VRP. The patch
exports a VRP function no_range_overflow that is called from match.pd via
tree.c. I realize that VRP didn't export any functions so far and if there's a
better/idiomatic way to do it, please tell me. Further improvements or "code
smell" hints very welcome.

Currently this should fix cases like

long foo(int i)
{
  return (long) (i + 1) - 2;
}

Commutativity of a PLUS_EXPR is not yet being exploited and I'm relying on
TYPE_OVERFLOW_UNDEFINED so far. The overflow checking could also be done more
elegantly I suppose.

This diff is extracted from the full patch which also includes the RTL level
fix which I haven't included here since it has not changed. I hope the
splitting didn't introduce new bugs but there are no regressions on s390x for
the full patch.