[Bug middle-end/61409] [4.9/5 regression] -Wmaybe-uninitialized false-positive with -O2

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61409

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com

--- Comment #10 from Jeffrey A. Law <law at redhat dot com> ---
Of particular interest is this PHI node at the start of BB6:

  # mw_1 = PHI <mw_9(D)(2), h_33(5)>


I vaguely remember that we had code that would optimize this case, specifically
we would ignore PHI alternatives associated with undefined uses.  If we did
that, then we'd create an equivalence mw_1 = h33 which would then propagate to
the use of mw_1 and replace it with h33 and avoid the false positive.  I
thought that was in the vrp/ccp propagation engine.