[Bug tree-optimization/64823] [5.0 Regression] false "may be used uninitialized", missed jump threading

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64823

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
      Known to work|                            |4.9.2
           Keywords|                            |diagnostic,
                   |                            |missed-optimization
   Last reconfirmed|                            |2015-01-28
          Component|c                           |tree-optimization
                 CC|                            |law at gcc dot gnu.org
     Ever confirmed|0                           |1
            Summary|[5.0 Regression] false "may |[5.0 Regression] false "may
                   |be used uninitialized"      |be used uninitialized",
                   |                            |missed jump threading
   Target Milestone|---                         |5.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
So the issue is we apply loop header copying here and correctly don't warn
about
the first loop body but figure the 2nd iteration is guarded with
effectively o->next != NULL which isn't the same predicate s was initialized
under.

This is also a missed jump-threading:

  <bb 2>:
  if (o_4(D) != 0B)
    goto <bb 3>;
  else
    goto <bb 15>;

  <bb 15>:
  goto <bb 4>;

  <bb 3>:
  s_7 = o_4(D)->se;

  <bb 4>:
  # s_16 = PHI <s_7(3), s_5(D)(15)>
  goto <bb 9>;

...

  <bb 9>:
  # o_10 = PHI <o_4(D)(4), o_11(8)>
  # s_13 = PHI <s_16(4), s_2(8)>
  if (o_10 != 0B)
    goto <bb 10>;
  else
    goto <bb 13>;

...

  <bb 13>:
  return;


Jeff - is DOM not able to see that?  The threading is in the loop header
but threads to the exit (that is, I'd thread the o_4(D) == 0 case to bb 13).
Looks like 4.9 does that in DOM1 already.