[Bug tree-optimization/54717] [4.8 Regression] Runtime regression: polyhedron test "rnflow" degraded

["hubicka at ucw dot cz" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54717

--- Comment #13 from Jan Hubicka <hubicka at ucw dot cz> 2012-11-14 19:43:00 UTC ---
> So for the loop that starting at bb 28 you can see the xxtrt_46 access was not
> put into pretemp. Possible reason is exactly as it was mentioned by Richard -
> there were extra candidates collected and this one become less anticipatable
> 
> Skipping partial partial redundancy for expression                    
> {array_ref<pretmp_8,0,4>,mem_ref<0B>,xxtrt_46(D)}@.MEM_30(D) (0165)   
>    not partially anticipated on any to be optimized for speed edges    
>   -----------------------------------------------------------------------
> Found partial partial redundancy for expression
>  {array_ref<pretmp_8,0,4>,mem_ref<0B>,xxtrt_46(D)}@.MEM_30(D) (0165)
> Created phi prephitmp_237 = PHI <_88(90), _85(29)>
>  in block 30

Hmm, interesting, what is the edge resonsible?
I would expect it to be the loopback edge and its frequency is:
;;   basic block 28, loop depth 0, count 0, freq 1998, maybe hot
;;    prev block 92, next block 94, flags: (NEW, REACHABLE)
;;    pred:       92 [100.0%, 180]  (FALLTHRU)
;;                96 [100.0%, 1818]  (FALLTHRU,DFS_BACK)
  # ival2_136 = PHI <ival2_62(92), ival2_144(96)>
  # ival2_140 = PHI <ival2_80(92), ival2_146(96)>
  _137 = (integer(kind=8)) ival2_136;
  _138 = _137 + -1;
  _139 = *xxtrt_25(D)[_138];
  _141 = (integer(kind=8)) ival2_140;
  _142 = _141 + -1;
  _143 = *xxtrt_25(D)[_142];
  if (_139 < _143)
    goto <bb 29>; 
  else            
    goto <bb 94>;

1818 that should be still hot.  Or isn't the heuristic backwards? I.e. I would
expect
the partial anticipance to sit on edge 92->28 (with freq 180) where we need to
insert
the computation to get the other path hot.

Honza