[Bug tree-optimization/84646] Missed optimisation for hoisting conditions outside nested loops

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84646

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aldyh at gcc dot gnu.org

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jeffrey A. Law from comment #2)
> The backwards threader discovers the threads, but refuses to optimize them
> because it thinks doing so will create an irreducible loop without enough
> benefit.

The reason is now:

Checking profitability of path (backwards):  bb:3 (2 insns) bb:7 (4 insns) bb:6
(1 insns) (latch) bb:5
  Control statement insns: 2
  Overall: 5 insns

 Registering value_relation (path_oracle) (j_17 < m_23(D)) (root: bb5)
path: 5->6->7->3->xx REJECTED (unreachable)

it seems the path oracle is confused here.


  <bb 3> [local count: 955630225]:
  if (running_14 != 0)
    goto <bb 4>; [50.00%]
  else
    goto <bb 6>; [50.00%]

  <bb 4> [local count: 477815112]:
  _1 = (long unsigned int) i_16;
  _2 = _1 * 4;
  _3 = p_25(D) + _2;
  _4 = *_3;
  _5 = (long unsigned int) j_17;
  _6 = _5 * 4;
  _7 = q_26(D) + _6;
  _8 = *_7;
  _9 = _4 * _8;
  sum_27 = _9 + sum_11;
  if (sum_27 > 19999)
    goto <bb 5>; [50.00%]
  else
    goto <bb 6>; [50.00%]

  <bb 5> [local count: 238907556]:

  <bb 6> [local count: 955630225]:
  # sum_10 = PHI <sum_11(3), 20000(5), sum_27(4)>
  # running_13 = PHI <0(3), 0(5), 1(4)>
  j_28 = j_17 + 1;

  <bb 7> [local count: 1073741824]:
  # sum_11 = PHI <sum_12(11), sum_10(6)>
  # running_14 = PHI <running_15(11), running_13(6)>
  # j_17 = PHI <0(11), j_28(6)>
  if (j_17 < m_23(D))
    goto <bb 3>; [89.00%]
  else
    goto <bb 8>; [11.00%]