spaceship_replacement cannot see through simplified set of FP conditionals

[Aldy Hernandez <aldyh@redhat.com>]

Hi.

With my upcoming patch enabling floating point VRP,
g++.dg/opt/pr94589-2.C is failing:

https://gcc.gnu.org/pipermail/gcc-patches/2022-July/598788.html

The problem is that phiopt no longer sees the following snippet,
because we have folded away the final conditional as true:

bool f5 (double i, double j)
{
  signed char __v$_M_value;
  signed char c$_M_value;
  bool D.8519;
  struct partial_ordering __v;
  struct partial_ordering c;
  unsigned int _11;
  bool _16;

  <bb 2> :
  if (i_2(D) != j_4(D))
    goto <bb 3>; [INV]
  else
    goto <bb 6>; [INV]

  <bb 3> :
  if (i_2(D) >= j_4(D))
    goto <bb 4>; [INV]
  else
    goto <bb 6>; [INV]

  <bb 4> :
  if (i_2(D) > j_4(D))
    goto <bb 6>; [INV]
  else
    goto <bb 5>; [INV]

  <bb 5> :

  <bb 6> :
  # c$_M_value_3 = PHI <-1(3), 0(2), 2(5), 1(4)>
  _11 = (unsigned int) c$_M_value_3;
  _16 = _11 <= 1;
  return _16;
}

This means that spaceship_replacement()  now sees one less argument to the PHI:

  <bb 2> :
  if (i_2(D) != j_4(D))
    goto <bb 3>; [INV]
  else
    goto <bb 5>; [INV]

  <bb 3> :
  if (i_2(D) >= j_4(D))
    goto <bb 4>; [INV]
  else
    goto <bb 5>; [INV]

  <bb 4> :

  <bb 5> :
  # c$_M_value_3 = PHI <-1(3), 0(2), 1(4)>

...and we bail because we are expecting 4 arguments:

      if (EDGE_COUNT (phi_bb->preds) != 4)
    return false;

Could someone give me a hand here? Can spaceship_replacement easily be
adapted to handle this case, or should we be simplifying this
elsewhere?

Thanks.
Aldy