[Bug tree-optimization/76957] [7 regression] FAIL: gcc.dg/graphite/scop-dsyr2k.c scan-tree-dump-times graphite "number of SCoPs

[amker at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=76957

--- Comment #9 from amker at gcc dot gnu.org ---
(In reply to amker from comment #8)
> I will have a look.  Thanks.

Tree dump is as below:

;; Function dsyr2k (dsyr2k, funcdef_no=0, decl_uid=4135, cgraph_uid=0,
symbol_order=3)

dsyr2k (long int N)
{
  int k;
  int j;
  int i;
  double _1;
  double _9;

  <bb 2> [0.47%]:
  if (N_17(D) > 0)
    goto <bb 3>; [85.00%]
  else
    goto <bb 12>; [15.00%]

  <bb 3> [0.40%]:
  goto <bb 11>; [100.00%]

  <bb 4> [12.75%]:

  <bb 5> [15.00%]:
  # j_3 = PHI <j_19(4), 0(11)>

  <bb 6> [85.00%]:
  # k_22 = PHI <k_21(7), j_3(5)>
  _1 = c[j_3][k_22];
  _9 = _1 + 0.0;
  c[j_3][k_22] = _9;
  k_21 = k_22 + 1;
  if (N_17(D) > k_21)
    goto <bb 7>; [82.35%]
  else
    goto <bb 8>; [17.65%]

  <bb 7> [70.00%]:
  goto <bb 6>; [100.00%]

  <bb 8> [15.00%]:
  j_19 = j_3 + 1;
  if (N_17(D) > j_19)
    goto <bb 4>; [85.00%]
  else
    goto <bb 9>; [15.00%]

  <bb 9> [2.65%]:
  i_18 = i_8 + 1;
  if (N_17(D) > i_18)
    goto <bb 10>; [85.00%]
  else
    goto <bb 12>; [15.00%]

  <bb 10> [2.25%]:

  <bb 11> [2.65%]:
  # i_8 = PHI <i_18(10), 0(3)>
  goto <bb 5>; [100.00%]

  <bb 12> [0.47%]:
  return;

}

the information is already there that (long int)j_3 >= N_17(D) is always false:
# j_3 = PHI <j_19(4), 0(11)>
and we have dominant condition for both PHI arguments:
if (N_17(D) > j_19)
and 
if (N_17(D) > 0)

We just need to find a way to use it.

It would be easy to fix if we have flow-sensitive value range, and support
symbolic range better in VRP.  Which we don't have either one for now.


Another choice is to improve simplify_using_initial_conditions or
simplify_using_outer_evolutions.  Function fact simplify_using_outer_evolutions
doesn't help at all in this case.