[Bug tree-optimization/33434] [4.3 Regression] -fipa-cp miscompilation

[jakub at gcc dot gnu dot org]

------- Comment #2 from jakub at gcc dot gnu dot org  2007-10-12 13:38 -------
Doesn't seem to look like a problem in ipa-cp to me, more like an inlining bug.
ipa-cp creates:
T.1 (a, b)
{
  _Bool D.1565;

<bb 2>:
  if (1)
    goto <bb 3>;
  else
    goto <bb 6>;

<bb 3>:
  # b_3 = PHI <1(2)>
  goto <bb 5>;

<bb 4>:
  k ={v} 1;

<bb 5>:
  # b_2 = PHI <b_3(3), b_1(4)>
  b_1 = b_2 + -1;
  if (b_2 != 0)
    goto <bb 4>;
  else
    goto <bb 8>;

<bb 6>:
  if (0)
    goto <bb 7>;
  else
    goto <bb 8>;

<bb 7>:
  __builtin_abort ();

<bb 8>:
  return;

}

and even the first dump during apply_inline looks correct, inside of main we
have:
...
  # BLOCK 4 freq:4999
  # b_6 = PHI <1(3)>
  goto <bb 6>;

  # BLOCK 5 freq:10000
  k ={v} 1;

  # BLOCK 6 freq:10000
  # b_5 = PHI <b_6(4), b_4(5)>
  b_4 = b_5 + -1;
  if (b_5 != 0)
    goto <bb 5>;
  else
    goto <bb 9>;

But during inlining both a and b are mark_sym_for_renaming and when update_ssa
is called, rewrite_update_stmt -> maybe_replace_use changes that
  if (b_5 != 0)
into incorrect:
  if (b_4 != 0)
so k is never set to 1.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33434