[Bug tree-optimization/26406] Fowardprop does harm for VRP to figure out if a point is non zero

[rguenth at gcc dot gnu dot org]

------- Comment #5 from rguenth at gcc dot gnu dot org  2006-02-22 12:47 -------
Of course we have the other case where we _have_ to propagate to optimize away
the test (gcc.dg/tree-ssa/20030807-2.c):

foo(int n)
{
  int *space = (int *)__builtin_alloca (n);

  if (space == 0)
    abort ();
  else
    bar (space);
}

where there is a 2nd use of space:

foo (n)
{
  int * space;
  void * D.1899;
  long unsigned int D.1898;

<bb 2>:
  D.1898_2 = (long unsigned int) n_1;
  D.1899_3 = __builtin_alloca (D.1898_2);
  space_4 = (int *) D.1899_3;
  if (space_4 == 0B) goto <L0>; else goto <L1>;

<L0>:;
  abort ();

<L1>:;
  bar (space_4);
  return;

}

So I think extending VRP is the only way to get both testcases optimized.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26406