[Bug tree-optimization/19038] [4.0 Regression] out-of ssa causing loops to have more than one BB

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2004-12-24 12:42 -------
A smaller test case:

int
f (int temp2, int xlvj_)
{
  int temp1;
 
  for(;;) {
    temp1 = temp2*xlvj_;
    temp2 = temp1;
    if (temp1)
      break;
  }
  return xlvj_;
}
                                                                               
               

After out-of-ssa this gives the following tree dump:

;; Function f (f)
                                                                               
               
Analyzing Edge Insertions.
f (temp2, xlvj_)
{
  int temp1;
  int D.1471;
                                                                               
               
<L4>:;
                                                                               
               
<L0>:;
  temp1 = temp2 * xlvj_;
  if (temp1 != 0) goto <L2>; else goto <L7>;
                                                                               
               
<L7>:;
  temp2 = temp1;
  goto <bb 1> (<L0>);
                                                                               
               
<L2>:;
  return xlvj_;
                                                                               
               
}



As far as I can tell the following test case is semantically
equivalent to the one above:

int
f (int temp2, int xlvj_)
{
  for(;;) {
    temp2 = temp2*xlvj_;
    if (temp2)
      break;
  }
  return xlvj_;
}

But the tree dumps are definitely not the same, in the smaller
test case we don't have the extra copy:

;; Function f (f)
                                                                               
               
Analyzing Edge Insertions.
f (temp2, xlvj_)
{
  int D.1470;
                                                                               
               
<L4>:;
                                                                               
               
<L0>:;
  temp2 = temp2 * xlvj_;
  if (temp2 != 0) goto <L2>; else goto <L0>;
                                                                               
               
<L2>:;
  return xlvj_;
                                                                               
               
}

So apparently something is preventing temp1 and temp2 from being
coalesced when going out of ssa.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19038