[Bug tree-optimization/26602] New: final_cleanup can mess up incoming frequencies

[Bug tree-optimization/26602] New: final_cleanup can mess up incoming frequencies

[pinskia at gcc dot gnu dot org]

An example (which I got from looking a little at varasm.c):
struct a
{
  int i;
};

int f(int *);

int g(struct a *b)
{
  struct a *b1 =b;
  int t = 0;
  for(;b < b1+3;b++)
    {
      if (b==0) return 1;
      int *c = &b->i;
      if (c == 0) return 1;
      t+=f(c);
    }
  return t;
}

-------

Currently we get:
Removing basic block 9
Merging blocks 2 and 3
g (b)
{
  struct a * b.35;
  int t;
  int D.1525;

<bb 2>:
  if (b == 0B) goto <L7>; else goto <L20>;

Invalid sum of incoming frequencies 2731, should be 2048
<L20>:;
  b.35 = b;
  t = 0;
  goto <bb 5> (<L2>);

<L0>:;
  if (b.35 == 0B) goto <L7>; else goto <L2>;

<L2>:;
  D.1525 = f (&b.35->i);
  t = t + D.1525;
  b.35 = b.35 + 4B;
  if (b.35 != (struct a *) (b + 12B)) goto <L0>; else goto <L21>;

<L7>:;
  t = 1;

Invalid sum of incoming frequencies 2156, should be 2875
<L21>:;
  return t;


-- 
           Summary: final_cleanup can mess up incoming frequencies
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26602

[Bug tree-optimization/26602] final_cleanup can mess up incoming frequencies

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2006-03-11 05:00 -------
Hmm, what is happening here is we fold the conditional and then update the
frequency for the next bb but not the ones after it which causes us to get the
warnings:
Invalid sum of incoming frequencies 2731, should be 2048

The numbers before the cleanup:
  # BLOCK 9 freq:719
  # BLOCK 3 freq:2156

After:
  # BLOCK 2 freq:2875


But:
  # BLOCK 6 freq:375
  # BLOCK 3 freq:2048


2875-375 == 2731 != 2048


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26602