[Bug tree-optimization/34677] New: PREs insert_fake_stores is mostly useless

[rguenth at gcc dot gnu dot org]

Because it only handles stores through direct INDIRECT_REFs.  So while we
handle
testcases like loadpre4.c:

int main(int *a, int argc)
{
  int b;
  int c;
  int i;
  int d, e;

  for (i = 0; i < argc; i++)
    {
      e = *a;
      *a = 9;
    }
  return d + e;
}

the much simpler

int a;

int foo(int argc)
{
  int b;
  int c;
  int i;
  int d, e;

  for (i = 0; i < argc; i++)
    {
      e = a;
      a = 9;
    }
  return d + e;
}

or the only slightly more complex

struct {
  int a;
  int large[100];
} x;

int foo(int argc)
{
  int b;
  int c;
  int i;
  int d, e;

  for (i = 0; i < argc; i++)
    {
      e = x.a;
      x.a = 9;
    }
  return d + e;
}

is not handled.  If you disable insertion of fake stores completely, only
two loadpre tests fail in tree-ssa.exp.  Did the new SCCVN obsolete some
cases where we needed the fake stores before but do not so now?


-- 
           Summary: PREs insert_fake_stores is mostly useless
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34677