Missed integer promotion for array_ref?

[Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>]

(last mail about all this unsigned vs. signed stuff...)

int a[1024];
int *bar(int i) { return a+i; }
int *foo(int i) { return &a[i]; }

is gimplified (for C++) to

int* bar(int) (i)
{
  int * D.1566;
  unsigned int i.0;
  unsigned int D.1568;
  int * D.1569;

  i.0 = (unsigned int) i;
  D.1568 = i.0 * 4;
  D.1569 = (int *) D.1568;
  D.1566 = D.1569 + &a[0];
  return D.1566;
}

int* foo(int) (i)
{
  int * D.1573;
  int i.1;

  i.1 = i;
  D.1573 = &a[i.1];
  return D.1573;
}

as i is not casted to unsigned int in foo(), &a[i] is not the same
as a + i here.  Of course for negative i the results of bar and foo
are undefined anyway, but this difference in signdedness of the offset
of ARRAY_REF confuses the optimizers (and made me dig into this).

Richard.

--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/