Re: [tree-ssa] Remove useless null pointer checks

[Jeff Sturm <jsturm at one-point dot com>]

On Wed, 30 Jul 2003 law@redhat.com wrote:
> This is a first cut at removal of useless null pointer checks based
> on the knowledge that after a memory load/store we know the pointer
> used must not be null.

This is excellent.  Java emits tons of useless null pointer checks.
Without this patch, the class:

class A {
  int i;
  final int j() {
    return i;
  }
  static int sum(A a) {
    return a.i + a.j();
  }
}

would optimize to (in method sum):

A.sum(A) (a)
{
  int T.3;
  A:: * j.5;
  int T.6;

  T.3 = a->i;
  if (a == 0B)
    {
      _Jv_ThrowNullPointerException ()
    };
  j.5 = (A:: *)j;
  T.6 = j.5 (a);
  int = T.3 + T.6;
  return T.3 + T.6;
}

With your patch, I get:

A.sum(A) (a)
{
  int T.4;
  A:: * j.6;
  int T.7;

  T.4 = a->i;
  j.6 = (A:: *)j;
  T.7 = j.6 (a);
  return T.4 + T.7;
}

Somehow the RTL optimizers never caught on to this.

A related question came up just the other day on the java list.  There are
often-called functions (e.g. allocators) in libgcj that cannot return
null.  Is there a good way to tell the tree optimizers about these, to
eliminate even more useless null pointer checks?

Jeff