Re: [autovect] [patch] Dependence tests and dimensions

[Sebastian Pop <sebastian dot pop at cri dot ensmp dot fr>]

Ira Rosen wrote:
> 
> Yes, but p can point to a[0], i.e., two data-refs can access the same
> memory even though their dimensions differ. Do you agree?
> 

Right.

But, why not keeping this simple test when we're sure both data refs
are arrays?  Something like this:

  /* When A and B are arrays and their dimensions differ, we directly
     initialize the relation to "there is no dependence": chrec_known.  */
  if (DR_BASE_OBJECT (a) && DR_BASE_OBJECT (b)
      && DR_NUM_DIMENSIONS (a) != DR_NUM_DIMENSIONS (b))
    {
      DDR_ARE_DEPENDENT (res) = chrec_known;
      return res;
    }

The test is so simple and avoids to process inhomogeneous cases in
later tests.