...

The interchanged loop traverses the iteration space top-to-bottom, left-to-right.

 for k   := ? to ? do

  for k   := ? to ? do

S: A[k  ,k  -k  ] := A[k  -1,k  -k  ]+A[k  ,k  -k  -1]

The inner loop is now vectorisable.

The skewed iteration space has N rows and 2N-1 columns, but still only actual statement instances.

What are the appropriate loop bounds?