S00 | ||||||||||||
S01 | S11 | |||||||||||
S02 | S12 | S22 | ||||||||||
S03 | S13 | S23 | S33 | |||||||||
S14 | S24 | S34 | ||||||||||
S25 | S35 | |||||||||||
S36 |
for k2 := ? to ? doThe inner loop is now vectorisable.
for k1 := ? to ? do
S: A[k1,k2-k1] := A[k1-1,k2-k1]+A[k1,k2-k1-1]
The skewed iteration space has N rows and 2N-1 columns, but still only N2 actual statement instances.
What are the appropriate loop bounds?