In practice, this should mean that WFA/diagonal transition is relatively better for unit costs and small affine costs, whereas edlib is better when affine costs are large. I did implement band doubling for affine costs as part of A*PA in a NW aligner that can handle arbitrary cost models, but for now this implementation is focused on the algorithm rather than efficiency, and it does not use any kind of SIMD and/or bit-packing. WFA vs band doubling for affine costs Link to heading