But looks more like they're giving up on people writing code for wide vectors, instead settling on trying to make the existing code faster.