[dingdingdang]

It always amazes me that this is possible (to some extend anyway, I mean, the base layer is binary so obviously simpler higher-end CPU instructions are possible!)

Is there any potential performance win in this? What I mean is; since this general direction could, in principle if not in practise, enable the targeting of say, the 5-10 most efficient CPU instructions rather than attempting to use the whole surface area... would this potentially be a win?