Strategies going forward

I have been doing a bit of thinking while on the bus, and I speculated if this here might be a plausible strategy for the thesis, going forward:

Supporting propositions:

• Our survey yields that Nikola looks most promising for further development
• Portfolio pricing using the binomial pricer is a nested data parallel problem, though not very intricate.
• NESL-style vectorisation deals with the entire class of nested data parallel problems, but it not always a net gain (\cite{copperhead}).

So:

We want to see how well we can equip Nikola with some version of the vectorisation transformation, and compare the resulting performance to the unvectorised Nikola (and Vector,R,C,CUDA,Repa) and try to identify what causes any significant differences in performance. Then we may conclude how much overhead/speedup our vectorisation transformation incurs, and point out plausible optimisations to the transformation. (Since portfolio pricing (i.e. map binom) looks rather innocent from a nested-data point of view, it might actually be usable for measuring the overhead of vectorisation.)