jules

To note: Sinsemilla with the default parameters `k := 10` yields `1537` constraints for a `1030` bit input. Sinsemilla with `k := 14` yields `1102` constraints for a `1036` bit...

Testing constraint counts for both bit windows: ``` sinsemilla 14 bit window: 1024 (1036) yields 1102 768 (770) yields 817 512 (518) yields 547 256 (266) yields 277 sinsemilla 10...

@ljedrz I've managed to squeeze out a lot of optimization, and managed around a 4-10% decrease in time taken to compute square roots (depending on field size, with `Fr` having...

For now we've decided to forgo the lookup variant of the Sarkar algorithm as the speedup may not be worth it.

This work should now be mostly wrapped up, cleaned and optimized for as far as I can manage naively. Next up to finish this will be to make it fit...

This idea is better suited as a CD pipeline which restarts testnet with the new master every time a PR is merged, to avoid long waiting times and general CI...

As an addendum to your point here, this can be combined very efficiently with GLV scalar multiplication, which can allow for ~50% speedups when performing group scalar muls. I've implemented...