Dev Ojha

For a circuit I'm working on, this saves ~20k to 30k constraints. (in general any circuit which has exponentiations of the same base to several constant exponents should get savings...

The circuit is not publicly available at the moment, but its on the order of 1-2 million constraints. These duplicate constraints arise from having to exponentiate a variable to multiple...

The turbo-plonk IR is really AIR, currently the only front-end for AIR I'm aware of is: https://github.com/GuildOfWeavers/AirScript, though this tracking issue from Coda seems to be a more limited starting...

> I think that perhaps in a plonk style argument, this sort of optimisation doesn't make too much sense, as we are using the Lagrange basis, rather than the monomial...

Even if we had an IR with optimizations, I think this should still be a useful optimization unique to R1CS / Plonk's intermediate representation. (Or I've misunderstood how the envisioned...

I've used that flame graph package and gotten it to work before. In golang theres a tool called `pprof` which is amazing though and I vastly prefer. e.g. here is...

Yes! (Didnt know there was a term for this)

> One potential direction is to let AllocatedFp uses a linear combination of variables instead of a symbolic LC, if it is optimized for the number of constraints. I think...

I feel like we will need to eventually remove `Copy` from Variable anyway. Even in the case of optimizing for density, I think we want to add some drop semantics...

In the case of density optimizations, I think we want to keep building it up in `LC` representation, and condense that into a saved `Symbolic LC` every time that variable...