Dev Ojha

I'm imagining a powerpoint, that has a brief descriptions of the various components of the stack, and known problems we face from them, e.g.: - P2P layer, security concerns, gets...

I'm in favor of wrapping k256 (https://github.com/RustCrypto/elliptic-curves/tree/master/k256). See Tony's blog post on it: https://iqlusion.blog/k256-crate-pure-rust-projective-secp256k1-library.

The current state of the FFT code is that the twiddle factors are computed once at the beginning of every FFT, and then re-used in every iteration of the FFT....

I took a look at the mixed addition formulae, and they are exactly the same for the short weierstrass curves. This leads me to believe that the speed difference there...

The FFT/MSM split depends on your circuit size and protocol choice. Agreed that improving it is important, especially at large circuit sizes. There are multiple polynomials. RS-IOP/AHP based SNARKs commit...

we could separate the unchecked serialization methods into its own trait, that is auto-implemented if canonical serialize is implemented

Im in favor of the approach of having 4 methods ``` * serialize_compressed * serialize_uncompressed * serialize_compressed_unchecked * serialize_uncompressed_unchecked ``` I think its fine to expose the four options here,...

I think we should also better separate out MSM benches in the benchmarks, the time delays are way too long for benching field arithmetic changes.

doubling it isn't great, because that introduces data dependencies into the core benchmark

The primary purpose is to make the speed of curve point subtraction equal to the speed of curve point addition. Currently they differ slightly by one field negation (https://github.com/arkworks-rs/curves/pull/41), however...