Async circuits, pt. 1 - async circuit runners

[mitschabaude]

Starts to address MinaFoundation/Core-Grants#4

This is the first step of supporting async circuits. It does not yet support async methods in contracts / zkprogram, but brings significant internal changes towards that support:

• Switch to a version of Pickles/Snarky which support async circuits under the hood. Internally, make the circuits that are passed to Pickles async functions
• Change low-level circuit runners -- constraintSystem() and runAndCheck() -- to be async, and support async callbacks in them. This is a necessary step because we run contract methods through these runners

bindings: o1-labs/o1js-bindings#248
mina: MinaProtocol/mina#15084
snarky: o1-labs/snarky#836

[mitschabaude]

wtf happened here??

[Trivo25]

wow :D

[mitschabaude]

it turns out that 298c4bd fixed a major bug - previously, reduce() thought there are no actions emitted in any method if the only method that emitted actions called reduce() itself

The new constraint count is the correct one 😅

[mitschabaude]

The last commit b47224f addressed the weird reduction by one constraint for many of our circuits

[Trivo25]

+1 also encountered them recently

[mitschabaude]

Just wanted to share that this massively improves our "witness generation" benchmark. The reason is that this witness generation step no longer maintains and updates a constraint system (it really doesn't need to).

I'm also expecting big improvements in uncached compilation since we run the circuit fewer times.

ecdsa - build constraint system

time: 3016.122ms ± 3.6%
change: -2.820% (p = 0.53 > 0.05)
Change within noise threshold.


ecdsa - witness generation

time: 885.910ms ± 3.0%
change: -57.055% (p = 0.00 < 0.05)
Performance has improved


ecdsa - compile (cached)

time: 8019.811ms ± 2.3%
change: -3.375% (p = 0.02 < 0.05)
Performance has improved