Bigger hard-coded circuits

[mitschabaude]

Hard coded circuit size increased by 16x, from $2^{16}$ to $2^{20}$

• This is not for merging, just for experiments with bigger circuits
• The bigger circuit size won't work for smart contracts, just with the Circuit API which represents plain kimchi circuits without recursion stuff

Mina branch: MinaProtocol/mina#12007

[zhfnjust]

still can not compile the recursion sha256.

error:

https://gist.github.com/zhfnjust/4996fc397b38562eb230c73a059e4a3f

[mitschabaude]

Yes as I wrote above, the bigger circuit size will only work with the Circuit API. Increasing sizes of ZkProgram would need a bit more digging. But I don't see the point of doing that because it just will run into the same OOM error.

Maybe the issue with the recursion SHA256 is that, since it includes a verifier circuit in addition to the SHA256 circuit, it is probably too big already with 1 SHA256 invocation, so needs to be split up further

[zhfnjust]

how to split up SHA256 circuit?

[mitschabaude]

how to split up SHA256 circuit?

you probably know the details better than me.. a first idea: looking at your code, it seems that both the g function and the compression function do their main work in a loop of 64 steps. So it seems plausible that this work can be split up into smaller portions of those loop, where an unfinished proof would maintain some current state as its public input

[mitschabaude]

unused

[xhliu]

We still need this.

[mitschabaude]

We still need this

Btw @xhliu we have a sha256 in the works that can hash ~10 blocks per circuit
#1285

[xhliu]

When can it be used? @mitschabaude