dev::cost: Estimate prover work for a given circuit
#753
Conversation
therealyingtong
changed the title
dev::cost: Estimate prover work for a given circuit
| // Number of FFTs in the 2^k domain | ||
| ffts: usize, | ||
| // Number of FFTs in the 2^extended_k domain | ||
| extended_ffts: usize, | ||
| // Number of multi-scalar multiplications of length 2^k | ||
| msms: usize, |
There was a problem hiding this comment.
We should also account for the cost of the transcript operations, which can be sizeable.
I don't know if the above is sufficiently granular to get a useable estimate of prover costs. OTOH, it may be sufficiently high-level that people can reliably just plug in their own estimates for the cost of each of these things, and get a reasonable result.
Your PR also reminded me that I started implementing verifier time cost estimation almost a year ago, but haven't had time to finish it. I was trying to base my estimates on the number of additions and multiplications, but deriving equations for them was rather laborious. I've opened #754 with my draft changes for comparison.
We should test both approaches against several different circuits to check how far off their estimates are.
therealyingtong
force-pushed
the
prover-work-cost
branch
from
17824fc
to
d09f2b9
Compare
Closes #749. Computes prover cost in terms of the number of:
2^kdomain,2^extended_kdomain, and2^k.What other prover operations are significant? (e.g. polynomial evaluations)