Batch ring proof

[swasilyev]

ATM twice faster for the batch of 10

Start: Verify
··End: Verify ..................................................................16.983ms
Start: Verify Batch KZG
··End: Verify Batch KZG ........................................................8.437ms

[swasilyev]

For N = 100

Start: Verify
··End: Verify ..................................................................175.106ms
Start: Verify Batch KZG
··End: Verify Batch KZG ........................................................22.411ms

[swasilyev]

TODO:

1. bench batch_size = 1
2. optimizations:

• merge muls by instance columns
• merge muls by nus with the linearization poly muls

3. fix final randomness
4. add instance to FS
5. support different instance columns (rings)
6. support different iops
7. support different domains
8. apply optimizations to the abstract pcs case
9. optimize single verification for kzg
10. some comments

[swasilyev]

Single proof (batch_size=1) verification is now also faster (due to the broken abstraction bounds)

Start: Verify
··End: Verify ..................................................................1.725ms
Start: Verify Batch KZG
··End: Verify Batch KZG ........................................................1.376ms

[gui1117]

so this is a for a single ring isn't it? it doesn't batch for different rings. Could batching for multiple rings also be improved (I mean in general, not necessarily in this PR).

could we also benchmark for 1000 and 10000 just to see the limit of improvement?

[swasilyev]

so this is a for a single ring isn't it? it doesn't batch for different rings. Could batching for multiple rings also be improved (I mean in general, not necessarily in this PR).

This is todo item #5.

could we also benchmark for 1000 and 10000 just to see the limit of improvement?

Kinda. I'll need to generate 1k different proofs (from scratch, as plonk doesn't support proof randomization) for that, that takes some time.

[davxy]

Requires g1 to be public

[davxy]

Needs some changes to KZG in fflonk (private fields and a couple of missing Clone bounds)

[davxy]

NOTE: I accidentally pushed here, but I have reverted to your changes. That is why you see a force push.

[davxy]

@swasilyev BatchVerifier struct : #66

ark-vrf integration draft: davxy/ark-vrf#67