Circuit Optimisations Tasklist

[iAmMichaelConnor]

All conceivable optimisations for all circuits: aztec.nr, protocol circuits.

Please feel free to directly edit the tasklist, and if something needs a long description, please create a linked issue to describe the optimisation.

Feel free to move tasks between the two categories (quite often the work will overlap both)

See also https://hackmd.io/@aztec-network/ryPlh16-A?type=view#Circuits

cc. @LeilaWang @LHerskind @sirasistant @MirandaWood @benesjan et al.

Aztec.nr

Beta Give feedback

1. Reduce constraint count of a function call #7092
   C-aztec.nr T-optimisation +2
   
2. Function data hash could just pack the inputs into a field #3057
   C-protocol-circuits T-optimisation +2
3. Consider removing siloed viewing keys to remove derivation constraints and header encryption constraints. (Needs discussion around whether this feature is worth keeping)
4. Consider state var methods which don't create redundant new notes & nullifiers. (See old debates).
5. Consider refactoring Token's transfer function with recursion #7142
   +0
   
6. Reduce constraint count of args_hash #7272
   +0
   
7. feat(StateVariables): Extend SharedMutable to work with multi-field values #5491
   S-blocked +1

Protocol Circuits

Beta Give feedback

1. Hash logs with poseidon client-side, and then prove they've been 'blobbed' sequencer-side (blobbing is a separate issue). (Needs discussion on whether logs are blobbed or simply emitted on L2 or both. Until then, we could implement sequencer-side hashing with sha256?).
2. Fix poseidon2 so it's ~71 gates to hash two fields (as per bberg blackbox benchmarks)
3. Switch to the data bus
4. Introducing tree epochs would shrink tree heights, but there's lots to understand before deciding to do that.
5. Introduce multiple sizes of Kernel Circuit (this might impact Base & Merge rollup designs, and cause us to require multiple versions of those).
6. Consider whether the current tree structure for the "Parity Circuits" is optimal (and consider renaming to make it clear that they relate to L1->L2 messaging)
7. Find out: "wen honkish verify_proof() function?" for cheap recursion
8. Consider this proposal to remove key derivation requests https://hackmd.io/Y7eEMJ5_RVmfyrJv1qjEJQ
9. Verify pre-constructed output (instead of constructing it in the circuit).
10. Combine all types of side effects and check their counters are incremented by 1 (instead of using range checks).
11. Reduce redundant is_empty() checks.
12. Create a sensitivity testing tool: for a selection of changes to the circuit constants, measure the impact on constraint counts