# Mopro progress reports

Document for progress reports on Mopro (Mobile Prover). Aiming to do check-ins ~every 2-4 weeks.

(There's also a grant document to cross-check with but link is private).

## Feb 1 - Apr 15

Done (abbreviated)
- Improve DevEx significantly with mopro-cli as main interface
- Demo https://www.loom.com/share/6ff382b0497c47aea9d0ef8b6e790dd8
- Finish up MoPerf: https://hackmd.io/5ItB2D50QcavF18cWIrmfQ
- Understand limits of client-side proving (4gb phone ~= 2m constraints np)
- Derisked enough for resuming AAMO project
- AAMO collab: https://hackmd.io/321WjevGRBWMSPYe6g7mcw
- Also with Proof of Passport
- Generally good feedback loops, RN support etc
- Found unexpected blockers with witness gen and app store
- Witness generation notes
- See https://github.com/oskarth/mopro/issues/109 for context
- Unable to get current dylib setup working for App Store upload
- Should be possible, we just couldn't figure it out
- Lead some teams to look for alt approach
- Found alt slower approach that works (wasm3)
- Non-trivial to integrate witnesscalc (C++) with current stack
- Native witness gen in Rust turns out to only work for some circuits
- Promising in terms of perf (much faster than witnesscalc)
- But non-trivial to make complete => more work needed
- Talks and workshops
- ETHTaipei (Taipei)
- Slides https://hackmd.io/@oskarth/S1yGjF8C6#/
- ZK Day workshop (Taipei)
- ZK Summit (Athens)

## Dec 1 - Jan 31

- Done
- Improve API and DevEx
- Simplify config with TOML file; start of CLI interface
- ark-zkey better integrated
- Misc bug fixes and clean up
- Outreach and talking to users
- Term Structure and ZK Identity talk at Taipei Blockchain Week
- Talk to teams to understand needs (ProofOfPassport, Myna, AA, etc)
- Support GPU grantee work
- Initial Android support and generate calldata (Vivian)
- Shape up Anon-Aadhaar ZK-ID Showcase App (AAMA)
- See https://hackmd.io/321WjevGRBWMSPYe6g7mcw?both=
- After circuit increased from ~160k to 1700k constraints technical risk went up
- We decided to pause this and derisk perf bottlenecks first before resuming
- Start MoPerf project collab with AA to find limits of client side proving on mobile
- See https://hackmd.io/5ItB2D50QcavF18cWIrmfQ?view=
- Understand unknowns and derisk these (see [GH issues](https://github.com/oskarth/mopro/issues/57) for more details)
- Good news: Able to run latest AA circuits on new iPhone and clear direction for most unknowns
- Next
- Finish up MoPerf project (clarity, perf, write-up)
- Resume Anon-Aadhaar showcase app (see doc above re scope)
- FYI
- This is for Dec-Jan; took some time off in middle
- Oskar off Feb 5 - 12; Vivian Feb 14 - 20
- Some lack of clarity re grant agreement

## Nov 6 - Nov 30

(3-4 weeks)

- Done
- ZKHack: RSA Verify and Anon-aadhaar on mobile together with Yanis from Anon-aadhaar
- [devfolio pres link](https://devfolio.co/projects/two-and-a-half-zk-pocketaadhaar-and-pocketkimchi-a8a6) ~6.5s (~5s witness gen, ~2s prover time); x10 faster vs browser on phone
- PoC of new proof system; (Kimchi, Plonkish) to show ease of integrating new proof systems
- RSA verify (zk email circuits) benchmark
- arkzkey: zkey bottleneck; sped up by x10 with custom arkzkey format to ~10s; also 50% file size
- ProgCrypto presentation: [slides](https://docs.google.com/presentation/d/1afIEgm8oYRvteWxUd04CcMOxChAiHaD55d5AKd0RkvY/edit#slide=id.g29a7b1abf9b_1_543)
- Planning at Devconnect: https://hackmd.io/@oskarth/mopro-devconnect
- Consider milestone 3 done (will turn talk into write-up too)
- Next
- Improve API and DevEx
- More robust API
- mopro CLI interface
- Docs and tutorials
- General clean up
- Improve performance: Address bottlenecks mentioned in presentation
- Integrate arkzkey and try to improve further
- Experiment with native witness generation
- Android support
- Write-up
- FYI
- Quite a lot in next; milestone 4 cleanup before EOY
- Oskar will take a bit of time off during holidays then continue in Jan
- After milestone 4 inital grant finished, but we decided to continue grant, just less rigid milestones

## Oct 9 - Nov 6

(4 weeks)

- Done
- Got it working on real device
- Lots of debugging related to witness generator wasmer memory restrictions problem; patched wasmer page size bug
- Lots of experimentation with dylib to pre-load resources, etc
- On modern iPhone (14 Pro Max), Keccak proof generation takes ~1.5s (~x20 vs comparable browser? see last note)
- DevEx work to make it easier to use and contribute to
- Onboarding second developer (Vivian), who helped with a bunch of DevExp work
- Improve scripts, tests, configs, CI, separating out examples, etc
- Generally making it easier for someone to run and use
- Library maturity: zkey experimentation
- Moving from doing in-line setup and .r1cs file to using zkey
- Works but (i) Too slow for big circuits (ii) Some API changes needed
- Experimentation with improving zkey loading/serializing, but not performant enough yet for e.g. Keccak
- Wrote up several issues related to future work
- GPU plan of attack: https://github.com/oskarth/mopro/issues/22 + subissues re micro benchmarking, etc
- Many others here: https://github.com/oskarth/mopro/issues - related to different platforms, proof systems, setup/zkey, DevEx, RSA example, etc
- Next
- Disseminate work done so far, give talk at ProgCrypto and involve others
- Fix zkey loading performance (proving key loading slow) and adjust API
- RSA example
- Blockers/FYI
- Blocker: Zkey loading too slow for Keccak256, >1m
- Some experimental work [here](https://github.com/oskarth/mopro/pull/26)
- FYI: Adjusting milestone 2 scope and considering it done
- Working on real device took longer than expected
- Less GPU focus for now (clear next steps), esp with x20 perf boost; revisit later
- Decided to focus more on DevEx to get tighter feedback loop and early usage

## Sep 18 - Oct 9

(3 weeks)

- Done
- Milestone 1 and PoC finished
- Went from doing everything in Rust to exposing API for inputs/setup/prove/verify
- Using a more involved circuit (Keccak256)
- Open sourced repo [here](https://github.com/oskarth/mopro) (not publicly announced just yet)
- Example app with basic Keccak proof and preliminary benchmark
- iOS simulator on M1 MBP Keccak256 (150k constraints) ~1.2-1.3s
- Comparable: ecdsa_verify 160k constraints ([link](https://github.com/personaelabs/efficient-zk-ecdsa#browser-proving)) in-browser ~35-40s
- Preliminary results, need to run on real device and sanity check (see REQ below)
- Demo of recent progress available [here](https://www.loom.com/share/1f36229c66534e16b5cf58abb5064add)
- Next
- Continue with benchmarking
- Run app on real device
- Do basic profiling (empirically identify bottlenecks)
- Find comparable and reliable benchmarks to communicate perf diff with confidence
- Improve API/UX for developers
- Lots that can be done here, need to identify most low-hanging fruit
- Depends on real-world demand (see REQ)
- Continue GPU usage trial
- Experiment with iOS GPU as a spike (likely requires quite a lot of new learning)
- Experiment with Arkworks GPU usage on laptop
- Revisit prior note and make plan of attack for incorporating recent advancements with GPU MSMs etc
- Blockers/FYI/Requests
- FYI: A lot of work spent on getting Rust/Swift to play nicely w.r.t. data types, serialization, error handling etc. This was identified as a risk in project scope, so good to get this done early.
- REQ1: It would be useful to have a simple standard and up to date benchmark to compare with for in-browser proving, e.g. with ECDSA/RSA/Keccak. I can set this up myself but if there are any good existing canonical ones it'd be better to use that, as opposed to doing a side quest for messaging around with JS/Web ZK env.
- REQ2: Prioritization regarding e.g. GPU vs API DevEx; if we have x20 perf improvement already it might be worth making it easier for Zupass/Aadhaard and similar projects to use at Devconnect.
- REQ3: Contact for Circom/Polygon person also looking into mobile proving; any other person you think I should talk to, especially early users.
- REQ4: I think it'd be useful to have mopro as a hacking target during Devconnect, either in terms of apps using it or improving tool. Any ideas for how to make this happen would be appreciated!

## Sep 4 - Sep 16

- Done
- Library architecture illustration of how things fit together, see drawing
- Setup separate components/libs for `mopro-{core, ffi, ios}` in [mopro repo](https://github.com/oskarth/mopro)
- Get Arkworks to work on iOS simulator (though still with some errors) with CocoaPod lib and UniFFI Swift bindings
- Evaluate easiest proof system/tooling for e2e, decide on circom-compat to start with but keeping it modular
- Next
- Expand PoC to make basic API (not just example but separate setup/prove/verify bindings), deal with errors etc; make tiny example app with buttons; use non-trivial circuit
- Start to scope GPU work
- Blockers/FYIs
- No real blockers
- FYI1: Keeping repo private for now until e2e works nicely, will open source soon but can add people if interest
- FYI2: Toolchain/scripting needs work in terms of updating static libs in iOS, working on real device and working nicely for diffeent architectures/environments (aarch64-apple-ios{,-sim}, x86_64-apple-ios different libs libs) - leaving this for now to get e2e nice

### Artifacts

Code: [mopro repo](https://github.com/oskarth/mopro)

#### Architecture

The following illustration shows how mopro and its components fit together into the wider ZKP ecosystem:


Zooming in a bit:


## Aug 21 - Sep 1

- Done
- Scope out initial work, write grant, milestones
- Initial research to understand options better (performance problems, current way, tooling, mobile bindings)
- Do PoC spike `rust-swift-spike` to get Rust working on iOS with UniFFI bindings
- Next
- Draw out library architecture and make illustration of how things fit together
- Take spike and split into components; integrate Arkworks for e2e proving on mobile
- Blockers/FYIs
- No real blockers
- FYI: A lot of fiddling with Xcode/iOS toolchain, refamiliarizing myself with ecosystem, opaque configs, different architectures for embedding libraries, linking issues etc - on the flip side this is good because shows mopro can solve a clear pain point