Benchmark ark-works for circom with wasm-bindgen

[vivianjeng]

Problem

We have explored this before: #202 (comment)
But looking carefully into the waku implementation, it doesn't enable rayon in their wasm-bindgen
(and the ark-groth16 has the feature parallel to enable rayon)
with current mopro implementation, we support wasm-bindgen-rayon
so we can try to benchmark circom groth16 prover for circuits

The result with parallel and without

keccak256 (debug mode)	with parallel	without parallel
time	10.98826125s	27.5738135s

Details

Try to use mopro build for web (by default it supports rayon)
and record the time it executes when generating a groth16 proof
(NOTE: without witness generation time, I think it is not possible to generate wasm bindings with current rust witness generator)
You can also refer to waku implementation: https://github.com/vacp2p/zerokit/blob/master/rln-wasm/src/lib.rs

Or you can load the wtns, zkey as bytes before generate the function
e.g.

const WTNS_BYTES: &[u8] = include_bytes!("./keccak256.wtns");
const ZKEY_BYTES: &[u8] = include_bytes!("./keccak256.zkey");

#[wasm_bindgen]
pub fn arkworks_prover() -> Result<JsValue, JsValue> {
    prove(...)
}

Acceptance criteria

Benchmark the execution time for the following circuits in the browser

	ark-works with rayon	snarkjs
sha256	- ms	- ms
keccak256	- ms	- ms
RSA	- ms	- ms
semaphore-32	- ms	- ms
Anon Aadhaar	- ms	- ms

Please check the circuit input, wasm, zkey here
https://github.com/zkmopro/android-benchmark-app/tree/main/ios/MoproApp

Next steps (optional)

Update documentation for https://zkmopro.org/docs/performance/

[sifnoc]

Successfully compiled ark-groth16 with wasm-pack in here: https://github.com/sifnoc/ark-groth16-wasm
However, I encountered a compilation error when attempting to use rust-witness in mopro-ffi. To proceed, I plan to use a pre-generated witness and zkey.

Could you provide guidance on obtaining the wtns file mentioned in this issue? Should I explore rust_witness for generating it, or is there another recommended approach?

[vivianjeng]

I think you can save the witness output to a file
and read the witness from the file (in wasm environment)
similar to the usage of srs in halo2 circuit
e.g. (from chatGPT)

use num_bigint::BigUint;
use std::fs::File;
use std::io::{self, Read, Write};

fn save_biguint_vec_to_file(vec: &[BigUint], file_path: &str) -> io::Result<()> {
    let mut file = File::create(file_path)?;
    for biguint in vec {
        let bytes = biguint.to_bytes_be();
        let length = bytes.len() as u64;
        // Write the length of the bytes
        file.write_all(&length.to_be_bytes())?;
        // Write the actual bytes
        file.write_all(&bytes)?;
    }
    Ok(())
}

fn load_biguint_vec_from_file(file_path: &str) -> io::Result<Vec<BigUint>> {
    let mut file = File::open(file_path)?;
    let mut buf = Vec::new();
    file.read_to_end(&mut buf)?;

    let mut biguint_vec = Vec::new();
    let mut cursor = 0;

    while cursor < buf.len() {
        // Read the length of the bytes (u64 is 8 bytes)
        let mut length_bytes = [0u8; 8];
        length_bytes.copy_from_slice(&buf[cursor..cursor + 8]);
        cursor += 8;
        let length = u64::from_be_bytes(length_bytes) as usize;

        // Read the actual bytes
        let bytes = &buf[cursor..cursor + length];
        cursor += length;

        // Convert bytes back to BigUint
        biguint_vec.push(BigUint::from_bytes_be(bytes));
    }

    Ok(biguint_vec)
}

and wtns is a file generated from snarkjs (you don't need to use it. Just an example)
https://github.com/iden3/snarkjs?tab=readme-ov-file#14-calculate-the-witness
To generate a wtns is by executing

snarkjs wtns calculate circuit_js/circuit.wasm input.json witness.wtns

The disadvantage of the wtns file is that the type might be different from the inputs in ark-works

[vivianjeng]

You can refer to how vacp2p does
https://github.com/vacp2p/zerokit/blob/00f8d039a8a51fb729381da8a46977983cb117da/rln/src/protocol.rs#L539
and the rln-wasm
https://github.com/vacp2p/zerokit/blob/00f8d039a8a51fb729381da8a46977983cb117da/rln-wasm/README.md