feat(icicle): Add icicle MSM support

[sagar-a16z]

• Rebased feat(icicle): Add icicle msm support #322
• Updated to latest icicle release
• Fixed it to use the GPU correctly
• Fall back to JOLT's CPU implementations when GPU is unavailable - much faster than the default CPU impl.
• Added msm benchmarks
• Overall Perf is up ~20%.

[wiz-a16z]

Wiz Scan Summary

Scanner	Findings
Vulnerabilities	9
Sensitive Data
Secrets
IaC Misconfigurations
Total	9

View scan details in Wiz

To detect these findings earlier in the dev lifecycle, try using Wiz Code VS Code Extension.

[sagar-a16z]

With the newest benchmark I added, the GPU perf is actually really good. (This is a WSL machine on a 5800X3D + 3080TI)

CPU

VariableBaseMSM::msm() [mode:JOLT CPU]
                        time:   [807.29 ms 810.86 ms 814.58 ms]

GPU

VariableBaseMSM::msm() [mode:Icicle]
                        time:   [153.71 ms 156.67 ms 159.43 ms]

It’s WAY faster. For some reason it's much slower when running the examples. Maybe because my CPU usage was low while this benchmark ran where as the CPU is pinned when running the examples or maybe because I have plenty of left over memory.

[moodlezoup]

great work! main question is: I think we need to move the icicle init into prove; otherwise if someone is e.g. loading the preprocessing from disk and running prove by itself, icicle won't be initialized. wdyt?

[moodlezoup]

Suggested change

	#[cfg(feature = "icicle")]
	let gpu_bases = Some(
	bases
	.par_iter()
	.map(|base| G1Projective::from_ark_affine(base))
	.collect(),
	);
	let max_num_bits = scalars
	.par_iter()
	.map(|s| s.clone().into_bigint().num_bits())
	.max()
	.unwrap();
	println!("Using max num bits: {}", max_num_bits);
	#[cfg(not(feature = "icicle"))]
	let gpu_bases = None;
	#[cfg(feature = "icicle")]
	let gpu_bases = Some(
	bases
	.par_iter()
	.map(|base| G1Projective::from_ark_affine(base))
	.collect(),
	);
	#[cfg(not(feature = "icicle"))]
	let gpu_bases = None;
	let max_num_bits = scalars
	.par_iter()
	.map(|s| s.clone().into_bigint().num_bits())
	.max()
	.unwrap();
	println!("Using max num bits: {}", max_num_bits);

[moodlezoup]

you should be able to just do

Suggested change

	let values: [u64; 4] = [
	rng.next_u64(),
	rng.next_u64(),
	rng.next_u64(),
	rng.next_u64(),
	];
	let bigint = ark_ff::BigInteger256::new(values);
	<Fr as JoltField>::from_bytes(&bigint.to_bytes_le())
	Fr::random(&mut rng)

[moodlezoup]

Suggested change

	let values: [u64; 4] = [
	rng.next_u64(),
	rng.next_u64(),
	rng.next_u64(),
	rng.next_u64(),
	];
	let bigint = ark_ff::BigInteger256::new(values);
	<Fr as JoltField>::from_bytes(&bigint.to_bytes_le())
	Fr::random(&mut rng)

[moodlezoup]

Should this be moved to prove?