miden-gpu 0.6.0

GPU acceleration for the Miden VM prover
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144

#![cfg(all(target_arch = "aarch64", target_os = "macos"))]

use std::time::Instant;

use miden_crypto::{Felt, FieldElement, Word, hash::rpo::Rpo256};
use miden_gpu::{
    HashFn,
    metal::{ColumnHasher, RowHasher, build_merkle_tree, utils::page_aligned_uninit_vector},
};

#[test]
fn gpu_rpo_from_columns() {
    let n = 8388608;
    let mut col0 = unsafe { page_aligned_uninit_vector(n) };
    let mut col1 = unsafe { page_aligned_uninit_vector(n) };
    let mut col2 = unsafe { page_aligned_uninit_vector(n) };
    let mut col3 = unsafe { page_aligned_uninit_vector(n) };
    let mut col4 = unsafe { page_aligned_uninit_vector(n) };
    let mut col5 = unsafe { page_aligned_uninit_vector(n) };
    let mut col6 = unsafe { page_aligned_uninit_vector(n) };
    let mut col7 = unsafe { page_aligned_uninit_vector(n) };
    col0.fill(Felt::ONE);
    col1.fill(Felt::ONE);
    col2.fill(Felt::ONE);
    col3.fill(Felt::ONE);
    col4.fill(Felt::ONE);
    col5.fill(Felt::ONE);
    col6.fill(Felt::ONE);
    col7.fill(Felt::ONE);
    let input_size = 8;

    let mut rpo = ColumnHasher::new(n, input_size, HashFn::Rpo256);

    let now = Instant::now();
    for i in 0..input_size {
        let col = match i % 8 {
            0 => &col0,
            1 => &col1,
            2 => &col2,
            3 => &col3,
            4 => &col4,
            5 => &col5,
            6 => &col6,
            7 => &col7,
            _ => unreachable!(),
        };

        rpo.update(col);
    }
    println!("Encode in {:?}", now.elapsed());

    let now = Instant::now();
    let hashes = pollster::block_on(rpo.finish());
    println!("Run in {:?}", now.elapsed());
    println!("Hashes: {:?}", hashes[0]);
    println!("Hashes1: {:?}", hashes[1]);
    println!("Hashes2: {:?}", hashes[1024]);

    let now = Instant::now();
    let merkle_tree = pollster::block_on(build_merkle_tree(&hashes, HashFn::Rpo256));
    println!("Root: {:?}", merkle_tree[0]);
    println!("Root1: {:?}", merkle_tree[1]);
    println!("Root1: {:?}", merkle_tree[2]);
    println!("Root1: {:?}", merkle_tree[3]);
    println!("Root1: {:?}", merkle_tree[merkle_tree.len() / 2 + 1]);
    println!("Merkle tree in {:?}", now.elapsed());
}

#[test]
fn gpu_rpo_from_rows() {
    let n = 8388608;
    let mut rows: Vec<[Felt; 8]> = unsafe { page_aligned_uninit_vector(n) };
    rows.fill([Felt::ONE; 8]);
    let mut rpo = RowHasher::new(n, 8, HashFn::Rpo256);

    let now = Instant::now();
    rpo.update(&rows);
    println!("Encode in {:?}", now.elapsed());

    let now = Instant::now();
    let hashes = pollster::block_on(rpo.finish());
    println!("Run in {:?}", now.elapsed());
    println!("Hashes (row): {:?}", hashes[0]);
    println!("Hashes1 (row): {:?}", hashes[1]);
    println!("Hashes2 (row): {:?}", hashes[1024]);
    hashes
        .iter()
        .zip(&hashes[1..])
        .enumerate()
        .for_each(|(i, (a, b))| assert_eq!(a, b, "mismatch at {i}"));

    let now = Instant::now();
    let merkle_tree = pollster::block_on(build_merkle_tree(&hashes, HashFn::Rpo256));
    println!("Root (row): {:?}", merkle_tree[0]);
    println!("Root1 (row): {:?}", merkle_tree[1]);
    println!("Root1 (row): {:?}", merkle_tree[2]);
    println!("Root1 (row): {:?}", merkle_tree[3]);
    println!("Root1 (row): {:?}", merkle_tree[merkle_tree.len() / 2 + 1]);
    println!("Merkle tree in {:?}", now.elapsed());
}

#[test]
fn compare_cpu_gpu_rpo() {
    let n = 32768;
    let mut rows: Vec<[Felt; 8]> = unsafe { page_aligned_uninit_vector(n) };
    rows.fill([Felt::ONE; 8]);
    let mut rpo = RowHasher::new(n, 8, HashFn::Rpo256);
    rpo.update(&rows);

    let hashes = pollster::block_on(rpo.finish());

    let cpu_elements = std::iter::repeat_n(Felt::ONE, 8).collect::<Vec<Felt>>();

    // hash using cpu implementation
    let cpu_rpo_digest = Rpo256::hash_elements(&cpu_elements);

    let cpu_fields: &[Felt] = cpu_rpo_digest.as_elements();

    let gpu_fields = hashes[0];

    gpu_fields.iter().zip(cpu_fields).for_each(|(gpu_field, cpu_field)| {
        assert_eq!(gpu_field.to_string(), cpu_field.to_string());
    });

    assert_eq!(hashes[0], hashes[1]);
    assert_eq!(hashes[1], hashes[2]);

    // Generate a Merkle tree using gpu
    let gpu_merkle_tree = pollster::block_on(build_merkle_tree(&hashes, HashFn::Rpo256));

    // Create the same input as for the gpu
    let cpu_merkle_digests = std::iter::repeat_n(cpu_rpo_digest, n).collect::<Vec<Word>>();
    // Generate a Merkle tree using cpu
    let cpu_merkle_tree = winter_crypto::MerkleTree::<Rpo256>::new(cpu_merkle_digests)
        .expect("failed to construct trace Merkle tree");

    // Check if roots are equal
    // The 0th element is always 0, 1st element is the root, 2 & 3 elements are childs of root, etc.
    let gpu_root = &gpu_merkle_tree[1];
    let cpu_root = cpu_merkle_tree.root().as_elements();
    gpu_root.iter().zip(cpu_root.iter()).for_each(|(gr, cr)| {
        assert_eq!(gr, cr);
    })
}