Skip to content

Commit

Permalink
avx blake (#518)
Browse files Browse the repository at this point in the history
  • Loading branch information
@spapinistarkware
spapinistarkware authored Mar 25, 2024
1 parent 60d1432 commit 9dbe5bc
Show file tree
Hide file tree
Showing 6 changed files with 612 additions and 34 deletions.
217 changes: 217 additions & 0 deletions src/commitment_scheme/blake2s_ref.rs
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,217 @@
//! An AVX512 implementation of the BLAKE2s compression function.
//! Based on https://github.com/oconnor663/blake2_simd/blob/master/blake2s/src/avx2.rs .
pub const IV: [u32; 8] = [
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A, 0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19,
];

pub const SIGMA: [[u8; 16]; 10] = [
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
[14, 10, 4, 8, 9, 15, 13, 6, 1, 12, 0, 2, 11, 7, 5, 3],
[11, 8, 12, 0, 5, 2, 15, 13, 10, 14, 3, 6, 7, 1, 9, 4],
[7, 9, 3, 1, 13, 12, 11, 14, 2, 6, 5, 10, 4, 0, 15, 8],
[9, 0, 5, 7, 2, 4, 10, 15, 14, 1, 11, 12, 6, 8, 3, 13],
[2, 12, 6, 10, 0, 11, 8, 3, 4, 13, 7, 5, 15, 14, 1, 9],
[12, 5, 1, 15, 14, 13, 4, 10, 0, 7, 6, 3, 9, 2, 8, 11],
[13, 11, 7, 14, 12, 1, 3, 9, 5, 0, 15, 4, 8, 6, 2, 10],
[6, 15, 14, 9, 11, 3, 0, 8, 12, 2, 13, 7, 1, 4, 10, 5],
[10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13, 0],
];

#[inline(always)]
fn add(a: u32, b: u32) -> u32 {
a.wrapping_add(b)
}

#[inline(always)]
fn xor(a: u32, b: u32) -> u32 {
a ^ b
}

#[inline(always)]
fn rot16(x: u32) -> u32 {
(x >> 16) | (x << (32 - 16))
}

#[inline(always)]
fn rot12(x: u32) -> u32 {
(x >> 12) | (x << (32 - 12))
}

#[inline(always)]
fn rot8(x: u32) -> u32 {
(x >> 8) | (x << (32 - 8))
}

#[inline(always)]
fn rot7(x: u32) -> u32 {
(x >> 7) | (x << (32 - 7))
}

#[inline(always)]
fn round(v: &mut [u32; 16], m: [u32; 16], r: usize) {
v[0] = add(v[0], m[SIGMA[r][0] as usize]);
v[1] = add(v[1], m[SIGMA[r][2] as usize]);
v[2] = add(v[2], m[SIGMA[r][4] as usize]);
v[3] = add(v[3], m[SIGMA[r][6] as usize]);
v[0] = add(v[0], v[4]);
v[1] = add(v[1], v[5]);
v[2] = add(v[2], v[6]);
v[3] = add(v[3], v[7]);
v[12] = xor(v[12], v[0]);
v[13] = xor(v[13], v[1]);
v[14] = xor(v[14], v[2]);
v[15] = xor(v[15], v[3]);
v[12] = rot16(v[12]);
v[13] = rot16(v[13]);
v[14] = rot16(v[14]);
v[15] = rot16(v[15]);
v[8] = add(v[8], v[12]);
v[9] = add(v[9], v[13]);
v[10] = add(v[10], v[14]);
v[11] = add(v[11], v[15]);
v[4] = xor(v[4], v[8]);
v[5] = xor(v[5], v[9]);
v[6] = xor(v[6], v[10]);
v[7] = xor(v[7], v[11]);
v[4] = rot12(v[4]);
v[5] = rot12(v[5]);
v[6] = rot12(v[6]);
v[7] = rot12(v[7]);
v[0] = add(v[0], m[SIGMA[r][1] as usize]);
v[1] = add(v[1], m[SIGMA[r][3] as usize]);
v[2] = add(v[2], m[SIGMA[r][5] as usize]);
v[3] = add(v[3], m[SIGMA[r][7] as usize]);
v[0] = add(v[0], v[4]);
v[1] = add(v[1], v[5]);
v[2] = add(v[2], v[6]);
v[3] = add(v[3], v[7]);
v[12] = xor(v[12], v[0]);
v[13] = xor(v[13], v[1]);
v[14] = xor(v[14], v[2]);
v[15] = xor(v[15], v[3]);
v[12] = rot8(v[12]);
v[13] = rot8(v[13]);
v[14] = rot8(v[14]);
v[15] = rot8(v[15]);
v[8] = add(v[8], v[12]);
v[9] = add(v[9], v[13]);
v[10] = add(v[10], v[14]);
v[11] = add(v[11], v[15]);
v[4] = xor(v[4], v[8]);
v[5] = xor(v[5], v[9]);
v[6] = xor(v[6], v[10]);
v[7] = xor(v[7], v[11]);
v[4] = rot7(v[4]);
v[5] = rot7(v[5]);
v[6] = rot7(v[6]);
v[7] = rot7(v[7]);

v[0] = add(v[0], m[SIGMA[r][8] as usize]);
v[1] = add(v[1], m[SIGMA[r][10] as usize]);
v[2] = add(v[2], m[SIGMA[r][12] as usize]);
v[3] = add(v[3], m[SIGMA[r][14] as usize]);
v[0] = add(v[0], v[5]);
v[1] = add(v[1], v[6]);
v[2] = add(v[2], v[7]);
v[3] = add(v[3], v[4]);
v[15] = xor(v[15], v[0]);
v[12] = xor(v[12], v[1]);
v[13] = xor(v[13], v[2]);
v[14] = xor(v[14], v[3]);
v[15] = rot16(v[15]);
v[12] = rot16(v[12]);
v[13] = rot16(v[13]);
v[14] = rot16(v[14]);
v[10] = add(v[10], v[15]);
v[11] = add(v[11], v[12]);
v[8] = add(v[8], v[13]);
v[9] = add(v[9], v[14]);
v[5] = xor(v[5], v[10]);
v[6] = xor(v[6], v[11]);
v[7] = xor(v[7], v[8]);
v[4] = xor(v[4], v[9]);
v[5] = rot12(v[5]);
v[6] = rot12(v[6]);
v[7] = rot12(v[7]);
v[4] = rot12(v[4]);
v[0] = add(v[0], m[SIGMA[r][9] as usize]);
v[1] = add(v[1], m[SIGMA[r][11] as usize]);
v[2] = add(v[2], m[SIGMA[r][13] as usize]);
v[3] = add(v[3], m[SIGMA[r][15] as usize]);
v[0] = add(v[0], v[5]);
v[1] = add(v[1], v[6]);
v[2] = add(v[2], v[7]);
v[3] = add(v[3], v[4]);
v[15] = xor(v[15], v[0]);
v[12] = xor(v[12], v[1]);
v[13] = xor(v[13], v[2]);
v[14] = xor(v[14], v[3]);
v[15] = rot8(v[15]);
v[12] = rot8(v[12]);
v[13] = rot8(v[13]);
v[14] = rot8(v[14]);
v[10] = add(v[10], v[15]);
v[11] = add(v[11], v[12]);
v[8] = add(v[8], v[13]);
v[9] = add(v[9], v[14]);
v[5] = xor(v[5], v[10]);
v[6] = xor(v[6], v[11]);
v[7] = xor(v[7], v[8]);
v[4] = xor(v[4], v[9]);
v[5] = rot7(v[5]);
v[6] = rot7(v[6]);
v[7] = rot7(v[7]);
v[4] = rot7(v[4]);
}

/// Performs a Blake2s compression.
pub fn compress(
h_vecs: [u32; 8],
msg_vecs: [u32; 16],
count_low: u32,
count_high: u32,
lastblock: u32,
lastnode: u32,
) -> [u32; 8] {
let mut v = [
h_vecs[0],
h_vecs[1],
h_vecs[2],
h_vecs[3],
h_vecs[4],
h_vecs[5],
h_vecs[6],
h_vecs[7],
IV[0],
IV[1],
IV[2],
IV[3],
xor(IV[4], count_low),
xor(IV[5], count_high),
xor(IV[6], lastblock),
xor(IV[7], lastnode),
];

round(&mut v, msg_vecs, 0);
round(&mut v, msg_vecs, 1);
round(&mut v, msg_vecs, 2);
round(&mut v, msg_vecs, 3);
round(&mut v, msg_vecs, 4);
round(&mut v, msg_vecs, 5);
round(&mut v, msg_vecs, 6);
round(&mut v, msg_vecs, 7);
round(&mut v, msg_vecs, 8);
round(&mut v, msg_vecs, 9);

[
xor(xor(h_vecs[0], v[0]), v[8]),
xor(xor(h_vecs[1], v[1]), v[9]),
xor(xor(h_vecs[2], v[2]), v[10]),
xor(xor(h_vecs[3], v[3]), v[11]),
xor(xor(h_vecs[4], v[4]), v[12]),
xor(xor(h_vecs[5], v[5]), v[13]),
xor(xor(h_vecs[6], v[6]), v[14]),
xor(xor(h_vecs[7], v[7]), v[15]),
]
}
1 change: 1 addition & 0 deletions src/commitment_scheme/mod.rs
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
pub mod blake2_hash;
pub mod blake2s_ref;
pub mod blake3_hash;
pub mod hasher;
pub mod merkle_decommitment;
Expand Down
Loading

0 comments on commit 9dbe5bc

Please sign in to comment.