logos-storage/plonky2
1
0
Fork 0
mirror of https://github.com/logos-storage/plonky2.git synced 2026-01-10 17:53:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

AVX2 vectorization of Poseidon S-box (#244)

Browse Source 
* AVX2 vectorization of Poseidon S-box

* Minor doc

* Microoptimization
This commit is contained in:
Jakub Nabaglo 2021-09-14 19:26:31 -07:00 committed by GitHub
parent 2ae9e34993
commit b411a275f9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 37 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/hash/poseidon.rs
View File

@ -519,6 +519,12 @@ impl Poseidon<8> for CrandallField {
fn mds_layer(state_: &[CrandallField; 8]) -> [CrandallField; 8] {
crate::hash::poseidon_neon::crandall_poseidon8_mds_neon(*state_)
}
#[cfg(target_feature="avx2")]
#[inline(always)]
fn sbox_layer(state: &mut [Self; 8]) {
crate::hash::poseidon_avx2::crandall_poseidon_sbox_avx2::<2>(state);
}
}
#[rustfmt::skip]
@ -744,6 +750,12 @@ impl Poseidon<12> for CrandallField {
fn mds_layer(state_: &[CrandallField; 12]) -> [CrandallField; 12] {
crate::hash::poseidon_neon::crandall_poseidon12_mds_neon(*state_)
}
#[cfg(target_feature="avx2")]
#[inline(always)]
fn sbox_layer(state: &mut [Self; 12]) {
crate::hash::poseidon_avx2::crandall_poseidon_sbox_avx2::<3>(state);
}
}
#[cfg(test)]

25
src/hash/poseidon_avx2.rs
View File

@ -220,3 +220,28 @@ pub unsafe fn crandall_poseidon_const_avx2<const PACKED_WIDTH: usize>(
packed_state[i] = packed_state[i].add_canonical_u64(packed_round_constants[i]);
}
}
#[inline(always)]
pub fn crandall_poseidon_sbox_avx2<const PACKED_WIDTH: usize>(
state: &mut [CrandallField; 4 * PACKED_WIDTH],
) {
// This function is manually interleaved to maximize instruction-level parallelism.
let packed_state = PackedCrandallAVX2::pack_slice_mut(state);
let mut x2 = [PackedCrandallAVX2::zero(); PACKED_WIDTH];
for i in 0..PACKED_WIDTH {
x2[i] = packed_state[i].square();
}
let mut x3 = [PackedCrandallAVX2::zero(); PACKED_WIDTH];
let mut x4 = [PackedCrandallAVX2::zero(); PACKED_WIDTH];
for i in 0..PACKED_WIDTH {
x3[i] = packed_state[i] * x2[i];
x4[i] = x2[i].square();
}
for i in 0..PACKED_WIDTH {
packed_state[i] = x3[i] * x4[i];
}
}