saymrwulf/risc0-curve25519-dalek-source
1
0
Fork 0
mirror of https://github.com/saymrwulf/risc0-curve25519-dalek-source.git synced 2026-09-04 20:03:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Replace some binary constants with named constants

Browse source
This commit is contained in:
Henry de Valence 2018-01-03 12:34:25 -08:00
parent dd80421094
commit d41dbb1fe4
2 changed files with 22 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
src/backend/avx2/edwards.rs
View file

@ -29,6 +29,7 @@ use traits::Identity;
use backend::avx2::field::FieldElement32x4;
use backend::avx2::field::{A_LANES, B_LANES, C_LANES, D_LANES, ALL_LANES};
use backend::avx2::field::D_LANES64;
use backend::avx2;
@ -155,6 +156,12 @@ impl ExtendedPoint {
let mut t0 = FieldElement32x4::zero();
let mut t1 = FieldElement32x4::zero();
// Want to compute (X1 Y1 Z1 X1+Y1).
// Not sure how to do this less expensively than computing
// (X1 Y1 Z1 T1) --(256bit shuffle)--> (X1 Y1 X1 Y1)
// (X1 Y1 X1 Y1) --(2x128b shuffle)--> (Y1 X1 Y1 X1)
// and then adding.
// Set t0 = (X1 Y1 X1 Y1)
t0.0[0] = _mm256_permute2x128_si256(P.0[0].into(), P.0[0].into(), 0b0000_0000).into();
t0.0[1] = _mm256_permute2x128_si256(P.0[1].into(), P.0[1].into(), 0b0000_0000).into();
@ -177,13 +184,15 @@ impl ExtendedPoint {
t0.0[4] = t0.0[4] + t1.0[4];
// Set t0 = (X1 Y1 Z1 X1+Y1)
t0.0[0] = _mm256_blend_epi32(t0.0[0].into(), P.0[0].into(), 0b01011111).into();
t0.0[1] = _mm256_blend_epi32(t0.0[1].into(), P.0[1].into(), 0b01011111).into();
t0.0[2] = _mm256_blend_epi32(t0.0[2].into(), P.0[2].into(), 0b01011111).into();
t0.0[3] = _mm256_blend_epi32(t0.0[3].into(), P.0[3].into(), 0b01011111).into();
t0.0[4] = _mm256_blend_epi32(t0.0[4].into(), P.0[4].into(), 0b01011111).into();
// why does this intrinsic take an i32 for the imm8 ???
t0.0[0] = _mm256_blend_epi32(P.0[0].into(), t0.0[0].into(), D_LANES as i32).into();
t0.0[1] = _mm256_blend_epi32(P.0[1].into(), t0.0[1].into(), D_LANES as i32).into();
t0.0[2] = _mm256_blend_epi32(P.0[2].into(), t0.0[2].into(), D_LANES as i32).into();
t0.0[3] = _mm256_blend_epi32(P.0[3].into(), t0.0[3].into(), D_LANES as i32).into();
t0.0[4] = _mm256_blend_epi32(P.0[4].into(), t0.0[4].into(), D_LANES as i32).into();
t1 = t0.square(0b11_00_00_00);
// Set t1 = t0^2, negating the D values
t1 = t0.square(D_LANES64);
// Now t1 = (S1 S2 S3 -S4)
@ -309,7 +318,8 @@ impl<'a, 'b> Add<&'b ExtendedPoint> for &'a ExtendedPoint {
// set t1 = (S0 S1 Z1 T1)
// set t0 = (S2 S3 Z2 T2)
for i in 0..5 {
t1.0[i] = _mm256_blend_epi32(t0.0[i].into(), P.0[i].into(), 0b11110000).into();
// why does this intrinsic take an i32 for the imm8 ???
t1.0[i] = _mm256_blend_epi32(t0.0[i].into(), P.0[i].into(), (C_LANES | D_LANES) as i32).into();
t0.0[i] = _mm256_permute2x128_si256(t0.0[i].into(), Q.0[i].into(), 49).into();
}

5
src/backend/avx2/field.rs
View file

@ -18,6 +18,11 @@ pub const B_LANES: u8 = 0b0000_1010;
pub const C_LANES: u8 = 0b0101_0000;
pub const D_LANES: u8 = 0b1010_0000;
pub const A_LANES64: u8 = 0b00_00_00_11;
pub const B_LANES64: u8 = 0b00_00_11_00;
pub const C_LANES64: u8 = 0b00_11_00_00;
pub const D_LANES64: u8 = 0b11_00_00_00;
pub const ALL_LANES: u8 = A_LANES | B_LANES | C_LANES | D_LANES;
use std::ops::Mul;