saymrwulf/curve25519-dalek-source
1
0
Fork 0
mirror of https://github.com/saymrwulf/curve25519-dalek-source.git synced 2026-09-04 20:24:10 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add comment to mul draft

Browse source
This commit is contained in:
Henry de Valence 2017-11-09 10:24:41 -08:00 committed by Henry de Valence
parent 4f6788c72d
commit e2a2b3a3b5
1 changed files with 12 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
src/avx2/field.rs
View file

@ -344,14 +344,23 @@ impl<'a, 'b> Mul<&'b FieldElement32x4> for &'a FieldElement32x4 {
unsafe { u32x8::from(_mm256_mul_epu32(x,y)) } unsafe { u32x8::from(_mm256_mul_epu32(x,y)) }
} }
let x19 = u32x8::new(19,0,19,0,19,0,19,0); let v19 = u32x8::new(19,0,19,0,19,0,19,0);
// XXX clean up this horrifying abomination
//
// The idea is to take the standard "schoolbook multiplication square" (see
// FieldElement32), and walk up each column from top to bottom, then left to right.
//
// Instead of multiplying by 19 in a precomputation, we overwrite the b[i] value with
// b[i]*19 as soon as we will no longer need it.
//
macro_rules! loop_body { macro_rules! loop_body {
($i:expr) => { ($i:expr) => {
let (ai, ai1) = unpack_pair(self.0[$i/2]); let (ai, ai1) = unpack_pair(self.0[$i/2]);
c[9] = c[9] + m(ai, b[(100 + 9-$i) % 10]); c[9] = c[9] + m(ai, b[(100 + 9-$i) % 10]);
b[(100 + 9-$i) % 10] = m_lo(b[(100 + 9-$i) % 10], x19); b[(100 + 9-$i) % 10] = m_lo(b[(100 + 9-$i) % 10], v19);
c[8] = c[8] + m(ai, b[(100 + 8-$i) % 10]); c[8] = c[8] + m(ai, b[(100 + 8-$i) % 10]);
c[7] = c[7] + m(ai, b[(100 + 7-$i) % 10]); c[7] = c[7] + m(ai, b[(100 + 7-$i) % 10]);
c[6] = c[6] + m(ai, b[(100 + 6-$i) % 10]); c[6] = c[6] + m(ai, b[(100 + 6-$i) % 10]);
@ -364,7 +373,7 @@ impl<'a, 'b> Mul<&'b FieldElement32x4> for &'a FieldElement32x4 {
let ai1_2 = ai1 + ai1; let ai1_2 = ai1 + ai1;
c[9] = c[9] + m(ai1, b[(100 + 9-($i+1)) % 10]); c[9] = c[9] + m(ai1, b[(100 + 9-($i+1)) % 10]);
b[(100 + 9-($i+1)) % 10] = m_lo(b[(100 + 9-($i+1)) % 10], x19); b[(100 + 9-($i+1)) % 10] = m_lo(b[(100 + 9-($i+1)) % 10], v19);
c[8] = c[8] + m(ai1_2, b[(100 + 8-($i+1)) % 10]); c[8] = c[8] + m(ai1_2, b[(100 + 8-($i+1)) % 10]);
c[7] = c[7] + m(ai1, b[(100 + 7-($i+1)) % 10]); c[7] = c[7] + m(ai1, b[(100 + 7-($i+1)) % 10]);
c[6] = c[6] + m(ai1_2, b[(100 + 6-($i+1)) % 10]); c[6] = c[6] + m(ai1_2, b[(100 + 6-($i+1)) % 10]);