diff --git a/src/backend/serial/u32/field.rs b/src/backend/serial/u32/field.rs index 4b06aae..86be81f 100644 --- a/src/backend/serial/u32/field.rs +++ b/src/backend/serial/u32/field.rs @@ -126,10 +126,13 @@ impl<'a, 'b> Mul<&'b FieldElement2625> for &'a FieldElement2625 { /// Helper function to multiply two 32-bit integers with 64 bits /// of output. #[inline(always)] - fn m(x: u32, y: u32) -> u64 { (x as u64) * (y as u64) } + fn m(x: u32, y: u32) -> u64 { + (x as u64) * (y as u64) + } // Alias self, _rhs for more readable formulas - let x: &[u32;10] = &self.0; let y: &[u32;10] = &_rhs.0; + let x: &[u32; 10] = &self.0; + let y: &[u32; 10] = &_rhs.0; // We assume that the input limbs x[i], y[i] are bounded by: // @@ -179,16 +182,16 @@ impl<'a, 'b> Mul<&'b FieldElement2625> for &'a FieldElement2625 { let x7_2 = 2 * x[7]; let x9_2 = 2 * x[9]; - let z0 = m(x[0],y[0]) + m(x1_2,y9_19) + m(x[2],y8_19) + m(x3_2,y7_19) + m(x[4],y6_19) + m(x5_2,y5_19) + m(x[6],y4_19) + m(x7_2,y3_19) + m(x[8],y2_19) + m(x9_2,y1_19); - let z1 = m(x[0],y[1]) + m(x[1],y[0]) + m(x[2],y9_19) + m(x[3],y8_19) + m(x[4],y7_19) + m(x[5],y6_19) + m(x[6],y5_19) + m(x[7],y4_19) + m(x[8],y3_19) + m(x[9],y2_19); - let z2 = m(x[0],y[2]) + m(x1_2,y[1]) + m(x[2],y[0]) + m(x3_2,y9_19) + m(x[4],y8_19) + m(x5_2,y7_19) + m(x[6],y6_19) + m(x7_2,y5_19) + m(x[8],y4_19) + m(x9_2,y3_19); - let z3 = m(x[0],y[3]) + m(x[1],y[2]) + m(x[2],y[1]) + m(x[3],y[0]) + m(x[4],y9_19) + m(x[5],y8_19) + m(x[6],y7_19) + m(x[7],y6_19) + m(x[8],y5_19) + m(x[9],y4_19); - let z4 = m(x[0],y[4]) + m(x1_2,y[3]) + m(x[2],y[2]) + m(x3_2,y[1]) + m(x[4],y[0]) + m(x5_2,y9_19) + m(x[6],y8_19) + m(x7_2,y7_19) + m(x[8],y6_19) + m(x9_2,y5_19); - let z5 = m(x[0],y[5]) + m(x[1],y[4]) + m(x[2],y[3]) + m(x[3],y[2]) + m(x[4],y[1]) + m(x[5],y[0]) + m(x[6],y9_19) + m(x[7],y8_19) + m(x[8],y7_19) + m(x[9],y6_19); - let z6 = m(x[0],y[6]) + m(x1_2,y[5]) + m(x[2],y[4]) + m(x3_2,y[3]) + m(x[4],y[2]) + m(x5_2,y[1]) + m(x[6],y[0]) + m(x7_2,y9_19) + m(x[8],y8_19) + m(x9_2,y7_19); - let z7 = m(x[0],y[7]) + m(x[1],y[6]) + m(x[2],y[5]) + m(x[3],y[4]) + m(x[4],y[3]) + m(x[5],y[2]) + m(x[6],y[1]) + m(x[7],y[0]) + m(x[8],y9_19) + m(x[9],y8_19); - let z8 = m(x[0],y[8]) + m(x1_2,y[7]) + m(x[2],y[6]) + m(x3_2,y[5]) + m(x[4],y[4]) + m(x5_2,y[3]) + m(x[6],y[2]) + m(x7_2,y[1]) + m(x[8],y[0]) + m(x9_2,y9_19); - let z9 = m(x[0],y[9]) + m(x[1],y[8]) + m(x[2],y[7]) + m(x[3],y[6]) + m(x[4],y[5]) + m(x[5],y[4]) + m(x[6],y[3]) + m(x[7],y[2]) + m(x[8],y[1]) + m(x[9],y[0]); + let z0 = m(x[0], y[0]) + m(x1_2, y9_19) + m(x[2], y8_19) + m(x3_2, y7_19) + m(x[4], y6_19) + m(x5_2, y5_19) + m(x[6], y4_19) + m(x7_2, y3_19) + m(x[8], y2_19) + m(x9_2, y1_19); + let z1 = m(x[0], y[1]) + m(x[1], y[0]) + m(x[2], y9_19) + m(x[3], y8_19) + m(x[4], y7_19) + m(x[5], y6_19) + m(x[6], y5_19) + m(x[7], y4_19) + m(x[8], y3_19) + m(x[9], y2_19); + let z2 = m(x[0], y[2]) + m(x1_2, y[1]) + m(x[2], y[0]) + m(x3_2, y9_19) + m(x[4], y8_19) + m(x5_2, y7_19) + m(x[6], y6_19) + m(x7_2, y5_19) + m(x[8], y4_19) + m(x9_2, y3_19); + let z3 = m(x[0], y[3]) + m(x[1], y[2]) + m(x[2], y[1]) + m(x[3], y[0]) + m(x[4], y9_19) + m(x[5], y8_19) + m(x[6], y7_19) + m(x[7], y6_19) + m(x[8], y5_19) + m(x[9], y4_19); + let z4 = m(x[0], y[4]) + m(x1_2, y[3]) + m(x[2], y[2]) + m(x3_2, y[1]) + m(x[4], y[0]) + m(x5_2, y9_19) + m(x[6], y8_19) + m(x7_2, y7_19) + m(x[8], y6_19) + m(x9_2, y5_19); + let z5 = m(x[0], y[5]) + m(x[1], y[4]) + m(x[2], y[3]) + m(x[3], y[2]) + m(x[4], y[1]) + m(x[5], y[0]) + m(x[6], y9_19) + m(x[7], y8_19) + m(x[8], y7_19) + m(x[9], y6_19); + let z6 = m(x[0], y[6]) + m(x1_2, y[5]) + m(x[2], y[4]) + m(x3_2, y[3]) + m(x[4], y[2]) + m(x5_2, y[1]) + m(x[6], y[0]) + m(x7_2, y9_19) + m(x[8], y8_19) + m(x9_2, y7_19); + let z7 = m(x[0], y[7]) + m(x[1], y[6]) + m(x[2], y[5]) + m(x[3], y[4]) + m(x[4], y[3]) + m(x[5], y[2]) + m(x[6], y[1]) + m(x[7], y[0]) + m(x[8], y9_19) + m(x[9], y8_19); + let z8 = m(x[0], y[8]) + m(x1_2, y[7]) + m(x[2], y[6]) + m(x3_2, y[5]) + m(x[4], y[4]) + m(x5_2, y[3]) + m(x[6], y[2]) + m(x7_2, y[1]) + m(x[8], y[0]) + m(x9_2, y9_19); + let z9 = m(x[0], y[9]) + m(x[1], y[8]) + m(x[2], y[7]) + m(x[3], y[6]) + m(x[4], y[5]) + m(x[5], y[4]) + m(x[6], y[3]) + m(x[7], y[2]) + m(x[8], y[1]) + m(x[9], y[0]); // How big is the contribution to z[i+j] from x[i], y[j]? // @@ -342,11 +345,11 @@ impl FieldElement2625 { debug_assert!(i < 9); if i % 2 == 0 { // Even limbs have 26 bits - z[i+1] += z[i] >> 26; + z[i + 1] += z[i] >> 26; z[i] &= LOW_26_BITS; } else { // Odd limbs have 25 bits - z[i+1] += z[i] >> 25; + z[i + 1] += z[i] >> 25; z[i] &= LOW_25_BITS; } } @@ -363,7 +366,7 @@ impl FieldElement2625 { // and z[5] < 2^25 + 2^13.0002 < 2^25.0004 (good enough) // Last carry has a multiplication by 19: - z[0] += 19*(z[9] >> 25); + z[0] += 19 * (z[9] >> 25); z[9] &= LOW_25_BITS; // Since z[9] < 2^64, c < 2^(64-25) = 2^39, @@ -374,8 +377,16 @@ impl FieldElement2625 { // and we're done. FieldElement2625([ - z[0] as u32, z[1] as u32, z[2] as u32, z[3] as u32, z[4] as u32, - z[5] as u32, z[6] as u32, z[7] as u32, z[8] as u32, z[9] as u32, + z[0] as u32, + z[1] as u32, + z[2] as u32, + z[3] as u32, + z[4] as u32, + z[5] as u32, + z[6] as u32, + z[7] as u32, + z[8] as u32, + z[9] as u32, ]) } @@ -390,7 +401,7 @@ impl FieldElement2625 { /// encoding of every field element should decode, re-encode to /// the canonical encoding, and check that the input was /// canonical. - pub fn from_bytes(data: &[u8; 32]) -> FieldElement2625 { //FeFromBytes + pub fn from_bytes(data: &[u8; 32]) -> FieldElement2625 { #[inline] fn load3(b: &[u8]) -> u64 { (b[0] as u64) | ((b[1] as u64) << 8) | ((b[2] as u64) << 16) @@ -449,7 +460,7 @@ impl FieldElement2625 { q = (h[8] + q) >> 26; q = (h[9] + q) >> 25; - debug_assert!( q == 0 || q == 1 ); + debug_assert!(q == 0 || q == 1); // Now we can compute r as r = h - pq = r - (2^255-19)q = r + 19q - 2^255q @@ -481,7 +492,7 @@ impl FieldElement2625 { // ... but instead of carrying the value // (h[9] >> 25) = q*2^255 into another limb, // discard it, subtracting the value from h. - debug_assert!( (h[9] >> 25) == 0 || (h[9] >> 25) == 1); + debug_assert!((h[9] >> 25) == 0 || (h[9] >> 25) == 1); h[9] = h[9] & LOW_25_BITS; let mut s = [0u8; 32]; @@ -529,39 +540,41 @@ impl FieldElement2625 { // Optimized version of multiplication for the case of squaring. // Pre- and post- conditions identical to multiplication function. let x = &self.0; - let x0_2 = 2 * x[0]; - let x1_2 = 2 * x[1]; - let x2_2 = 2 * x[2]; - let x3_2 = 2 * x[3]; - let x4_2 = 2 * x[4]; - let x5_2 = 2 * x[5]; - let x6_2 = 2 * x[6]; - let x7_2 = 2 * x[7]; - let x5_19 = 19 * x[5]; - let x6_19 = 19 * x[6]; - let x7_19 = 19 * x[7]; - let x8_19 = 19 * x[8]; - let x9_19 = 19 * x[9]; + let x0_2 = 2 * x[0]; + let x1_2 = 2 * x[1]; + let x2_2 = 2 * x[2]; + let x3_2 = 2 * x[3]; + let x4_2 = 2 * x[4]; + let x5_2 = 2 * x[5]; + let x6_2 = 2 * x[6]; + let x7_2 = 2 * x[7]; + let x5_19 = 19 * x[5]; + let x6_19 = 19 * x[6]; + let x7_19 = 19 * x[7]; + let x8_19 = 19 * x[8]; + let x9_19 = 19 * x[9]; /// Helper function to multiply two 32-bit integers with 64 bits /// of output. #[inline(always)] - fn m(x: u32, y: u32) -> u64 { (x as u64) * (y as u64) } + fn m(x: u32, y: u32) -> u64 { + (x as u64) * (y as u64) + } // This block is rearranged so that instead of doing a 32-bit multiplication by 38, we do a // 64-bit multiplication by 2 on the results. This is because lg(38) is too big: we would // have less than 1 bit of headroom left, which is too little. - let mut z = [0u64;10]; - z[0] = m(x[0],x[0]) + m(x2_2,x8_19) + m(x4_2,x6_19) + (m(x1_2,x9_19) + m(x3_2,x7_19) + m(x[5],x5_19))*2; - z[1] = m(x0_2,x[1]) + m(x3_2,x8_19) + m(x5_2,x6_19) + (m(x[2],x9_19) + m(x[4],x7_19))*2; - z[2] = m(x0_2,x[2]) + m(x1_2,x[1]) + m(x4_2,x8_19) + m(x[6],x6_19) + (m(x3_2,x9_19) + m(x5_2,x7_19))*2; - z[3] = m(x0_2,x[3]) + m(x1_2,x[2]) + m(x5_2,x8_19) + (m(x[4],x9_19) + m(x[6],x7_19))*2; - z[4] = m(x0_2,x[4]) + m(x1_2,x3_2) + m(x[2],x[2]) + m(x6_2,x8_19) + (m(x5_2,x9_19) + m(x[7],x7_19))*2; - z[5] = m(x0_2,x[5]) + m(x1_2,x[4]) + m(x2_2,x[3]) + m(x7_2,x8_19) + m(x[6],x9_19)*2; - z[6] = m(x0_2,x[6]) + m(x1_2,x5_2) + m(x2_2,x[4]) + m(x3_2,x[3]) + m(x[8],x8_19) + m(x7_2,x9_19)*2; - z[7] = m(x0_2,x[7]) + m(x1_2,x[6]) + m(x2_2,x[5]) + m(x3_2,x[4]) + m(x[8],x9_19)*2; - z[8] = m(x0_2,x[8]) + m(x1_2,x7_2) + m(x2_2,x[6]) + m(x3_2,x5_2) + m(x[4],x[4]) + m(x[9],x9_19)*2; - z[9] = m(x0_2,x[9]) + m(x1_2,x[8]) + m(x2_2,x[7]) + m(x3_2,x[6]) + m(x4_2,x[5]) ; + let mut z = [0u64; 10]; + z[0] = m(x[0], x[0]) + m(x2_2, x8_19) + m(x4_2, x6_19) + (m(x1_2, x9_19) + m(x3_2, x7_19) + m(x[5], x5_19)) * 2; + z[1] = m(x0_2, x[1]) + m(x3_2, x8_19) + m(x5_2, x6_19) + (m(x[2], x9_19) + m(x[4], x7_19) ) * 2; + z[2] = m(x0_2, x[2]) + m(x1_2, x[1]) + m(x4_2, x8_19) + m(x[6], x6_19) + (m(x3_2, x9_19) + m(x5_2, x7_19)) * 2; + z[3] = m(x0_2, x[3]) + m(x1_2, x[2]) + m(x5_2, x8_19) + (m(x[4], x9_19) + m(x[6], x7_19) ) * 2; + z[4] = m(x0_2, x[4]) + m(x1_2, x3_2) + m(x[2], x[2]) + m(x6_2, x8_19) + (m(x5_2, x9_19) + m(x[7], x7_19)) * 2; + z[5] = m(x0_2, x[5]) + m(x1_2, x[4]) + m(x2_2, x[3]) + m(x7_2, x8_19) + m(x[6], x9_19) * 2; + z[6] = m(x0_2, x[6]) + m(x1_2, x5_2) + m(x2_2, x[4]) + m(x3_2, x[3]) + m(x[8], x8_19) + m(x7_2, x9_19) * 2; + z[7] = m(x0_2, x[7]) + m(x1_2, x[6]) + m(x2_2, x[5]) + m(x3_2, x[4]) + m(x[8], x9_19) * 2; + z[8] = m(x0_2, x[8]) + m(x1_2, x7_2) + m(x2_2, x[6]) + m(x3_2, x5_2) + m(x[4], x[4]) + m(x[9], x9_19) * 2; + z[9] = m(x0_2, x[9]) + m(x1_2, x[8]) + m(x2_2, x[7]) + m(x3_2, x[6]) + m(x4_2, x[5]) ; z } diff --git a/src/backend/serial/u32/scalar.rs b/src/backend/serial/u32/scalar.rs index 672cb89..dc28e35 100644 --- a/src/backend/serial/u32/scalar.rs +++ b/src/backend/serial/u32/scalar.rs @@ -72,15 +72,15 @@ impl Scalar29 { let top_mask = (1u32 << 24) - 1; let mut s = Scalar29::zero(); - s[ 0] = words[0] & mask; - s[ 1] = ((words[0] >> 29) | (words[1] << 3)) & mask; - s[ 2] = ((words[1] >> 26) | (words[2] << 6)) & mask; - s[ 3] = ((words[2] >> 23) | (words[3] << 9)) & mask; - s[ 4] = ((words[3] >> 20) | (words[4] << 12)) & mask; - s[ 5] = ((words[4] >> 17) | (words[5] << 15)) & mask; - s[ 6] = ((words[5] >> 14) | (words[6] << 18)) & mask; - s[ 7] = ((words[6] >> 11) | (words[7] << 21)) & mask; - s[ 8] = (words[7] >> 8) & top_mask; + s[0] = words[0] & mask; + s[1] = ((words[0] >> 29) | (words[1] << 3)) & mask; + s[2] = ((words[1] >> 26) | (words[2] << 6)) & mask; + s[3] = ((words[2] >> 23) | (words[3] << 9)) & mask; + s[4] = ((words[3] >> 20) | (words[4] << 12)) & mask; + s[5] = ((words[4] >> 17) | (words[5] << 15)) & mask; + s[6] = ((words[5] >> 14) | (words[6] << 18)) & mask; + s[7] = ((words[6] >> 11) | (words[7] << 21)) & mask; + s[8] = (words[7] >> 8) & top_mask; s } @@ -129,38 +129,38 @@ impl Scalar29 { pub fn to_bytes(&self) -> [u8; 32] { let mut s = [0u8; 32]; - s[0] = (self.0[ 0] >> 0) as u8; - s[1] = (self.0[ 0] >> 8) as u8; - s[2] = (self.0[ 0] >> 16) as u8; - s[3] = ((self.0[ 0] >> 24) | (self.0[ 1] << 5)) as u8; - s[4] = (self.0[ 1] >> 3) as u8; - s[5] = (self.0[ 1] >> 11) as u8; - s[6] = (self.0[ 1] >> 19) as u8; - s[7] = ((self.0[ 1] >> 27) | (self.0[ 2] << 2)) as u8; - s[8] = (self.0[ 2] >> 6) as u8; - s[9] = (self.0[ 2] >> 14) as u8; - s[10] = ((self.0[ 2] >> 22) | (self.0[ 3] << 7)) as u8; - s[11] = (self.0[ 3] >> 1) as u8; - s[12] = (self.0[ 3] >> 9) as u8; - s[13] = (self.0[ 3] >> 17) as u8; - s[14] = ((self.0[ 3] >> 25) | (self.0[ 4] << 4)) as u8; - s[15] = (self.0[ 4] >> 4) as u8; - s[16] = (self.0[ 4] >> 12) as u8; - s[17] = (self.0[ 4] >> 20) as u8; - s[18] = ((self.0[ 4] >> 28) | (self.0[ 5] << 1)) as u8; - s[19] = (self.0[ 5] >> 7) as u8; - s[20] = (self.0[ 5] >> 15) as u8; - s[21] = ((self.0[ 5] >> 23) | (self.0[ 6] << 6)) as u8; - s[22] = (self.0[ 6] >> 2) as u8; - s[23] = (self.0[ 6] >> 10) as u8; - s[24] = (self.0[ 6] >> 18) as u8; - s[25] = ((self.0[ 6] >> 26) | (self.0[ 7] << 3)) as u8; - s[26] = (self.0[ 7] >> 5) as u8; - s[27] = (self.0[ 7] >> 13) as u8; - s[28] = (self.0[ 7] >> 21) as u8; - s[29] = (self.0[ 8] >> 0) as u8; - s[30] = (self.0[ 8] >> 8) as u8; - s[31] = (self.0[ 8] >> 16) as u8; + s[ 0] = (self.0[0] >> 0) as u8; + s[ 1] = (self.0[0] >> 8) as u8; + s[ 2] = (self.0[0] >> 16) as u8; + s[ 3] = ((self.0[0] >> 24) | (self.0[1] << 5)) as u8; + s[ 4] = (self.0[1] >> 3) as u8; + s[ 5] = (self.0[1] >> 11) as u8; + s[ 6] = (self.0[1] >> 19) as u8; + s[ 7] = ((self.0[1] >> 27) | (self.0[2] << 2)) as u8; + s[ 8] = (self.0[2] >> 6) as u8; + s[ 9] = (self.0[2] >> 14) as u8; + s[10] = ((self.0[2] >> 22) | (self.0[3] << 7)) as u8; + s[11] = (self.0[3] >> 1) as u8; + s[12] = (self.0[3] >> 9) as u8; + s[13] = (self.0[3] >> 17) as u8; + s[14] = ((self.0[3] >> 25) | (self.0[4] << 4)) as u8; + s[15] = (self.0[4] >> 4) as u8; + s[16] = (self.0[4] >> 12) as u8; + s[17] = (self.0[4] >> 20) as u8; + s[18] = ((self.0[4] >> 28) | (self.0[5] << 1)) as u8; + s[19] = (self.0[5] >> 7) as u8; + s[20] = (self.0[5] >> 15) as u8; + s[21] = ((self.0[5] >> 23) | (self.0[6] << 6)) as u8; + s[22] = (self.0[6] >> 2) as u8; + s[23] = (self.0[6] >> 10) as u8; + s[24] = (self.0[6] >> 18) as u8; + s[25] = ((self.0[6] >> 26) | (self.0[7] << 3)) as u8; + s[26] = (self.0[7] >> 5) as u8; + s[27] = (self.0[7] >> 13) as u8; + s[28] = (self.0[7] >> 21) as u8; + s[29] = (self.0[8] >> 0) as u8; + s[30] = (self.0[8] >> 8) as u8; + s[31] = (self.0[8] >> 16) as u8; s } @@ -212,23 +212,23 @@ impl Scalar29 { pub (crate) fn mul_internal(a: &Scalar29, b: &Scalar29) -> [u64; 17] { let mut z = [0u64; 17]; - z[0] = m(a[0],b[0]); // c00 - z[1] = m(a[0],b[1]) + m(a[1],b[0]); // c01 - z[2] = m(a[0],b[2]) + m(a[1],b[1]) + m(a[2],b[0]); // c02 - z[3] = m(a[0],b[3]) + m(a[1],b[2]) + m(a[2],b[1]) + m(a[3],b[0]); // c03 - z[4] = m(a[0],b[4]) + m(a[1],b[3]) + m(a[2],b[2]) + m(a[3],b[1]) + m(a[4],b[0]); // c04 - z[5] = m(a[1],b[4]) + m(a[2],b[3]) + m(a[3],b[2]) + m(a[4],b[1]); // c05 - z[6] = m(a[2],b[4]) + m(a[3],b[3]) + m(a[4],b[2]); // c06 - z[7] = m(a[3],b[4]) + m(a[4],b[3]); // c07 - z[8] = (m(a[4],b[4])).wrapping_sub(z[3]); // c08 - c03 + z[0] = m(a[0], b[0]); // c00 + z[1] = m(a[0], b[1]) + m(a[1], b[0]); // c01 + z[2] = m(a[0], b[2]) + m(a[1], b[1]) + m(a[2], b[0]); // c02 + z[3] = m(a[0], b[3]) + m(a[1], b[2]) + m(a[2], b[1]) + m(a[3], b[0]); // c03 + z[4] = m(a[0], b[4]) + m(a[1], b[3]) + m(a[2], b[2]) + m(a[3], b[1]) + m(a[4], b[0]); // c04 + z[5] = m(a[1], b[4]) + m(a[2], b[3]) + m(a[3], b[2]) + m(a[4], b[1]); // c05 + z[6] = m(a[2], b[4]) + m(a[3], b[3]) + m(a[4], b[2]); // c06 + z[7] = m(a[3], b[4]) + m(a[4], b[3]); // c07 + z[8] = (m(a[4], b[4])).wrapping_sub(z[3]); // c08 - c03 - z[10] = z[5].wrapping_sub(m(a[5],b[5])); // c05mc10 - z[11] = z[6].wrapping_sub(m(a[5],b[6]) + m(a[6],b[5])); // c06mc11 - z[12] = z[7].wrapping_sub(m(a[5],b[7]) + m(a[6],b[6]) + m(a[7],b[5])); // c07mc12 - z[13] = m(a[5],b[8]) + m(a[6],b[7]) + m(a[7],b[6]) + m(a[8],b[5]); // c13 - z[14] = m(a[6],b[8]) + m(a[7],b[7]) + m(a[8],b[6]); // c14 - z[15] = m(a[7],b[8]) + m(a[8],b[7]); // c15 - z[16] = m(a[8],b[8]); // c16 + z[10] = z[5].wrapping_sub(m(a[5], b[5])); // c05mc10 + z[11] = z[6].wrapping_sub(m(a[5], b[6]) + m(a[6], b[5])); // c06mc11 + z[12] = z[7].wrapping_sub(m(a[5], b[7]) + m(a[6], b[6]) + m(a[7], b[5])); // c07mc12 + z[13] = m(a[5], b[8]) + m(a[6], b[7]) + m(a[7], b[6]) + m(a[8], b[5]); // c13 + z[14] = m(a[6], b[8]) + m(a[7], b[7]) + m(a[8], b[6]); // c14 + z[15] = m(a[7], b[8]) + m(a[8], b[7]); // c15 + z[16] = m(a[8], b[8]); // c16 z[ 5] = z[10].wrapping_sub(z[ 0]); // c05mc10 - c00 z[ 6] = z[11].wrapping_sub(z[ 1]); // c06mc11 - c01 @@ -239,27 +239,27 @@ impl Scalar29 { z[11] = z[16].wrapping_add(z[11]); // c16 + c06mc11 let aa = [ - a[0]+a[5], - a[1]+a[6], - a[2]+a[7], - a[3]+a[8] + a[0] + a[5], + a[1] + a[6], + a[2] + a[7], + a[3] + a[8] ]; let bb = [ - b[0]+b[5], - b[1]+b[6], - b[2]+b[7], - b[3]+b[8] + b[0] + b[5], + b[1] + b[6], + b[2] + b[7], + b[3] + b[8] ]; - z[ 5] = (m(aa[0],bb[0])) .wrapping_add(z[ 5]); // c20 + c05mc10 - c00 - z[ 6] = (m(aa[0],bb[1]) + m(aa[1],bb[0])) .wrapping_add(z[ 6]); // c21 + c06mc11 - c01 - z[ 7] = (m(aa[0],bb[2]) + m(aa[1],bb[1]) + m(aa[2],bb[0])) .wrapping_add(z[ 7]); // c22 + c07mc12 - c02 - z[ 8] = (m(aa[0],bb[3]) + m(aa[1],bb[2]) + m(aa[2],bb[1]) + m(aa[3],bb[0])) .wrapping_add(z[ 8]); // c23 + c08mc13 - c03 - z[ 9] = (m(aa[0], b[4]) + m(aa[1],bb[3]) + m(aa[2],bb[2]) + m(aa[3],bb[1]) + m(a[4],bb[0])).wrapping_sub(z[ 9]); // c24 - c14 - c04 - z[10] = ( m(aa[1], b[4]) + m(aa[2],bb[3]) + m(aa[3],bb[2]) + m(a[4],bb[1])).wrapping_sub(z[10]); // c25 - c15 - c05mc10 - z[11] = ( m(aa[2], b[4]) + m(aa[3],bb[3]) + m(a[4],bb[2])).wrapping_sub(z[11]); // c26 - c16 - c06mc11 - z[12] = ( m(aa[3], b[4]) + m(a[4],bb[3])).wrapping_sub(z[12]); // c27 - c07mc12 + z[ 5] = (m(aa[0], bb[0])) .wrapping_add(z[ 5]); // c20 + c05mc10 - c00 + z[ 6] = (m(aa[0], bb[1]) + m(aa[1], bb[0])) .wrapping_add(z[ 6]); // c21 + c06mc11 - c01 + z[ 7] = (m(aa[0], bb[2]) + m(aa[1], bb[1]) + m(aa[2], bb[0])) .wrapping_add(z[ 7]); // c22 + c07mc12 - c02 + z[ 8] = (m(aa[0], bb[3]) + m(aa[1], bb[2]) + m(aa[2], bb[1]) + m(aa[3], bb[0])) .wrapping_add(z[ 8]); // c23 + c08mc13 - c03 + z[ 9] = (m(aa[0], b[4]) + m(aa[1], bb[3]) + m(aa[2], bb[2]) + m(aa[3], bb[1]) + m(a[4], bb[0])).wrapping_sub(z[ 9]); // c24 - c14 - c04 + z[10] = ( m(aa[1], b[4]) + m(aa[2], bb[3]) + m(aa[3], bb[2]) + m(a[4], bb[1])).wrapping_sub(z[10]); // c25 - c15 - c05mc10 + z[11] = ( m(aa[2], b[4]) + m(aa[3], bb[3]) + m(a[4], bb[2])).wrapping_sub(z[11]); // c26 - c16 - c06mc11 + z[12] = ( m(aa[3], b[4]) + m(a[4], bb[3])).wrapping_sub(z[12]); // c27 - c07mc12 z } @@ -269,34 +269,34 @@ impl Scalar29 { #[rustfmt::skip] // keep alignment of calculations fn square_internal(a: &Scalar29) -> [u64; 17] { let aa = [ - a[0]*2, - a[1]*2, - a[2]*2, - a[3]*2, - a[4]*2, - a[5]*2, - a[6]*2, - a[7]*2 + a[0] * 2, + a[1] * 2, + a[2] * 2, + a[3] * 2, + a[4] * 2, + a[5] * 2, + a[6] * 2, + a[7] * 2 ]; [ - m( a[0],a[0]), - m(aa[0],a[1]), - m(aa[0],a[2]) + m( a[1],a[1]), - m(aa[0],a[3]) + m(aa[1],a[2]), - m(aa[0],a[4]) + m(aa[1],a[3]) + m( a[2],a[2]), - m(aa[0],a[5]) + m(aa[1],a[4]) + m(aa[2],a[3]), - m(aa[0],a[6]) + m(aa[1],a[5]) + m(aa[2],a[4]) + m( a[3],a[3]), - m(aa[0],a[7]) + m(aa[1],a[6]) + m(aa[2],a[5]) + m(aa[3],a[4]), - m(aa[0],a[8]) + m(aa[1],a[7]) + m(aa[2],a[6]) + m(aa[3],a[5]) + m( a[4],a[4]), - m(aa[1],a[8]) + m(aa[2],a[7]) + m(aa[3],a[6]) + m(aa[4],a[5]), - m(aa[2],a[8]) + m(aa[3],a[7]) + m(aa[4],a[6]) + m( a[5],a[5]), - m(aa[3],a[8]) + m(aa[4],a[7]) + m(aa[5],a[6]), - m(aa[4],a[8]) + m(aa[5],a[7]) + m( a[6],a[6]), - m(aa[5],a[8]) + m(aa[6],a[7]), - m(aa[6],a[8]) + m( a[7],a[7]), - m(aa[7],a[8]), - m( a[8],a[8]), + m( a[0], a[0]), + m(aa[0], a[1]), + m(aa[0], a[2]) + m( a[1], a[1]), + m(aa[0], a[3]) + m(aa[1], a[2]), + m(aa[0], a[4]) + m(aa[1], a[3]) + m( a[2], a[2]), + m(aa[0], a[5]) + m(aa[1], a[4]) + m(aa[2], a[3]), + m(aa[0], a[6]) + m(aa[1], a[5]) + m(aa[2], a[4]) + m( a[3], a[3]), + m(aa[0], a[7]) + m(aa[1], a[6]) + m(aa[2], a[5]) + m(aa[3], a[4]), + m(aa[0], a[8]) + m(aa[1], a[7]) + m(aa[2], a[6]) + m(aa[3], a[5]) + m( a[4], a[4]), + m(aa[1], a[8]) + m(aa[2], a[7]) + m(aa[3], a[6]) + m(aa[4], a[5]), + m(aa[2], a[8]) + m(aa[3], a[7]) + m(aa[4], a[6]) + m( a[5], a[5]), + m(aa[3], a[8]) + m(aa[4], a[7]) + m(aa[5], a[6]), + m(aa[4], a[8]) + m(aa[5], a[7]) + m( a[6], a[6]), + m(aa[5], a[8]) + m(aa[6], a[7]), + m(aa[6], a[8]) + m( a[7], a[7]), + m(aa[7], a[8]), + m( a[8], a[8]), ] } diff --git a/src/backend/serial/u64/field.rs b/src/backend/serial/u64/field.rs index 00d08f2..fb1059b 100644 --- a/src/backend/serial/u64/field.rs +++ b/src/backend/serial/u64/field.rs @@ -139,11 +139,11 @@ impl<'a, 'b> Mul<&'b FieldElement51> for &'a FieldElement51 { let b4_19 = b[4] * 19; // Multiply to get 128-bit coefficients of output - let c0: u128 = m(a[0],b[0]) + m(a[4],b1_19) + m(a[3],b2_19) + m(a[2],b3_19) + m(a[1],b4_19); - let mut c1: u128 = m(a[1],b[0]) + m(a[0],b[1]) + m(a[4],b2_19) + m(a[3],b3_19) + m(a[2],b4_19); - let mut c2: u128 = m(a[2],b[0]) + m(a[1],b[1]) + m(a[0],b[2]) + m(a[4],b3_19) + m(a[3],b4_19); - let mut c3: u128 = m(a[3],b[0]) + m(a[2],b[1]) + m(a[1],b[2]) + m(a[0],b[3]) + m(a[4],b4_19); - let mut c4: u128 = m(a[4],b[0]) + m(a[3],b[1]) + m(a[2],b[2]) + m(a[1],b[3]) + m(a[0],b[4]); + let c0: u128 = m(a[0], b[0]) + m(a[4], b1_19) + m(a[3], b2_19) + m(a[2], b3_19) + m(a[1], b4_19); + let mut c1: u128 = m(a[1], b[0]) + m(a[0], b[1]) + m(a[4], b2_19) + m(a[3], b3_19) + m(a[2], b4_19); + let mut c2: u128 = m(a[2], b[0]) + m(a[1], b[1]) + m(a[0], b[2]) + m(a[4], b3_19) + m(a[3], b4_19); + let mut c3: u128 = m(a[3], b[0]) + m(a[2], b[1]) + m(a[1], b[2]) + m(a[0], b[3]) + m(a[4], b4_19); + let mut c4: u128 = m(a[4], b[0]) + m(a[3], b[1]) + m(a[2], b[2]) + m(a[1], b[3]) + m(a[0] , b[4]); // How big are the c[i]? We have // @@ -388,7 +388,7 @@ impl FieldElement51 { // Now we can compute r as r = h - pq = r - (2^255-19)q = r + 19q - 2^255q - limbs[0] += 19*q; + limbs[0] += 19 * q; // Now carry the result to compute r + 19q ... let low_51_bit_mask = (1u64 << 51) - 1; @@ -406,38 +406,38 @@ impl FieldElement51 { // Now arrange the bits of the limbs. let mut s = [0u8;32]; - s[ 0] = limbs[0] as u8; - s[ 1] = (limbs[0] >> 8) as u8; - s[ 2] = (limbs[0] >> 16) as u8; - s[ 3] = (limbs[0] >> 24) as u8; - s[ 4] = (limbs[0] >> 32) as u8; - s[ 5] = (limbs[0] >> 40) as u8; + s[ 0] = limbs[0] as u8; + s[ 1] = (limbs[0] >> 8) as u8; + s[ 2] = (limbs[0] >> 16) as u8; + s[ 3] = (limbs[0] >> 24) as u8; + s[ 4] = (limbs[0] >> 32) as u8; + s[ 5] = (limbs[0] >> 40) as u8; s[ 6] = ((limbs[0] >> 48) | (limbs[1] << 3)) as u8; - s[ 7] = (limbs[1] >> 5) as u8; - s[ 8] = (limbs[1] >> 13) as u8; - s[ 9] = (limbs[1] >> 21) as u8; - s[10] = (limbs[1] >> 29) as u8; - s[11] = (limbs[1] >> 37) as u8; + s[ 7] = (limbs[1] >> 5) as u8; + s[ 8] = (limbs[1] >> 13) as u8; + s[ 9] = (limbs[1] >> 21) as u8; + s[10] = (limbs[1] >> 29) as u8; + s[11] = (limbs[1] >> 37) as u8; s[12] = ((limbs[1] >> 45) | (limbs[2] << 6)) as u8; - s[13] = (limbs[2] >> 2) as u8; - s[14] = (limbs[2] >> 10) as u8; - s[15] = (limbs[2] >> 18) as u8; - s[16] = (limbs[2] >> 26) as u8; - s[17] = (limbs[2] >> 34) as u8; - s[18] = (limbs[2] >> 42) as u8; + s[13] = (limbs[2] >> 2) as u8; + s[14] = (limbs[2] >> 10) as u8; + s[15] = (limbs[2] >> 18) as u8; + s[16] = (limbs[2] >> 26) as u8; + s[17] = (limbs[2] >> 34) as u8; + s[18] = (limbs[2] >> 42) as u8; s[19] = ((limbs[2] >> 50) | (limbs[3] << 1)) as u8; - s[20] = (limbs[3] >> 7) as u8; - s[21] = (limbs[3] >> 15) as u8; - s[22] = (limbs[3] >> 23) as u8; - s[23] = (limbs[3] >> 31) as u8; - s[24] = (limbs[3] >> 39) as u8; + s[20] = (limbs[3] >> 7) as u8; + s[21] = (limbs[3] >> 15) as u8; + s[22] = (limbs[3] >> 23) as u8; + s[23] = (limbs[3] >> 31) as u8; + s[24] = (limbs[3] >> 39) as u8; s[25] = ((limbs[3] >> 47) | (limbs[4] << 4)) as u8; - s[26] = (limbs[4] >> 4) as u8; - s[27] = (limbs[4] >> 12) as u8; - s[28] = (limbs[4] >> 20) as u8; - s[29] = (limbs[4] >> 28) as u8; - s[30] = (limbs[4] >> 36) as u8; - s[31] = (limbs[4] >> 44) as u8; + s[26] = (limbs[4] >> 4) as u8; + s[27] = (limbs[4] >> 12) as u8; + s[28] = (limbs[4] >> 20) as u8; + s[29] = (limbs[4] >> 28) as u8; + s[30] = (limbs[4] >> 36) as u8; + s[31] = (limbs[4] >> 44) as u8; // High bit should be zero. debug_assert!((s[31] & 0b1000_0000u8) == 0u8); @@ -453,7 +453,9 @@ impl FieldElement51 { /// Multiply two 64-bit integers with 128 bits of output. #[inline(always)] - fn m(x: u64, y: u64) -> u128 { (x as u128) * (y as u128) } + fn m(x: u64, y: u64) -> u128 { + (x as u128) * (y as u128) + } let mut a: [u64; 5] = self.0; diff --git a/src/backend/serial/u64/scalar.rs b/src/backend/serial/u64/scalar.rs index 3cd9511..e8ad36f 100644 --- a/src/backend/serial/u64/scalar.rs +++ b/src/backend/serial/u64/scalar.rs @@ -74,11 +74,11 @@ impl Scalar52 { let top_mask = (1u64 << 48) - 1; let mut s = Scalar52::zero(); - s[ 0] = words[0] & mask; - s[ 1] = ((words[0] >> 52) | (words[1] << 12)) & mask; - s[ 2] = ((words[1] >> 40) | (words[2] << 24)) & mask; - s[ 3] = ((words[2] >> 28) | (words[3] << 36)) & mask; - s[ 4] = (words[3] >> 16) & top_mask; + s[0] = words[0] & mask; + s[1] = ((words[0] >> 52) | (words[1] << 12)) & mask; + s[2] = ((words[1] >> 40) | (words[2] << 24)) & mask; + s[3] = ((words[2] >> 28) | (words[3] << 36)) & mask; + s[4] = (words[3] >> 16) & top_mask; s } @@ -97,16 +97,16 @@ impl Scalar52 { let mut lo = Scalar52::zero(); let mut hi = Scalar52::zero(); - lo[0] = words[ 0] & mask; - lo[1] = ((words[ 0] >> 52) | (words[ 1] << 12)) & mask; - lo[2] = ((words[ 1] >> 40) | (words[ 2] << 24)) & mask; - lo[3] = ((words[ 2] >> 28) | (words[ 3] << 36)) & mask; - lo[4] = ((words[ 3] >> 16) | (words[ 4] << 48)) & mask; - hi[0] = (words[ 4] >> 4) & mask; - hi[1] = ((words[ 4] >> 56) | (words[ 5] << 8)) & mask; - hi[2] = ((words[ 5] >> 44) | (words[ 6] << 20)) & mask; - hi[3] = ((words[ 6] >> 32) | (words[ 7] << 32)) & mask; - hi[4] = words[ 7] >> 20 ; + lo[0] = words[0] & mask; + lo[1] = ((words[0] >> 52) | (words[ 1] << 12)) & mask; + lo[2] = ((words[1] >> 40) | (words[ 2] << 24)) & mask; + lo[3] = ((words[2] >> 28) | (words[ 3] << 36)) & mask; + lo[4] = ((words[3] >> 16) | (words[ 4] << 48)) & mask; + hi[0] = (words[4] >> 4) & mask; + hi[1] = ((words[4] >> 56) | (words[ 5] << 8)) & mask; + hi[2] = ((words[5] >> 44) | (words[ 6] << 20)) & mask; + hi[3] = ((words[6] >> 32) | (words[ 7] << 32)) & mask; + hi[4] = words[7] >> 20 ; lo = Scalar52::montgomery_mul(&lo, &constants::R); // (lo * R) / R = lo hi = Scalar52::montgomery_mul(&hi, &constants::RR); // (hi * R^2) / R = hi * R @@ -119,16 +119,16 @@ impl Scalar52 { pub fn to_bytes(&self) -> [u8; 32] { let mut s = [0u8; 32]; - s[0] = (self.0[ 0] >> 0) as u8; - s[1] = (self.0[ 0] >> 8) as u8; - s[2] = (self.0[ 0] >> 16) as u8; - s[3] = (self.0[ 0] >> 24) as u8; - s[4] = (self.0[ 0] >> 32) as u8; - s[5] = (self.0[ 0] >> 40) as u8; - s[6] = ((self.0[ 0] >> 48) | (self.0[ 1] << 4)) as u8; - s[7] = (self.0[ 1] >> 4) as u8; - s[8] = (self.0[ 1] >> 12) as u8; - s[9] = (self.0[ 1] >> 20) as u8; + s[ 0] = (self.0[ 0] >> 0) as u8; + s[ 1] = (self.0[ 0] >> 8) as u8; + s[ 2] = (self.0[ 0] >> 16) as u8; + s[ 3] = (self.0[ 0] >> 24) as u8; + s[ 4] = (self.0[ 0] >> 32) as u8; + s[ 5] = (self.0[ 0] >> 40) as u8; + s[ 6] = ((self.0[ 0] >> 48) | (self.0[ 1] << 4)) as u8; + s[ 7] = (self.0[ 1] >> 4) as u8; + s[ 8] = (self.0[ 1] >> 12) as u8; + s[ 9] = (self.0[ 1] >> 20) as u8; s[10] = (self.0[ 1] >> 28) as u8; s[11] = (self.0[ 1] >> 36) as u8; s[12] = (self.0[ 1] >> 44) as u8; @@ -200,15 +200,15 @@ impl Scalar52 { pub (crate) fn mul_internal(a: &Scalar52, b: &Scalar52) -> [u128; 9] { let mut z = [0u128; 9]; - z[0] = m(a[0],b[0]); - z[1] = m(a[0],b[1]) + m(a[1],b[0]); - z[2] = m(a[0],b[2]) + m(a[1],b[1]) + m(a[2],b[0]); - z[3] = m(a[0],b[3]) + m(a[1],b[2]) + m(a[2],b[1]) + m(a[3],b[0]); - z[4] = m(a[0],b[4]) + m(a[1],b[3]) + m(a[2],b[2]) + m(a[3],b[1]) + m(a[4],b[0]); - z[5] = m(a[1],b[4]) + m(a[2],b[3]) + m(a[3],b[2]) + m(a[4],b[1]); - z[6] = m(a[2],b[4]) + m(a[3],b[3]) + m(a[4],b[2]); - z[7] = m(a[3],b[4]) + m(a[4],b[3]); - z[8] = m(a[4],b[4]); + z[0] = m(a[0], b[0]); + z[1] = m(a[0], b[1]) + m(a[1], b[0]); + z[2] = m(a[0], b[2]) + m(a[1], b[1]) + m(a[2], b[0]); + z[3] = m(a[0], b[3]) + m(a[1], b[2]) + m(a[2], b[1]) + m(a[3], b[0]); + z[4] = m(a[0], b[4]) + m(a[1], b[3]) + m(a[2], b[2]) + m(a[3], b[1]) + m(a[4], b[0]); + z[5] = m(a[1], b[4]) + m(a[2], b[3]) + m(a[3], b[2]) + m(a[4], b[1]); + z[6] = m(a[2], b[4]) + m(a[3], b[3]) + m(a[4], b[2]); + z[7] = m(a[3], b[4]) + m(a[4], b[3]); + z[8] = m(a[4], b[4]); z } @@ -218,22 +218,22 @@ impl Scalar52 { #[rustfmt::skip] // keep alignment of return calculations fn square_internal(a: &Scalar52) -> [u128; 9] { let aa = [ - a[0]*2, - a[1]*2, - a[2]*2, - a[3]*2, + a[0] * 2, + a[1] * 2, + a[2] * 2, + a[3] * 2, ]; [ - m( a[0],a[0]), - m(aa[0],a[1]), - m(aa[0],a[2]) + m( a[1],a[1]), - m(aa[0],a[3]) + m(aa[1],a[2]), - m(aa[0],a[4]) + m(aa[1],a[3]) + m( a[2],a[2]), - m(aa[1],a[4]) + m(aa[2],a[3]), - m(aa[2],a[4]) + m( a[3],a[3]), - m(aa[3],a[4]), - m(a[4],a[4]) + m( a[0], a[0]), + m(aa[0], a[1]), + m(aa[0], a[2]) + m( a[1], a[1]), + m(aa[0], a[3]) + m(aa[1], a[2]), + m(aa[0], a[4]) + m(aa[1], a[3]) + m( a[2], a[2]), + m(aa[1], a[4]) + m(aa[2], a[3]), + m(aa[2], a[4]) + m( a[3], a[3]), + m(aa[3], a[4]), + m(a[4], a[4]) ] } @@ -245,7 +245,7 @@ impl Scalar52 { #[inline(always)] fn part1(sum: u128) -> (u128, u64) { let p = (sum as u64).wrapping_mul(constants::LFACTOR) & ((1u64 << 52) - 1); - ((sum + m(p,constants::L[0])) >> 52, p) + ((sum + m(p, constants::L[0])) >> 52, p) } #[inline(always)] @@ -259,20 +259,20 @@ impl Scalar52 { // the first half computes the Montgomery adjustment factor n, and begins adding n*l to make limbs divisible by R let (carry, n0) = part1( limbs[0]); - let (carry, n1) = part1(carry + limbs[1] + m(n0,l[1])); - let (carry, n2) = part1(carry + limbs[2] + m(n0,l[2]) + m(n1,l[1])); - let (carry, n3) = part1(carry + limbs[3] + m(n1,l[2]) + m(n2,l[1])); - let (carry, n4) = part1(carry + limbs[4] + m(n0,l[4]) + m(n2,l[2]) + m(n3,l[1])); + let (carry, n1) = part1(carry + limbs[1] + m(n0, l[1])); + let (carry, n2) = part1(carry + limbs[2] + m(n0, l[2]) + m(n1, l[1])); + let (carry, n3) = part1(carry + limbs[3] + m(n1, l[2]) + m(n2, l[1])); + let (carry, n4) = part1(carry + limbs[4] + m(n0, l[4]) + m(n2, l[2]) + m(n3, l[1])); // limbs is divisible by R now, so we can divide by R by simply storing the upper half as the result - let (carry, r0) = part2(carry + limbs[5] + m(n1,l[4]) + m(n3,l[2]) + m(n4,l[1])); - let (carry, r1) = part2(carry + limbs[6] + m(n2,l[4]) + m(n4,l[2])); - let (carry, r2) = part2(carry + limbs[7] + m(n3,l[4]) ); - let (carry, r3) = part2(carry + limbs[8] + m(n4,l[4])); + let (carry, r0) = part2(carry + limbs[5] + m(n1, l[4]) + m(n3, l[2]) + m(n4, l[1])); + let (carry, r1) = part2(carry + limbs[6] + m(n2,l[4]) + m(n4, l[2])); + let (carry, r2) = part2(carry + limbs[7] + m(n3, l[4]) ); + let (carry, r3) = part2(carry + limbs[8] + m(n4, l[4])); let r4 = carry as u64; // result may be >= l, so attempt to subtract l - Scalar52::sub(&Scalar52([r0,r1,r2,r3,r4]), l) + Scalar52::sub(&Scalar52([r0, r1, r2, r3, r4]), l) } /// Compute `a * b` (mod l) diff --git a/src/backend/vector/avx2/field.rs b/src/backend/vector/avx2/field.rs index afc9dbb..6e7da01 100644 --- a/src/backend/vector/avx2/field.rs +++ b/src/backend/vector/avx2/field.rs @@ -624,31 +624,31 @@ impl FieldElement2625x4 { let (x6, x7) = unpack_pair(self.0[3]); let (x8, x9) = unpack_pair(self.0[4]); - let x0_2 = x0 << 1; - let x1_2 = x1 << 1; - let x2_2 = x2 << 1; - let x3_2 = x3 << 1; - let x4_2 = x4 << 1; - let x5_2 = x5 << 1; - let x6_2 = x6 << 1; - let x7_2 = x7 << 1; + let x0_2 = x0 << 1; + let x1_2 = x1 << 1; + let x2_2 = x2 << 1; + let x3_2 = x3 << 1; + let x4_2 = x4 << 1; + let x5_2 = x5 << 1; + let x6_2 = x6 << 1; + let x7_2 = x7 << 1; - let x5_19 = m_lo(v19, x5); - let x6_19 = m_lo(v19, x6); - let x7_19 = m_lo(v19, x7); - let x8_19 = m_lo(v19, x8); - let x9_19 = m_lo(v19, x9); + let x5_19 = m_lo(v19, x5); + let x6_19 = m_lo(v19, x6); + let x7_19 = m_lo(v19, x7); + let x8_19 = m_lo(v19, x8); + let x9_19 = m_lo(v19, x9); - let mut z0 = m(x0, x0) + m(x2_2,x8_19) + m(x4_2,x6_19) + ((m(x1_2,x9_19) + m(x3_2,x7_19) + m(x5,x5_19)) << 1); - let mut z1 = m(x0_2,x1) + m(x3_2,x8_19) + m(x5_2,x6_19) + ((m(x2,x9_19) + m(x4,x7_19)) << 1); - let mut z2 = m(x0_2,x2) + m(x1_2,x1) + m(x4_2,x8_19) + m(x6,x6_19) + ((m(x3_2,x9_19) + m(x5_2,x7_19)) << 1); - let mut z3 = m(x0_2,x3) + m(x1_2,x2) + m(x5_2,x8_19) + ((m(x4,x9_19) + m(x6,x7_19)) << 1); - let mut z4 = m(x0_2,x4) + m(x1_2,x3_2) + m(x2, x2) + m(x6_2,x8_19) + ((m(x5_2,x9_19) + m(x7,x7_19)) << 1); - let mut z5 = m(x0_2,x5) + m(x1_2,x4) + m(x2_2,x3) + m(x7_2,x8_19) + ((m(x6,x9_19)) << 1); - let mut z6 = m(x0_2,x6) + m(x1_2,x5_2) + m(x2_2,x4) + m(x3_2,x3) + m(x8,x8_19) + ((m(x7_2,x9_19)) << 1); - let mut z7 = m(x0_2,x7) + m(x1_2,x6) + m(x2_2,x5) + m(x3_2,x4) + ((m(x8,x9_19)) << 1); - let mut z8 = m(x0_2,x8) + m(x1_2,x7_2) + m(x2_2,x6) + m(x3_2,x5_2) + m(x4,x4) + ((m(x9,x9_19)) << 1); - let mut z9 = m(x0_2,x9) + m(x1_2,x8) + m(x2_2,x7) + m(x3_2,x6) + m(x4_2,x5); + let mut z0 = m(x0, x0) + m(x2_2, x8_19) + m(x4_2, x6_19) + ((m(x1_2, x9_19) + m(x3_2, x7_19) + m(x5, x5_19)) << 1); + let mut z1 = m(x0_2, x1) + m(x3_2, x8_19) + m(x5_2, x6_19) + ((m(x2, x9_19) + m(x4, x7_19)) << 1); + let mut z2 = m(x0_2, x2) + m(x1_2, x1) + m(x4_2, x8_19) + m(x6, x6_19) + ((m(x3_2, x9_19) + m(x5_2, x7_19)) << 1); + let mut z3 = m(x0_2, x3) + m(x1_2, x2) + m(x5_2, x8_19) + ((m(x4, x9_19) + m(x6, x7_19)) << 1); + let mut z4 = m(x0_2, x4) + m(x1_2, x3_2) + m(x2, x2) + m(x6_2, x8_19) + ((m(x5_2, x9_19) + m(x7, x7_19)) << 1); + let mut z5 = m(x0_2, x5) + m(x1_2, x4) + m(x2_2, x3) + m(x7_2, x8_19) + ((m(x6, x9_19)) << 1); + let mut z6 = m(x0_2, x6) + m(x1_2, x5_2) + m(x2_2, x4) + m(x3_2, x3) + m(x8, x8_19) + ((m(x7_2, x9_19)) << 1); + let mut z7 = m(x0_2, x7) + m(x1_2, x6) + m(x2_2, x5) + m(x3_2, x4) + ((m(x8, x9_19)) << 1); + let mut z8 = m(x0_2, x8) + m(x1_2, x7_2) + m(x2_2, x6) + m(x3_2, x5_2) + m(x4, x4) + ((m(x9, x9_19)) << 1); + let mut z9 = m(x0_2, x9) + m(x1_2, x8) + m(x2_2, x7) + m(x3_2, x6) + m(x4_2, x5) ; // The biggest z_i is bounded as z_i < 249*2^(51 + 2*b); // if b < 1.5 we get z_i < 4485585228861014016. @@ -828,16 +828,16 @@ impl<'a, 'b> Mul<&'b FieldElement2625x4> for &'a FieldElement2625x4 { let x7_2 = x7 + x7; let x9_2 = x9 + x9; - let z0 = m(x0,y0) + m(x1_2,y9_19) + m(x2,y8_19) + m(x3_2,y7_19) + m(x4,y6_19) + m(x5_2,y5_19) + m(x6,y4_19) + m(x7_2,y3_19) + m(x8,y2_19) + m(x9_2,y1_19); - let z1 = m(x0,y1) + m(x1,y0) + m(x2,y9_19) + m(x3,y8_19) + m(x4,y7_19) + m(x5,y6_19) + m(x6,y5_19) + m(x7,y4_19) + m(x8,y3_19) + m(x9,y2_19); - let z2 = m(x0,y2) + m(x1_2,y1) + m(x2,y0) + m(x3_2,y9_19) + m(x4,y8_19) + m(x5_2,y7_19) + m(x6,y6_19) + m(x7_2,y5_19) + m(x8,y4_19) + m(x9_2,y3_19); - let z3 = m(x0,y3) + m(x1,y2) + m(x2,y1) + m(x3,y0) + m(x4,y9_19) + m(x5,y8_19) + m(x6,y7_19) + m(x7,y6_19) + m(x8,y5_19) + m(x9,y4_19); - let z4 = m(x0,y4) + m(x1_2,y3) + m(x2,y2) + m(x3_2,y1) + m(x4,y0) + m(x5_2,y9_19) + m(x6,y8_19) + m(x7_2,y7_19) + m(x8,y6_19) + m(x9_2,y5_19); - let z5 = m(x0,y5) + m(x1,y4) + m(x2,y3) + m(x3,y2) + m(x4,y1) + m(x5,y0) + m(x6,y9_19) + m(x7,y8_19) + m(x8,y7_19) + m(x9,y6_19); - let z6 = m(x0,y6) + m(x1_2,y5) + m(x2,y4) + m(x3_2,y3) + m(x4,y2) + m(x5_2,y1) + m(x6,y0) + m(x7_2,y9_19) + m(x8,y8_19) + m(x9_2,y7_19); - let z7 = m(x0,y7) + m(x1,y6) + m(x2,y5) + m(x3,y4) + m(x4,y3) + m(x5,y2) + m(x6,y1) + m(x7,y0) + m(x8,y9_19) + m(x9,y8_19); - let z8 = m(x0,y8) + m(x1_2,y7) + m(x2,y6) + m(x3_2,y5) + m(x4,y4) + m(x5_2,y3) + m(x6,y2) + m(x7_2,y1) + m(x8,y0) + m(x9_2,y9_19); - let z9 = m(x0,y9) + m(x1,y8) + m(x2,y7) + m(x3,y6) + m(x4,y5) + m(x5,y4) + m(x6,y3) + m(x7,y2) + m(x8,y1) + m(x9,y0); + let z0 = m(x0, y0) + m(x1_2, y9_19) + m(x2, y8_19) + m(x3_2, y7_19) + m(x4, y6_19) + m(x5_2, y5_19) + m(x6, y4_19) + m(x7_2, y3_19) + m(x8, y2_19) + m(x9_2, y1_19); + let z1 = m(x0, y1) + m(x1, y0) + m(x2, y9_19) + m(x3, y8_19) + m(x4, y7_19) + m(x5, y6_19) + m(x6, y5_19) + m(x7, y4_19) + m(x8, y3_19) + m(x9, y2_19); + let z2 = m(x0, y2) + m(x1_2, y1) + m(x2, y0) + m(x3_2, y9_19) + m(x4, y8_19) + m(x5_2, y7_19) + m(x6, y6_19) + m(x7_2, y5_19) + m(x8, y4_19) + m(x9_2, y3_19); + let z3 = m(x0, y3) + m(x1, y2) + m(x2, y1) + m(x3, y0) + m(x4, y9_19) + m(x5, y8_19) + m(x6, y7_19) + m(x7, y6_19) + m(x8, y5_19) + m(x9, y4_19); + let z4 = m(x0, y4) + m(x1_2, y3) + m(x2, y2) + m(x3_2, y1) + m(x4, y0) + m(x5_2, y9_19) + m(x6, y8_19) + m(x7_2, y7_19) + m(x8, y6_19) + m(x9_2, y5_19); + let z5 = m(x0, y5) + m(x1, y4) + m(x2, y3) + m(x3, y2) + m(x4, y1) + m(x5, y0) + m(x6, y9_19) + m(x7, y8_19) + m(x8, y7_19) + m(x9, y6_19); + let z6 = m(x0, y6) + m(x1_2, y5) + m(x2, y4) + m(x3_2, y3) + m(x4, y2) + m(x5_2, y1) + m(x6, y0) + m(x7_2, y9_19) + m(x8, y8_19) + m(x9_2, y7_19); + let z7 = m(x0, y7) + m(x1, y6) + m(x2, y5) + m(x3, y4) + m(x4, y3) + m(x5, y2) + m(x6, y1) + m(x7, y0) + m(x8, y9_19) + m(x9, y8_19); + let z8 = m(x0, y8) + m(x1_2, y7) + m(x2, y6) + m(x3_2, y5) + m(x4, y4) + m(x5_2, y3) + m(x6, y2) + m(x7_2, y1) + m(x8, y0) + m(x9_2, y9_19); + let z9 = m(x0, y9) + m(x1, y8) + m(x2, y7) + m(x3, y6) + m(x4, y5) + m(x5, y4) + m(x6, y3) + m(x7, y2) + m(x8, y1) + m(x9, y0); // The bounds on z[i] are the same as in the serial 32-bit code // and the comment below is copied from there: