From 00d8b6ea4f177a7dcf9ba06c8c08cc0e303e6fdf Mon Sep 17 00:00:00 2001
From: Henry de Valence <hdevalence@hdevalence.ca>
Date: Wed, 13 Jun 2018 12:32:42 -0700
Subject: [PATCH] Change `negate_D`, `negate_D_lazy` to `impl Neg`,
 `negate_lazy`

---
 src/backend/avx2/edwards.rs |   9 ++--
 src/backend/avx2/field.rs   | 101 +++++++++++++-----------------------
 2 files changed, 40 insertions(+), 70 deletions(-)

diff --git a/src/backend/avx2/edwards.rs b/src/backend/avx2/edwards.rs
index be154dd..73db3db 100644
--- a/src/backend/avx2/edwards.rs
+++ b/src/backend/avx2/edwards.rs
@@ -176,7 +176,7 @@ impl From<ExtendedPoint> for CachedPoint {
         x.scale_by_curve_constants();
 
         // x = (121666*S2 121666*S3 2*121666*Z2 -2*121665*T2)
-        x.negate_D();
+        x = x.blend(-x, Lanes::D);
 
         CachedPoint(x)
     }
@@ -210,10 +210,9 @@ impl<'a> Neg for &'a CachedPoint {
     type Output = CachedPoint;
 
     fn neg(self) -> CachedPoint {
-        let mut neg = *self;
-        neg.0.swap_AB();
-        neg.0.negate_D_lazy();
-        neg
+        let mut coords = self.0;
+        coords.swap_AB();
+        CachedPoint(coords.blend(coords.negate_lazy(), Lanes::D))
     }
 }
 
diff --git a/src/backend/avx2/field.rs b/src/backend/avx2/field.rs
index d5ac6a8..c7ee80b 100644
--- a/src/backend/avx2/field.rs
+++ b/src/backend/avx2/field.rs
@@ -24,7 +24,7 @@ pub const D_LANES64: u8 = 0b11_00_00_00;
 
 pub const ALL_LANES: u8 = A_LANES | B_LANES | C_LANES | D_LANES;
 
-use core::ops::{Add, Mul};
+use core::ops::{Add, Mul, Neg};
 use core::simd::{i32x8, u32x8, u64x4, IntoBits};
 
 use backend::avx2::constants::{P_TIMES_16_HI, P_TIMES_16_LO, P_TIMES_2_HI, P_TIMES_2_LO};
@@ -195,73 +195,20 @@ impl FieldElement32x4 {
         return out;
     }
 
-    /// Negate the \\(D\\) variable of \\((A,B,C,D)\\).
+    /// Given \\((A,B,C,D)\\), compute \\((-A,-B,-C,-D)\\), without
+    /// performing a reduction.
     ///
     /// Input limbs must be less than the limbs of \\(2p\\), i.e., freshly reduced.
-    pub fn negate_D_lazy(&mut self) {
-        unsafe {
-            use core::arch::x86_64::_mm256_blend_epi32;
-            self.0[0] = _mm256_blend_epi32(
-                self.0[0].into_bits(),
-                (P_TIMES_2_LO - self.0[0]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[1] = _mm256_blend_epi32(
-                self.0[1].into_bits(),
-                (P_TIMES_2_HI - self.0[1]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[2] = _mm256_blend_epi32(
-                self.0[2].into_bits(),
-                (P_TIMES_2_HI - self.0[2]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[3] = _mm256_blend_epi32(
-                self.0[3].into_bits(),
-                (P_TIMES_2_HI - self.0[3]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[4] = _mm256_blend_epi32(
-                self.0[4].into_bits(),
-                (P_TIMES_2_HI - self.0[4]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-        }
-    }
-
-    /// Negate the \\(D\\) variable of \\((A,B,C,D)\\).
     ///
-    /// Input limbs must be less than the limbs of \\(2p\\), i.e., freshly reduced.
-    pub fn negate_D(&mut self) {
-        unsafe {
-            use core::arch::x86_64::_mm256_blend_epi32;
-            self.0[0] = _mm256_blend_epi32(
-                self.0[0].into_bits(),
-                (P_TIMES_16_LO - self.0[0]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[1] = _mm256_blend_epi32(
-                self.0[1].into_bits(),
-                (P_TIMES_16_HI - self.0[1]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[2] = _mm256_blend_epi32(
-                self.0[2].into_bits(),
-                (P_TIMES_16_HI - self.0[2]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[3] = _mm256_blend_epi32(
-                self.0[3].into_bits(),
-                (P_TIMES_16_HI - self.0[3]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-            self.0[4] = _mm256_blend_epi32(
-                self.0[4].into_bits(),
-                (P_TIMES_16_HI - self.0[4]).into_bits(),
-                D_LANES as i32,
-            ).into_bits();
-        }
-        self.reduce32();
+    /// The output limbs are bounded by \\(2p\\).
+    pub fn negate_lazy(&self) -> FieldElement32x4 {
+        FieldElement32x4([
+            P_TIMES_2_LO - self.0[0],
+            P_TIMES_2_HI - self.0[1],
+            P_TIMES_2_HI - self.0[2],
+            P_TIMES_2_HI - self.0[3],
+            P_TIMES_2_HI - self.0[4],
+        ])
     }
 
     /// Given `self = (A,B,C,D)`, set `self = (B,A,C,D)`
@@ -639,6 +586,30 @@ impl FieldElement32x4 {
     }
 }
 
+impl Neg for FieldElement32x4 {
+    type Output = FieldElement32x4;
+
+    /// Given \\((A,B,C,D)\\), compute \\((-A,-B,-C,-D)\\), and
+    /// perform a reduction.
+    ///
+    /// The input limbs can be any size.
+    ///
+    /// The output limbs are freshly reduced.
+    #[inline]
+    fn neg(self) -> FieldElement32x4 {
+        let mut neg = FieldElement32x4([
+            P_TIMES_16_LO - self.0[0],
+            P_TIMES_16_HI - self.0[1],
+            P_TIMES_16_HI - self.0[2],
+            P_TIMES_16_HI - self.0[3],
+            P_TIMES_16_HI - self.0[4],
+        ]);
+        neg.reduce32();
+
+        neg
+    }
+}
+
 impl Add<FieldElement32x4> for FieldElement32x4 {
     type Output = FieldElement32x4;
     #[inline]