diff --git a/Cargo.toml b/Cargo.toml
index e317204..ac1b8d6 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -59,6 +59,7 @@ rand = "0.3"
 generic-array = "^0.8"
 digest = "0.6"
 arrayref = "0.3.4"
+stdsimd = { git = "https://github.com/hdevalence/stdsimd", branch="feature/more-avx2" }
 
 [build-dependencies.serde]
 version = "1.0"
diff --git a/build.rs b/build.rs
index 13d9715..4fba073 100644
--- a/build.rs
+++ b/build.rs
@@ -21,6 +21,8 @@ use std::path::Path;
 // For instance, this shouldn't exist here at all, but it does.
 #[cfg(feature = "serde")]
 extern crate serde;
+#[cfg(feature = "yolocrypto")]
+extern crate stdsimd;
 
 // Public modules
 
diff --git a/src/avx2/edwards.rs b/src/backend/avx2/edwards.rs
similarity index 99%
rename from src/avx2/edwards.rs
rename to src/backend/avx2/edwards.rs
index c283e08..4db18f9 100644
--- a/src/avx2/edwards.rs
+++ b/src/backend/avx2/edwards.rs
@@ -25,8 +25,8 @@ use scalar::Scalar;
 
 use traits::Identity;
 
-use avx2::field::FieldElement32x4;
-use avx2::field::P_TIMES_2;
+use backend::avx2::field::FieldElement32x4;
+use backend::avx2::field::P_TIMES_2;
 
 /// A point on Curve25519, represented in an AVX2-friendly format.
 #[derive(Copy, Clone, Debug)]
@@ -324,17 +324,18 @@ impl<'a, 'b> Mul<&'b Scalar> for &'a EdwardsBasepointTable {
     /// takes `-8 ≤ x < 8` and `[16^2i * B, ..., 8 * 16^2i * B]`,
     /// and returns `x * 16^2i * B` in constant time.
     fn mul(self, scalar: &'b Scalar) -> ExtendedPoint {
+        use traits::select_precomputed_point;
         let e = scalar.to_radix_16();
         let mut h = ExtendedPoint::identity();
 
         for i in (0..64).filter(|x| x % 2 == 1) {
-            h = &h + &edwards::select_precomputed_point(e[i], &self.0[i/2]);
+            h = &h + &select_precomputed_point(e[i], &self.0[i/2]);
         }
 
         h = h.mult_by_pow_2(4);
 
         for i in (0..64).filter(|x| x % 2 == 0) {
-            h = &h + &edwards::select_precomputed_point(e[i], &self.0[i/2]);
+            h = &h + &select_precomputed_point(e[i], &self.0[i/2]);
         }
 
         h
@@ -395,7 +396,7 @@ pub fn multiscalar_mult<'a, 'b, I, J>(scalars: I, points: J) -> ExtendedPoint
     where I: IntoIterator<Item = &'a Scalar>,
           J: IntoIterator<Item = &'b ExtendedPoint>
 {
-    use edwards::select_precomputed_point;
+    use traits::select_precomputed_point;
     //assert_eq!(scalars.len(), points.len());
 
     let lookup_tables: Vec<_> = points.into_iter()
diff --git a/src/avx2/field.rs b/src/backend/avx2/field.rs
similarity index 100%
rename from src/avx2/field.rs
rename to src/backend/avx2/field.rs
diff --git a/src/avx2/mod.rs b/src/backend/avx2/mod.rs
similarity index 100%
rename from src/avx2/mod.rs
rename to src/backend/avx2/mod.rs
diff --git a/src/backend/mod.rs b/src/backend/mod.rs
index be3a347..b6845da 100644
--- a/src/backend/mod.rs
+++ b/src/backend/mod.rs
@@ -28,3 +28,7 @@ pub mod u32;
 #[cfg(feature="radix_51")]
 pub mod u64;
 
+/// Code using AVX2.
+#[cfg(all(feature="yolocrypto", not(feature="radix_51")))]
+pub mod avx2;
+
diff --git a/src/lib.rs b/src/lib.rs
index c3a4255..2106e88 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -94,9 +94,5 @@ pub(crate) mod field;
 // Arithmetic backends (using u32, u64, etc) live here
 pub(crate) mod backend;
 
-// XXX this should be in backend
-#[cfg(all(feature="yolocrypto", not(feature="radix_51")))]
-pub(crate) mod avx2;
-
 // Internal curve models which are not part of the public API.
 pub(crate) mod curve_models;