horcrux.dynamic_detect.patch

diff --git a/horcrux/src/gf2n.rs b/horcrux/src/gf2n.rs
index 5bc0221..501d455 100644
--- a/horcrux/src/gf2n.rs
+++ b/horcrux/src/gf2n.rs
@@ -294,13 +294,9 @@ fn mul_clmul_u64<const NWORDS: usize, const A: usize, const B: usize, const C: u
     GF2n::<u64, NWORDS, A, B, C>::propagate_carries(words, carry)
 }
 
-#[cfg(all(
-    feature = "clmul",
-    target_arch = "aarch64",
-    target_feature = "neon",
-    target_feature = "aes"
-))]
-fn mul_clmul_u64<const NWORDS: usize, const A: usize, const B: usize, const C: usize>(
+#[cfg(all(feature = "clmul", target_arch = "aarch64"))]
+#[target_feature(enable = "neon", enable = "aes")]
+unsafe fn mul_clmul_u64<const NWORDS: usize, const A: usize, const B: usize, const C: usize>(
     x: &GF2n<u64, NWORDS, A, B, C>,
     y: &GF2n<u64, NWORDS, A, B, C>,
 ) -> GF2n<u64, NWORDS, A, B, C> {
@@ -316,7 +312,7 @@ fn mul_clmul_u64<const NWORDS: usize, const A: usize, const B: usize, const C: u
         for j in 0..NWORDS {
             let yj = y.words[j];
             // Safety: target_feature's "neon" and "aes" are available in this function.
-            let clmul: u128 = unsafe { vmull_p64(xi, yj) };
+            let clmul: u128 = vmull_p64(xi, yj);
             let low: u64 = clmul as u64;
             let high: u64 = (clmul >> 64) as u64;
 
@@ -540,12 +536,7 @@ impl<W: Word, const NWORDS: usize, const A: usize, const B: usize, const C: usiz
             target_feature = "sse2",
             target_feature = "pclmulqdq"
         ),
-        all(
-            feature = "clmul",
-            target_arch = "aarch64",
-            target_feature = "neon",
-            target_feature = "aes"
-        )
+        all(feature = "clmul", target_arch = "aarch64")
     ))]
     fn propagate_carries(mut words: [W; NWORDS], carry: [W; NWORDS]) -> Self {
         if NWORDS == 1 {
@@ -672,19 +663,11 @@ impl<W: Word, const NWORDS: usize, const A: usize, const B: usize, const C: usiz
     type Output = Self;
 
     fn mul(self, other: &Self) -> Self {
-        #[cfg(any(
-            all(
-                feature = "clmul",
-                target_arch = "x86_64",
-                target_feature = "sse2",
-                target_feature = "pclmulqdq"
-            ),
-            all(
-                feature = "clmul",
-                target_arch = "aarch64",
-                target_feature = "neon",
-                target_feature = "aes"
-            )
+        #[cfg(all(
+            feature = "clmul",
+            target_arch = "x86_64",
+            target_feature = "sse2",
+            target_feature = "pclmulqdq"
         ))]
         if W::NBITS == 64 {
             // Safety: W == u64 when NBITS == 64.
@@ -696,6 +679,21 @@ impl<W: Word, const NWORDS: usize, const A: usize, const B: usize, const C: usiz
             let result: &Self = unsafe { std::mem::transmute(&tmp) };
             return *result;
         }
+        #[cfg(all(feature = "clmul", target_arch = "aarch64"))]
+        if W::NBITS == 64
+            && std::arch::is_aarch64_feature_detected!("neon")
+            && std::arch::is_aarch64_feature_detected!("aes")
+        {
+            // Safety: W == u64 when NBITS == 64.
+            let x: &GF2n<u64, NWORDS, A, B, C> = unsafe { std::mem::transmute(&self) };
+            // Safety: W == u64 when NBITS == 64.
+            let y: &GF2n<u64, NWORDS, A, B, C> = unsafe { std::mem::transmute(other) };
+            // Safety: target_feature's "neon" and "aes" are available in this block.
+            let tmp: GF2n<u64, NWORDS, A, B, C> = unsafe { mul_clmul_u64(x, y) };
+            // Safety: W == u64 when NBITS == 64.
+            let result: &Self = unsafe { std::mem::transmute(&tmp) };
+            return *result;
+        }
         self.mul_as_add(other)
     }
 }