diff --git a/src/lib.rs b/src/lib.rs
index 0394f58..53b7393 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -82,6 +82,8 @@ use u256_impl::u256;
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 use pulp::cast;
 use pulp::u64x4;
+#[allow(unused_imports)]
+use pulp::u64x8;
 
 #[doc(hidden)]
 pub mod prime;
diff --git a/src/native32.rs b/src/native32.rs
index 715e717..85650d3 100644
--- a/src/native32.rs
+++ b/src/native32.rs
@@ -1,6 +1,8 @@
 use aligned_vec::avec;
 
 use pulp::u32x8;
+#[allow(unused_imports)]
+use pulp::{u32x16, u64x8};
 
 /// Negacyclic NTT plan for multiplying two 32bit polynomials.
 #[derive(Clone, Debug)]
diff --git a/src/native64.rs b/src/native64.rs
index 034a3e8..a45b36c 100644
--- a/src/native64.rs
+++ b/src/native64.rs
@@ -1,5 +1,7 @@
 use aligned_vec::avec;
 
+#[allow(unused_imports)]
+use pulp::{b8, u32x16, u64x8};
 use pulp::{i32x4, m64x4, u32x4, u32x8, u64x4};
 
 pub(crate) use crate::native32::mul_mod32;
@@ -438,8 +440,8 @@ fn reconstruct_32bit_01234_avx2(
     let sign = simd.cmp_gt_u32x8(v4, half_p4);
     let sign: [i32x4; 2] = pulp::cast(sign);
     // sign extend so that -1i32 becomes -1i64
-    let sign0: m64x4 = pulp::cast(simd.convert_i32x4_to_i64x4(sign[0]));
-    let sign1: m64x4 = pulp::cast(simd.convert_i32x4_to_i64x4(sign[1]));
+    let sign0: m64x4 = unsafe { core::mem::transmute(simd.convert_i32x4_to_i64x4(sign[0])) };
+    let sign1: m64x4 = unsafe { core::mem::transmute(simd.convert_i32x4_to_i64x4(sign[1])) };
 
     let v0: [u32x4; 2] = pulp::cast(v0);
     let v1: [u32x4; 2] = pulp::cast(v1);
diff --git a/src/native_binary32.rs b/src/native_binary32.rs
index 18fa72c..8047cb7 100644
--- a/src/native_binary32.rs
+++ b/src/native_binary32.rs
@@ -1,6 +1,8 @@
 use aligned_vec::avec;
 
 use pulp::u32x8;
+#[allow(unused_imports)]
+use pulp::{u32x16, u64x8};
 
 use crate::native32::mul_mod32;
 
diff --git a/src/native_binary64.rs b/src/native_binary64.rs
index eaa09d0..f8427dc 100644
--- a/src/native_binary64.rs
+++ b/src/native_binary64.rs
@@ -1,5 +1,7 @@
 use aligned_vec::avec;
 
+#[allow(unused_imports)]
+use pulp::{b8, u32x16, u64x8};
 use pulp::{i32x4, m64x4, u32x4, u32x8, u64x4};
 
 pub(crate) use crate::native32::mul_mod32;
@@ -109,8 +111,8 @@ fn reconstruct_32bit_012_avx2(
     let sign = simd.cmp_gt_u32x8(v2, half_p2);
     let sign: [i32x4; 2] = pulp::cast(sign);
     // sign extend so that -1i32 becomes -1i64
-    let sign0: m64x4 = pulp::cast(simd.convert_i32x4_to_i64x4(sign[0]));
-    let sign1: m64x4 = pulp::cast(simd.convert_i32x4_to_i64x4(sign[1]));
+    let sign0: m64x4 = unsafe { core::mem::transmute(simd.convert_i32x4_to_i64x4(sign[0])) };
+    let sign1: m64x4 = unsafe { core::mem::transmute(simd.convert_i32x4_to_i64x4(sign[1])) };
 
     let v0: [u32x4; 2] = pulp::cast(v0);
     let v1: [u32x4; 2] = pulp::cast(v1);
diff --git a/src/prime32.rs b/src/prime32.rs
index 9cd0ccc..a4a2486 100644
--- a/src/prime32.rs
+++ b/src/prime32.rs
@@ -8,6 +8,8 @@ use aligned_vec::{avec, ABox};
 
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 use pulp::cast;
+#[allow(unused_imports)]
+use pulp::u32x16;
 use pulp::{u32x4, u32x8};
 
 const RECURSION_THRESHOLD: usize = 2048;
diff --git a/src/prime32/generic.rs b/src/prime32/generic.rs
index b28322c..a45b1b7 100644
--- a/src/prime32/generic.rs
+++ b/src/prime32/generic.rs
@@ -5,6 +5,8 @@ use core::iter::zip;
 use pulp::u32x8;
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 use pulp::{as_arrays, as_arrays_mut, cast};
+#[allow(unused_imports)]
+use pulp::{b16, u32x16};
 
 #[inline(always)]
 pub(crate) fn add(p: u32, a: u32, b: u32) -> u32 {
diff --git a/src/prime32/less_than_30bit.rs b/src/prime32/less_than_30bit.rs
index 670633b..8ef083d 100644
--- a/src/prime32/less_than_30bit.rs
+++ b/src/prime32/less_than_30bit.rs
@@ -1,3 +1,5 @@
+#[allow(unused_imports)]
+use pulp::u32x16;
 use pulp::u32x8;
 
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
diff --git a/src/prime32/less_than_31bit.rs b/src/prime32/less_than_31bit.rs
index ef68039..a1eb796 100644
--- a/src/prime32/less_than_31bit.rs
+++ b/src/prime32/less_than_31bit.rs
@@ -1,3 +1,5 @@
+#[allow(unused_imports)]
+use pulp::u32x16;
 use pulp::u32x8;
 
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
diff --git a/src/prime32/shoup.rs b/src/prime32/shoup.rs
index 7bbb99b..f3265bb 100644
--- a/src/prime32/shoup.rs
+++ b/src/prime32/shoup.rs
@@ -3,6 +3,8 @@ use crate::Butterfly;
 use core::iter::zip;
 use pulp::cast;
 
+#[allow(unused_imports)]
+use pulp::u32x16;
 use pulp::u32x8;
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 use pulp::{as_arrays, as_arrays_mut};
diff --git a/src/prime64.rs b/src/prime64.rs
index d1447c9..3d03f44 100644
--- a/src/prime64.rs
+++ b/src/prime64.rs
@@ -4,6 +4,8 @@ use aligned_vec::{avec, ABox};
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 use pulp::cast;
 use pulp::u64x4;
+#[allow(unused_imports)]
+use pulp::u64x8;
 
 const RECURSION_THRESHOLD: usize = 1024;
 
diff --git a/src/prime64/generic_solinas.rs b/src/prime64/generic_solinas.rs
index 1f9ab7e..f696492 100644
--- a/src/prime64/generic_solinas.rs
+++ b/src/prime64/generic_solinas.rs
@@ -5,6 +5,8 @@ use core::{fmt::Debug, iter::zip};
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 use pulp::cast;
 use pulp::u64x4;
+#[allow(unused_imports)]
+use pulp::{b8, u64x8};
 
 pub(crate) trait PrimeModulus: Debug + Copy {
     type Div: Debug + Copy;
diff --git a/src/prime64/less_than_50bit.rs b/src/prime64/less_than_50bit.rs
index 39edbe9..b6c7c75 100644
--- a/src/prime64/less_than_50bit.rs
+++ b/src/prime64/less_than_50bit.rs
@@ -1,3 +1,5 @@
+use pulp::u64x8;
+
 #[inline(always)]
 pub(crate) fn fwd_butterfly_avx512(
     simd: crate::V4IFma,
diff --git a/src/prime64/less_than_51bit.rs b/src/prime64/less_than_51bit.rs
index 26c22ab..607ad8b 100644
--- a/src/prime64/less_than_51bit.rs
+++ b/src/prime64/less_than_51bit.rs
@@ -1,3 +1,5 @@
+use pulp::u64x8;
+
 #[inline(always)]
 pub(crate) fn fwd_butterfly_avx512(
     simd: crate::V4IFma,
diff --git a/src/prime64/less_than_62bit.rs b/src/prime64/less_than_62bit.rs
index d6837c8..a6258a8 100644
--- a/src/prime64/less_than_62bit.rs
+++ b/src/prime64/less_than_62bit.rs
@@ -1,4 +1,6 @@
 use pulp::u64x4;
+#[allow(unused_imports)]
+use pulp::u64x8;
 
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 #[cfg(feature = "nightly")]
diff --git a/src/prime64/less_than_63bit.rs b/src/prime64/less_than_63bit.rs
index aebeda9..c52b43f 100644
--- a/src/prime64/less_than_63bit.rs
+++ b/src/prime64/less_than_63bit.rs
@@ -1,4 +1,6 @@
 use pulp::u64x4;
+#[allow(unused_imports)]
+use pulp::u64x8;
 
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 #[cfg(feature = "nightly")]
diff --git a/src/prime64/shoup.rs b/src/prime64/shoup.rs
index e106a9c..6f06191 100644
--- a/src/prime64/shoup.rs
+++ b/src/prime64/shoup.rs
@@ -5,6 +5,8 @@ use core::iter::zip;
 use pulp::cast;
 
 use pulp::u64x4;
+#[allow(unused_imports)]
+use pulp::u64x8;
 
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
 #[cfg(feature = "nightly")]