diff --git a/src/non_zero.rs b/src/non_zero.rs
index fe91b1ee7..7ffd06bb2 100644
--- a/src/non_zero.rs
+++ b/src/non_zero.rs
@@ -84,6 +84,20 @@ where
 }
 
 impl NonZero<Limb> {
+    /// Creates a new non-zero limb in a const context.
+    /// Panics if the value is zero.
+    ///
+    /// In future versions of Rust it should be possible to replace this with
+    /// `NonZero::new(…).unwrap()`
+    // TODO: Remove when `Self::new` and `CtOption::unwrap` support `const fn`
+    pub const fn new_unwrap(n: Limb) -> Self {
+        if n.is_nonzero().is_true_vartime() {
+            Self(n)
+        } else {
+            panic!("Invalid value: zero")
+        }
+    }
+
     /// Create a [`NonZero<Limb>`] from a [`NonZeroU8`] (const-friendly)
     // TODO(tarcieri): replace with `const impl From<NonZeroU8>` when stable
     pub const fn from_u8(n: NonZeroU8) -> Self {
diff --git a/src/primitives.rs b/src/primitives.rs
index 5d79a5f6c..3a0ae58e3 100644
--- a/src/primitives.rs
+++ b/src/primitives.rs
@@ -65,9 +65,3 @@ pub(crate) const fn mac(a: Word, b: Word, c: Word, carry: Word) -> (Word, Word)
     let ret = a + (b * c) + carry;
     (ret as Word, (ret >> Word::BITS) as Word)
 }
-
-/// Computes `(a * b) % d`.
-#[inline(always)]
-pub(crate) const fn mul_rem(a: Word, b: Word, d: Word) -> Word {
-    ((a as WideWord * b as WideWord) % (d as WideWord)) as Word
-}
diff --git a/src/uint/boxed/mul_mod.rs b/src/uint/boxed/mul_mod.rs
index 8c6b7ee15..340bf76ff 100644
--- a/src/uint/boxed/mul_mod.rs
+++ b/src/uint/boxed/mul_mod.rs
@@ -1,9 +1,9 @@
 //! [`BoxedUint`] modular multiplication operations.
 
 use crate::{
+    div_limb::mul_rem,
     modular::{BoxedMontyForm, BoxedMontyParams},
-    primitives::mul_rem,
-    BoxedUint, Limb, MulMod, Odd, WideWord, Word,
+    BoxedUint, Limb, MulMod, NonZero, Odd, WideWord, Word,
 };
 
 impl BoxedUint {
@@ -42,7 +42,11 @@ impl BoxedUint {
         // We implicitly assume `LIMBS > 0`, because `Uint<0>` doesn't compile.
         // Still the case `LIMBS == 1` needs special handling.
         if self.nlimbs() == 1 {
-            let reduced = mul_rem(self.limbs[0].0, rhs.limbs[0].0, Word::MIN.wrapping_sub(c.0));
+            let reduced = mul_rem(
+                self.limbs[0],
+                rhs.limbs[0],
+                NonZero::<Limb>::new_unwrap(Limb(Word::MIN.wrapping_sub(c.0))),
+            );
             return Self::from(reduced);
         }
 
diff --git a/src/uint/div_limb.rs b/src/uint/div_limb.rs
index 28f876517..4f50032cc 100644
--- a/src/uint/div_limb.rs
+++ b/src/uint/div_limb.rs
@@ -280,6 +280,14 @@ pub(crate) const fn rem_limb_with_reciprocal_wide<const L: usize>(
     Limb(r >> reciprocal.shift)
 }
 
+/// Computes `(a * b) % d`.
+#[inline(always)]
+pub(crate) const fn mul_rem(a: Limb, b: Limb, d: NonZero<Limb>) -> Limb {
+    let rec = Reciprocal::new(d);
+    let (hi, lo) = mulhilo(a.0, b.0);
+    rem_limb_with_reciprocal(&Uint::from_words([lo, hi]), &rec)
+}
+
 #[cfg(test)]
 mod tests {
     use super::{div2by1, Reciprocal};
diff --git a/src/uint/mul_mod.rs b/src/uint/mul_mod.rs
index 103ef8d87..97dc1ee33 100644
--- a/src/uint/mul_mod.rs
+++ b/src/uint/mul_mod.rs
@@ -1,8 +1,8 @@
 //! [`Uint`] modular multiplication operations.
 
 use crate::{
+    div_limb::mul_rem,
     modular::{MontyForm, MontyParams},
-    primitives::mul_rem,
     Concat, Limb, MulMod, NonZero, Split, Uint, WideWord, Word,
 };
 
@@ -53,8 +53,12 @@ impl<const LIMBS: usize> Uint<LIMBS> {
         // We implicitly assume `LIMBS > 0`, because `Uint<0>` doesn't compile.
         // Still the case `LIMBS == 1` needs special handling.
         if LIMBS == 1 {
-            let reduced = mul_rem(self.limbs[0].0, rhs.limbs[0].0, Word::MIN.wrapping_sub(c.0));
-            return Self::from_word(reduced);
+            let reduced = mul_rem(
+                self.limbs[0],
+                rhs.limbs[0],
+                NonZero::<Limb>::new_unwrap(Limb(Word::MIN.wrapping_sub(c.0))),
+            );
+            return Self::from_word(reduced.0);
         }
 
         let (lo, hi) = self.split_mul(rhs);