diff --git a/include/cuco/detail/extent/extent.inl b/include/cuco/detail/extent/extent.inl
index c04f6c329..13f056a57 100644
--- a/include/cuco/detail/extent/extent.inl
+++ b/include/cuco/detail/extent/extent.inl
@@ -18,7 +18,7 @@
 
 #include <cuco/detail/error.hpp>
 #include <cuco/detail/prime.hpp>  // TODO move to detail/extent/
-#include <cuco/detail/utility/math.hpp>
+#include <cuco/detail/utility/math.cuh>
 #include <cuco/detail/utils.hpp>
 #include <cuco/utility/fast_int.cuh>
 
diff --git a/include/cuco/detail/prime.hpp b/include/cuco/detail/prime.hpp
index c788fa245..4ef0a35e3 100644
--- a/include/cuco/detail/prime.hpp
+++ b/include/cuco/detail/prime.hpp
@@ -16,7 +16,7 @@
 
 #pragma once
 
-#include <cuco/detail/utility/math.hpp>
+#include <cuco/detail/utility/math.cuh>
 
 #include <algorithm>
 #include <array>
diff --git a/include/cuco/detail/utility/cuda.hpp b/include/cuco/detail/utility/cuda.hpp
index 9a2c526d8..49445e02f 100644
--- a/include/cuco/detail/utility/cuda.hpp
+++ b/include/cuco/detail/utility/cuda.hpp
@@ -16,7 +16,7 @@
 #pragma once
 
 #include <cuco/detail/error.hpp>
-#include <cuco/detail/utility/math.hpp>
+#include <cuco/detail/utility/math.cuh>
 
 namespace cuco {
 namespace detail {
diff --git a/include/cuco/detail/utility/math.hpp b/include/cuco/detail/utility/math.cuh
similarity index 84%
rename from include/cuco/detail/utility/math.hpp
rename to include/cuco/detail/utility/math.cuh
index 47484d6ad..c2715f6fa 100644
--- a/include/cuco/detail/utility/math.hpp
+++ b/include/cuco/detail/utility/math.cuh
@@ -15,7 +15,7 @@
 
 #pragma once
 
-#include <type_traits>
+#include <cuda/std/type_traits>
 
 namespace cuco {
 namespace detail {
@@ -35,10 +35,10 @@ namespace detail {
  * @return Ceiling of the integer division
  */
 template <typename T, typename U>
-constexpr T int_div_ceil(T dividend, U divisor) noexcept
+__host__ __device__ constexpr T int_div_ceil(T dividend, U divisor) noexcept
 {
-  static_assert(std::is_integral_v<T>);
-  static_assert(std::is_integral_v<U>);
+  static_assert(cuda::std::is_integral_v<T>);
+  static_assert(cuda::std::is_integral_v<U>);
   return (dividend + divisor - 1) / divisor;
 }