diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt
index 6a317600e..2e5a8926e 100644
--- a/src/core/CMakeLists.txt
+++ b/src/core/CMakeLists.txt
@@ -33,6 +33,7 @@ add_mp_units_module(
             include/mp-units/bits/core_gmf.h
             include/mp-units/bits/double_width_int.h
             include/mp-units/bits/fixed_point.h
+            include/mp-units/bits/int_power.h
             include/mp-units/bits/get_associated_quantity.h
             include/mp-units/bits/hacks.h
             include/mp-units/bits/module_macros.h
diff --git a/src/core/include/mp-units/bits/double_width_int.h b/src/core/include/mp-units/bits/double_width_int.h
index 1b822441c..fd4fd93fb 100644
--- a/src/core/include/mp-units/bits/double_width_int.h
+++ b/src/core/include/mp-units/bits/double_width_int.h
@@ -22,57 +22,24 @@
 
 #pragma once
 
-#include <mp-units/bits/hacks.h>       // IWYU pragma: keep
+#include <mp-units/bits/hacks.h>  // IWYU pragma: keep
+#include <mp-units/bits/int_power.h>
 #include <mp-units/ext/type_traits.h>  // IWYU pragma: keep
 
 #ifndef MP_UNITS_IN_MODULE_INTERFACE
 #ifdef MP_UNITS_IMPORT_STD
 import std;
 #else
-#include <algorithm>
-#include <bit>
 #include <concepts>
+#include <cstddef>
 #include <cstdint>
-#include <cstdlib>
 #include <limits>
-#include <tuple>
 #include <type_traits>
-#include <version>
-// <cmath> becomes partially freestanding in C++26
-// before that, there is no guarantee about this header even existing
-// (GCC 14 has it, but actively #error's out)
-#if __STDC_HOSTED__
-#include <cmath>
-#endif
 #endif
 #endif
 
 namespace mp_units::detail {
 
-template<typename T>
-constexpr std::size_t integer_rep_width_v = std::numeric_limits<std::make_unsigned_t<T>>::digits;
-
-template<std::floating_point T>
-[[nodiscard]] consteval T int_power(T base, int exponent)
-{
-#if __STDC_HOSTED__ && defined(__cpp_lib_constexpr_cmath) && __cpp_lib_constexpr_cmath >= 202202L
-  return std::ldexp(base, exponent);
-#else
-  if (exponent < 0) {
-    base = T{1} / base;
-    exponent = -exponent;
-  }
-  T ret = 1;
-  while (exponent) {
-    if (exponent & 1) ret *= base;
-    exponent >>= 1;
-    base *= base;
-  }
-  return ret;
-
-#endif
-}
-
 // A double-width integer synthesized from two base-width integers.  On platforms without a
 // native `__int128` (notably MSVC), `int128_t` / `uint128_t` are aliases for this template
 // instantiated on (u)int64_t; the operator set is therefore comprehensive enough for it to
diff --git a/src/core/include/mp-units/bits/fixed_point.h b/src/core/include/mp-units/bits/fixed_point.h
index 4376797ee..1b0383641 100644
--- a/src/core/include/mp-units/bits/fixed_point.h
+++ b/src/core/include/mp-units/bits/fixed_point.h
@@ -22,10 +22,16 @@
 
 #pragma once
 
-#include <mp-units/bits/double_width_int.h>
-#include <mp-units/bits/hacks.h>       // IWYU pragma: keep
+#include <mp-units/bits/hacks.h>  // IWYU pragma: keep
+#include <mp-units/bits/int_power.h>
 #include <mp-units/ext/type_traits.h>  // IWYU pragma: keep
 
+// `double_width_int` only emulates `int128_t` on platforms without native `__int128` (notably
+// MSVC).  On every other platform `int128_t` is the builtin and dwint is unused in library code
+#if !defined(__SIZEOF_INT128__)
+#include <mp-units/bits/double_width_int.h>
+#endif
+
 #ifndef MP_UNITS_IN_MODULE_INTERFACE
 #ifdef MP_UNITS_IMPORT_STD
 import std;
@@ -74,9 +80,6 @@ constexpr std::size_t max_native_width = 64;
 template<typename T>
 constexpr bool is_signed_v = std::is_signed_v<T>;
 
-template<typename T>
-constexpr bool is_signed_v<double_width_int<T>> = double_width_int<T>::is_signed;
-
 #if defined(__SIZEOF_INT128__)
 
 MP_UNITS_DIAGNOSTIC_PUSH
@@ -95,6 +98,13 @@ inline constexpr bool is_signed_v<unsigned __int128> = false;
 
 MP_UNITS_DIAGNOSTIC_POP
 
+#else
+
+// Only needed when `double_width_int` doubles as `int128_t`; the dwint header was included
+// for that purpose above.
+template<typename T>
+constexpr bool is_signed_v<double_width_int<T>> = double_width_int<T>::is_signed;
+
 #endif
 
 template<typename T>
diff --git a/src/core/include/mp-units/bits/int_power.h b/src/core/include/mp-units/bits/int_power.h
new file mode 100644
index 000000000..c245b3030
--- /dev/null
+++ b/src/core/include/mp-units/bits/int_power.h
@@ -0,0 +1,76 @@
+// The MIT License (MIT)
+//
+// Copyright (c) 2018 Mateusz Pusz
+//
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+//
+// The above copyright notice and this permission notice shall be included in all
+// copies or substantial portions of the Software.
+//
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+#pragma once
+
+#include <mp-units/bits/hacks.h>  // IWYU pragma: keep
+
+#ifndef MP_UNITS_IN_MODULE_INTERFACE
+#ifdef MP_UNITS_IMPORT_STD
+import std;
+#else
+#include <concepts>
+#include <cstddef>
+#include <limits>
+#include <type_traits>
+#include <version>
+// <cmath> becomes partially freestanding in C++26
+// before that, there is no guarantee about this header even existing
+// (GCC 14 has it, but actively #error's out)
+#if __STDC_HOSTED__
+#include <cmath>
+#endif
+#endif
+#endif
+
+namespace mp_units::detail {
+
+// Representation width (in bits) of an integral type, as the digit count of its unsigned
+// counterpart.  Lives in its own header so that both `double_width_int.h` and
+// `fixed_point.h` can use it without one having to include the other.
+template<typename T>
+constexpr std::size_t integer_rep_width_v = std::numeric_limits<std::make_unsigned_t<T>>::digits;
+
+// `2^exponent` (or more precisely `base * 2^exponent`) as a compile-time floating-point
+// constant.  Used by `double_width_int`'s long-double conversions and by `fixed_point`'s
+// long-double constructor.
+template<std::floating_point T>
+[[nodiscard]] consteval T int_power(T base, int exponent)
+{
+#if __STDC_HOSTED__ && defined(__cpp_lib_constexpr_cmath) && __cpp_lib_constexpr_cmath >= 202202L
+  return std::ldexp(base, exponent);
+#else
+  if (exponent < 0) {
+    base = T{1} / base;
+    exponent = -exponent;
+  }
+  T ret = 1;
+  while (exponent) {
+    if (exponent & 1) ret *= base;
+    exponent >>= 1;
+    base *= base;
+  }
+  return ret;
+#endif
+}
+
+}  // namespace mp_units::detail