diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -70,6 +70,10 @@
   __algorithm/pop_heap.h
   __algorithm/prev_permutation.h
   __algorithm/pstl_any_all_none_of.h
+  __algorithm/pstl_backend.h
+  __algorithm/pstl_backends/cpu_backend.h
+  __algorithm/pstl_backends/cpu_backends/serial.h
+  __algorithm/pstl_backends/cpu_backends/thread.h
   __algorithm/pstl_fill.h
   __algorithm/pstl_find.h
   __algorithm/pstl_for_each.h
diff --git a/libcxx/include/__algorithm/pstl_backend.h b/libcxx/include/__algorithm/pstl_backend.h
new file mode 100644
--- /dev/null
+++ b/libcxx/include/__algorithm/pstl_backend.h
@@ -0,0 +1,96 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ALGORITHM_PSTL_BACKEND_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKEND_H
+
+#include <__algorithm/pstl_backends/cpu_backend.h>
+#include <__config>
+#include <execution>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+#if !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+/*
+TODO: Documentation of how backends work
+
+A PSTL parallel backend is a tag type to which the following functions are associated, at minimum:
+
+  template <class _ExecutionPolicy, class _Iterator, class _Func>
+  void __for_each(_Backend, _ExecutionPolicy&&, _Iterator __first, _Iterator __last, _Func __f);
+
+  template <class _ExecutionPolicy, class _Iterator, class _Tp, class _BinOp>
+  _Tp __reduce(_Backend, _ExecutionPolicy&&, _Iterator __first, _Iterator __last, _Tp const& __value, _BinOp __op);
+
+  etc...
+
+The following functions are optional but can be provided. If provided, they are used by the corresponding
+algorithms, otherwise they are implemented in terms of the basis operations mentioned above:
+
+  template <class _ExecutionPolicy, class _Iterator, class _Size, class _Func>
+  void __for_each_n(_Backend, _ExecutionPolicy&&, _Iterator __first, _Size __n, _Func __f);
+
+  etc...
+
+*/
+
+void __pstl_for_each();
+void __pstl_for_each_n();
+
+template <class _ExecutionPolicy>
+struct __select_backend;
+
+template <>
+struct __select_backend<std::execution::sequenced_policy> {
+  using type = __cpu_backend_tag;
+};
+
+template <>
+struct __select_backend<std::execution::unsequenced_policy> {
+  using type = __cpu_backend_tag;
+};
+
+// TODO: Make this a proper customization point
+#define _PSTL_PAR_BACKEND_SERIAL
+
+#  if defined(_PSTL_PAR_BACKEND_STD_THREAD)
+#    include <__algorithm/pstl_backends/thread.h>
+template <> struct __select_backend<std::parallel_policy> { using type = __std_thread_backend; };
+template <> struct __select_backend<std::parallel_unsequenced_policy> { using type = __std_thread_backend; };
+
+#  elif defined(_PSTL_PAR_BACKEND_GCD)
+#    include <__algorithm/pstl_backends/gcd.h>
+template <> struct __select_backend<std::parallel_policy> { using type = __gcd_backend; };
+template <> struct __select_backend<std::parallel_unsequenced_policy> { using type = __gcd_backend; };
+
+#  elif defined(_PSTL_PAR_BACKEND_TBB)
+#    include <__algorithm/pstl_backends/tbb.h>
+template <> struct __select_backend<std::parallel_policy> { using type = __tbb_backend; };
+template <> struct __select_backend<std::parallel_unsequenced_policy> { using type = __tbb_backend; };
+
+#  elif defined(_PSTL_PAR_BACKEND_SERIAL)
+template <> struct __select_backend<std::execution::parallel_policy> { using type = __cpu_backend_tag; };
+template <> struct __select_backend<std::execution::parallel_unsequenced_policy> { using type = __cpu_backend_tag; };
+
+#  else
+
+// ...New vendors can add parallel backends here...
+
+#    error "Invalid choice of a PSTL parallel backend"
+#  endif
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && _LIBCPP_STD_VER >= 17
+
+#endif // _LIBCPP___ALGORITHM_PSTL_BACKEND_H
diff --git a/libcxx/include/__algorithm/pstl_backends/cpu_backend.h b/libcxx/include/__algorithm/pstl_backends/cpu_backend.h
new file mode 100644
--- /dev/null
+++ b/libcxx/include/__algorithm/pstl_backends/cpu_backend.h
@@ -0,0 +1,63 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKEND_BACKEND_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKEND_BACKEND_H
+
+#include <__config>
+
+/*
+  TODO: Document the parallel backend
+*/
+
+#ifdef _LIBCPP_HAS_NO_THREADS
+#  include <__algorithm/pstl_backends/cpu_backends/thread.h>
+#else
+#  include <__algorithm/pstl_backends/cpu_backends/serial.h>
+#endif
+
+#include <__algorithm/for_each.h>
+#include <__iterator/iterator_traits.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/terminate_on_exception.h>
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+struct __cpu_backend_tag {};
+
+template <class _Iterator, class _DifferenceType, class _Function>
+_LIBCPP_HIDE_FROM_ABI _Iterator __simd_walk_1(_Iterator __first, _DifferenceType __n, _Function __f) noexcept {
+  // _PSTL_PRAGMA_SIMD
+  for (_DifferenceType __i = 0; __i < __n; ++__i)
+    __f(__first[__i]);
+
+  return __first + __n;
+}
+
+template <class _ExecutionPolicy, class _ForwardIterator, class _Functor>
+void __pstl_for_each(__cpu_backend_tag, _ForwardIterator __first, _ForwardIterator __last, _Functor __func) {
+  if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+                __is_cpp17_random_access_iterator<_ForwardIterator>::value) {
+    std::__terminate_on_exception([&] {
+      std::__par_backend::__parallel_for(
+          __first, __last, [__func](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+            std::__pstl_for_each<__remove_parallel_policy_t<_ExecutionPolicy>>(
+                __cpu_backend_tag{}, __brick_first, __brick_last, __func);
+          });
+    });
+  } else if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
+                       __is_cpp17_random_access_iterator<_ForwardIterator>::value) {
+    std::__simd_walk_1(__first, __last - __first, __func);
+  } else {
+    std::for_each(__first, __last, __func);
+  }
+}
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKEND_BACKEND_H
diff --git a/libcxx/include/__algorithm/pstl_backends/cpu_backends/serial.h b/libcxx/include/__algorithm/pstl_backends/cpu_backends/serial.h
new file mode 100644
--- /dev/null
+++ b/libcxx/include/__algorithm/pstl_backends/cpu_backends/serial.h
@@ -0,0 +1,57 @@
+// -*- C++ -*-
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_SERIAL_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_SERIAL_H
+
+#include <__algorithm/for_each.h>
+#include <__algorithm/for_each_n.h>
+#include <__config>
+#include <__numeric/reduce.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+#if !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+namespace __par_backend::inline __serial_cpu_backend {
+
+//
+// Mandatory customization points
+//
+template <class _RandomAccessIterator, class _Fp>
+_LIBCPP_HIDE_FROM_ABI void __parallel_for(_RandomAccessIterator __first, _RandomAccessIterator __last, _Fp __f) {
+  __f(__first, __last);
+}
+
+template <class _RandomAccessIterator, class _Tp, class _BinOp>
+_LIBCPP_HIDE_FROM_ABI _Tp
+__parallel_reduce(_RandomAccessIterator __first, _RandomAccessIterator __last, _Tp const& __value, _BinOp __op) {
+  return std::reduce(__first, __last, __value, __op);
+}
+
+// etc...
+
+//
+// Optional customization points
+//
+template <class _Iterator, class _Size, class _Fp>
+_LIBCPP_HIDE_FROM_ABI void __pstl_for_each_n(_Iterator __first, _Size __n, _Fp __f) {
+  std::for_each_n(__first, __n, __f);
+}
+} // namespace __par_backend::__serial_cpu_backend
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && && _LIBCPP_STD_VER >= 17
+
+#endif // _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_SERIAL_H
diff --git a/libcxx/include/__algorithm/pstl_backends/cpu_backends/thread.h b/libcxx/include/__algorithm/pstl_backends/cpu_backends/thread.h
new file mode 100644
--- /dev/null
+++ b/libcxx/include/__algorithm/pstl_backends/cpu_backends/thread.h
@@ -0,0 +1,47 @@
+// -*- C++ -*-
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_THREAD_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_THREAD_H
+
+#include <__algorithm/for_each.h>
+#include <__config>
+#include <__numeric/reduce.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+#if !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+namespace __par_backend::inline __thread_cpu_backend {
+
+template <class _RandomAccessIterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI void
+__parallel_for(_RandomAccessIterator __first, _RandomAccessIterator __last, _Functor __func) {
+  // TODO: implement
+}
+
+template <class _ExecutionPolicy, class _RandomAccessIterator, class _Tp, class _BinOp>
+_LIBCPP_HIDE_FROM_ABI _Tp
+__parallel_reduce(_RandomAccessIterator __first, _RandomAccessIterator __last, _Tp const& __value, _BinOp __op) {
+  // TODO: Implement
+}
+
+// etc...
+
+} // namespace __par_backend::__thread_cpu_backend
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // !defined(_LIBCPP_HAS_NO_INCOMPLETE_PSTL) && _LIBCPP_STD_VER >= 17
+
+#endif // _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_THREAD_H
diff --git a/libcxx/include/__algorithm/pstl_backends/some_gpu_backend.h b/libcxx/include/__algorithm/pstl_backends/some_gpu_backend.h
new file mode 100644
--- /dev/null
+++ b/libcxx/include/__algorithm/pstl_backends/some_gpu_backend.h
@@ -0,0 +1,28 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ALGORITHM_PSTL_BACKENDS_SOME_GPU_BACKEND_H
+#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_SOME_GPU_BACKEND_H
+
+#include <__config>
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+struct __some_gpu_backend_tag {};
+
+// Mandatory algorithms
+template <class _ExecutionPolicy, class _ForwardIterator, class _Functor>
+void __pstl_for_each(__some_gpu_backend_tag, _ForwardIterator __first, _ForwardIterator __last, _Functor __func) {
+  // Implement
+}
+
+// ...
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP___ALGORITHM_PSTL_BACKENDS_SOME_GPU_BACKEND_H
diff --git a/libcxx/include/__algorithm/pstl_for_each.h b/libcxx/include/__algorithm/pstl_for_each.h
--- a/libcxx/include/__algorithm/pstl_for_each.h
+++ b/libcxx/include/__algorithm/pstl_for_each.h
@@ -11,12 +11,14 @@
 
 #include <__algorithm/for_each.h>
 #include <__algorithm/for_each_n.h>
+#include <__algorithm/pstl_backend.h>
 #include <__config>
 #include <__iterator/iterator_traits.h>
 #include <__pstl/internal/parallel_backend.h>
 #include <__pstl/internal/unseq_backend_simd.h>
 #include <__type_traits/is_execution_policy.h>
 #include <__type_traits/remove_cvref.h>
+#include <__type_traits/void_t.h>
 #include <__utility/terminate_on_exception.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
@@ -30,40 +32,33 @@
 template <class _ExecutionPolicy,
           class _ForwardIterator,
           class _Function,
-          enable_if_t<is_execution_policy_v<__remove_cvref_t<_ExecutionPolicy>>, int> = 0>
+          class _RawPolicy = __remove_cvref_t<_ExecutionPolicy>,
+          enable_if_t<is_execution_policy_v<_RawPolicy>, int> = 0>
 _LIBCPP_HIDE_FROM_ABI void
 for_each(_ExecutionPolicy&& __policy, _ForwardIterator __first, _ForwardIterator __last, _Function __func) {
-  if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
-                __is_cpp17_random_access_iterator<_ForwardIterator>::value) {
-    std::__terminate_on_exception([&] {
-      __pstl::__par_backend::__parallel_for(
-          {},
-          __policy,
-          __first,
-          __last,
-          [&__policy, __func](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
-            std::for_each(std::__remove_parallel_policy(__policy), __brick_first, __brick_last, __func);
-          });
-    });
-  } else if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
-                       __is_cpp17_random_access_iterator<_ForwardIterator>::value) {
-    __pstl::__unseq_backend::__simd_walk_1(__first, __last - __first, __func);
-  } else {
-    std::for_each(__first, __last, __func);
-  }
+  using _Backend = typename __select_backend<_RawPolicy>::type;
+  std::__pstl_for_each<_RawPolicy>(_Backend{}, std::move(__first), std::move(__last), std::move(__func));
 }
 
 template <class _ExecutionPolicy,
           class _ForwardIterator,
           class _Size,
           class _Function,
-          enable_if_t<is_execution_policy_v<__remove_cvref_t<_ExecutionPolicy>>, int> = 0>
+          class _RawPolicy = __remove_cvref_t<_ExecutionPolicy>,
+          enable_if_t<is_execution_policy_v<_RawPolicy>, int> = 0>
 _LIBCPP_HIDE_FROM_ABI void
 for_each_n(_ExecutionPolicy&& __policy, _ForwardIterator __first, _Size __size, _Function __func) {
-  if constexpr (__is_cpp17_random_access_iterator<_ForwardIterator>::value) {
-    std::for_each(__policy, __first, __first + __size, __func);
+  using _Backend = typename __select_backend<_RawPolicy>::type;
+  if constexpr (requires {std::__pstl_for_each_n(_Backend{}, __first, __size, __func); }) {
+    __pstl_for_each_n<_RawPolicy>(_Backend{}, std::move(__first), __size, std::move(__func));
   } else {
-    std::for_each_n(__first, __size, __func);
+    // Default implementation
+    if constexpr (__is_cpp17_random_access_iterator<_ForwardIterator>::value &&
+                  __is_parallel_execution_policy_v<_RawPolicy>) {
+      std::for_each(__policy, __first, __first + __size, __func);
+    } else {
+      std::for_each_n(__first, __size, __func);
+    }
   }
 }
 
diff --git a/libcxx/include/__type_traits/is_execution_policy.h b/libcxx/include/__type_traits/is_execution_policy.h
--- a/libcxx/include/__type_traits/is_execution_policy.h
+++ b/libcxx/include/__type_traits/is_execution_policy.h
@@ -36,10 +36,13 @@
 template <class _Tp>
 inline constexpr bool __is_parallel_execution_policy_v = __is_parallel_execution_policy_impl<__remove_cvref_t<_Tp>>;
 
+template <class _ExecutionPolicy>
+_LIBCPP_HIDE_FROM_ABI auto __remove_parallel_policy();
+
 // Removes the "parallel" part of an execution policy.
 // For example, turns par_unseq into unseq, and par into seq.
 template <class _ExecutionPolicy>
-_LIBCPP_HIDE_FROM_ABI const auto& __remove_parallel_policy(_ExecutionPolicy&&);
+using __remove_parallel_policy_t = decltype(std::__remove_parallel_policy<_ExecutionPolicy>());
 
 _LIBCPP_END_NAMESPACE_STD
 
diff --git a/libcxx/include/execution b/libcxx/include/execution
--- a/libcxx/include/execution
+++ b/libcxx/include/execution
@@ -135,12 +135,11 @@
 struct is_execution_policy : bool_constant<is_execution_policy_v<_Tp>> {};
 
 template <class _ExecutionPolicy>
-_LIBCPP_HIDE_FROM_ABI const auto& __remove_parallel_policy(_ExecutionPolicy&&) {
-  using _ExecPol = __remove_cvref_t<_ExecutionPolicy>;
-  if constexpr (is_same_v<_ExecPol, execution::parallel_policy>) {
-    return execution::seq;
-  } else if constexpr (is_same_v<_ExecPol, execution::parallel_unsequenced_policy>) {
-    return execution::__unseq;
+_LIBCPP_HIDE_FROM_ABI auto __remove_parallel_policy() {
+  if constexpr (is_same_v<_ExecutionPolicy, execution::parallel_policy>) {
+    return execution::sequenced_policy(execution::__disable_user_instantiations_tag{});
+  } else if constexpr (is_same_v<_ExecutionPolicy, execution::parallel_unsequenced_policy>) {
+    return execution::__unsequenced_policy{execution::__disable_user_instantiations_tag{}};
   }
 }