cuda-cccl 0.1.3.2.0.dev438__cp313-cp313-manylinux_2_24_aarch64.whl → 0.3.1__cp313-cp313-manylinux_2_24_aarch64.whl

cuda/cccl/headers/include/cuda/std/__floating_point/cast.h

@@ -76,25 +76,25 @@ template <class _To, class _From>
 #if _CCCL_HAS_NVFP8_E8M0()
     else if constexpr (is_same_v<_To, __nv_fp8_e8m0>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp8_e8m0>(::__nv_cvt_float_to_e8m0(__v, __NV_NOSAT, cudaRoundZero));
+      return ::cuda::std::__fp_from_storage<__nv_fp8_e8m0>(::__nv_cvt_float_to_e8m0(__v, __NV_NOSAT, ::cudaRoundZero));
     }
 #endif // _CCCL_HAS_NVFP8_E8M0()
 #if _CCCL_HAS_NVFP6_E2M3()
     else if constexpr (is_same_v<_To, __nv_fp6_e2m3>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp6_e2m3>(::__nv_cvt_float_to_fp6(__v, __NV_E2M3, cudaRoundNearest));
+      return ::cuda::std::__fp_from_storage<__nv_fp6_e2m3>(::__nv_cvt_float_to_fp6(__v, __NV_E2M3, ::cudaRoundNearest));
     }
 #endif // _CCCL_HAS_NVFP6_E2M3()
 #if _CCCL_HAS_NVFP6_E3M2()
     else if constexpr (is_same_v<_To, __nv_fp6_e3m2>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp6_e3m2>(::__nv_cvt_float_to_fp6(__v, __NV_E3M2, cudaRoundNearest));
+      return ::cuda::std::__fp_from_storage<__nv_fp6_e3m2>(::__nv_cvt_float_to_fp6(__v, __NV_E3M2, ::cudaRoundNearest));
     }
 #endif // _CCCL_HAS_NVFP6_E3M2()
 #if _CCCL_HAS_NVFP4_E2M1()
     else if constexpr (is_same_v<_To, __nv_fp4_e2m1>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp4_e2m1>(::__nv_cvt_float_to_fp4(__v, __NV_E2M1, cudaRoundNearest));
+      return ::cuda::std::__fp_from_storage<__nv_fp4_e2m1>(::__nv_cvt_float_to_fp4(__v, __NV_E2M1, ::cudaRoundNearest));
     }
 #endif // _CCCL_HAS_NVFP4_E2M1()
     else
@@ -145,25 +145,28 @@ template <class _To, class _From>
 #if _CCCL_HAS_NVFP8_E8M0()
     else if constexpr (is_same_v<_To, __nv_fp8_e8m0>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp8_e8m0>(::__nv_cvt_double_to_e8m0(__v, __NV_NOSAT, cudaRoundZero));
+      return ::cuda::std::__fp_from_storage<__nv_fp8_e8m0>(::__nv_cvt_double_to_e8m0(__v, __NV_NOSAT, ::cudaRoundZero));
     }
 #endif // _CCCL_HAS_NVFP8_E8M0()
 #if _CCCL_HAS_NVFP6_E2M3()
     else if constexpr (is_same_v<_To, __nv_fp6_e2m3>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp6_e2m3>(::__nv_cvt_double_to_fp6(__v, __NV_E2M3, cudaRoundNearest));
+      return ::cuda::std::__fp_from_storage<__nv_fp6_e2m3>(
+        ::__nv_cvt_double_to_fp6(__v, __NV_E2M3, ::cudaRoundNearest));
     }
 #endif // _CCCL_HAS_NVFP6_E2M3()
 #if _CCCL_HAS_NVFP6_E3M2()
     else if constexpr (is_same_v<_To, __nv_fp6_e3m2>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp6_e3m2>(::__nv_cvt_double_to_fp6(__v, __NV_E3M2, cudaRoundNearest));
+      return ::cuda::std::__fp_from_storage<__nv_fp6_e3m2>(
+        ::__nv_cvt_double_to_fp6(__v, __NV_E3M2, ::cudaRoundNearest));
     }
 #endif // _CCCL_HAS_NVFP6_E3M2()
 #if _CCCL_HAS_NVFP4_E2M1()
     else if constexpr (is_same_v<_To, __nv_fp4_e2m1>)
     {
-      return ::cuda::std::__fp_from_storage<__nv_fp4_e2m1>(::__nv_cvt_double_to_fp4(__v, __NV_E2M1, cudaRoundNearest));
+      return ::cuda::std::__fp_from_storage<__nv_fp4_e2m1>(
+        ::__nv_cvt_double_to_fp4(__v, __NV_E2M1, ::cudaRoundNearest));
     }
 #endif // _CCCL_HAS_NVFP4_E2M1()
     else
@@ -352,28 +355,28 @@ template <class _To, class _From>
     else if constexpr (is_same_v<_To, __nv_fp8_e8m0>)
     {
       return ::cuda::std::__fp_from_storage<__nv_fp8_e8m0>(
-        ::__nv_cvt_bfloat16raw_to_e8m0(__v, __NV_NOSAT, cudaRoundZero));
+        ::__nv_cvt_bfloat16raw_to_e8m0(__v, __NV_NOSAT, ::cudaRoundZero));
     }
 #  endif // _CCCL_HAS_NVFP8_E8M0()
 #  if _CCCL_HAS_NVFP6_E2M3()
     else if constexpr (is_same_v<_To, __nv_fp6_e2m3>)
     {
       return ::cuda::std::__fp_from_storage<__nv_fp6_e2m3>(
-        ::__nv_cvt_bfloat16raw_to_fp6(__v, __NV_E2M3, cudaRoundNearest));
+        ::__nv_cvt_bfloat16raw_to_fp6(__v, __NV_E2M3, ::cudaRoundNearest));
     }
 #  endif // _CCCL_HAS_NVFP6_E2M3()
 #  if _CCCL_HAS_NVFP6_E3M2()
     else if constexpr (is_same_v<_To, __nv_fp6_e3m2>)
     {
       return ::cuda::std::__fp_from_storage<__nv_fp6_e3m2>(
-        ::__nv_cvt_bfloat16raw_to_fp6(__v, __NV_E3M2, cudaRoundNearest));
+        ::__nv_cvt_bfloat16raw_to_fp6(__v, __NV_E3M2, ::cudaRoundNearest));
     }
 #  endif // _CCCL_HAS_NVFP6_E3M2()
 #  if _CCCL_HAS_NVFP4_E2M1()
     else if constexpr (is_same_v<_To, __nv_fp4_e2m1>)
     {
       return ::cuda::std::__fp_from_storage<__nv_fp4_e2m1>(
-        ::__nv_cvt_bfloat16raw_to_fp4(__v, __NV_E2M1, cudaRoundNearest));
+        ::__nv_cvt_bfloat16raw_to_fp4(__v, __NV_E2M1, ::cudaRoundNearest));
     }
 #  endif // _CCCL_HAS_NVFP4_E2M1()
     else

cuda/cccl/headers/include/cuda/std/__floating_point/cuda_fp_types.h

@@ -55,6 +55,9 @@ _CCCL_DIAG_SUPPRESS_MSVC(4100) // unreferenced formal parameter
 _CCCL_DIAG_POP
 #endif // _CCCL_HAS_NVFP4()
 
+// crt/device_fp128_functions.h is available in CUDA 12.8+.
+// _CCCL_HAS_FLOAT128() checks the *compiler* compatibility with __float128.
+// We also need to check the toolkit version to ensure the compatibility with nvc++.
 #if _CCCL_HAS_FLOAT128() && _CCCL_DEVICE_COMPILATION() && _CCCL_CTK_AT_LEAST(12, 8)
 #  if !_CCCL_COMPILER(NVRTC)
 _CCCL_DIAG_PUSH

cuda/cccl/headers/include/cuda/std/__floating_point/fp.h

@@ -11,7 +11,7 @@
 #ifndef _CUDA_STD___FLOATING_POINT_FP_H
 #define _CUDA_STD___FLOATING_POINT_FP_H
 
-#include <cuda/__cccl_config>
+#include <cuda/std/detail/__config>
 
 #if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
 #  pragma GCC system_header

cuda/cccl/headers/include/cuda/std/__mdspan/mdspan.h

@@ -439,7 +439,8 @@ public:
   [[nodiscard]] _CCCL_API constexpr bool is_exhaustive() const
     noexcept(noexcept(::cuda::std::declval<const mapping_type&>().is_exhaustive()))
   {
-    return mapping().is_exhaustive();
+    auto __tmp = mapping(); // workaround for clang with nodiscard
+    return __tmp.is_exhaustive();
   }
   [[nodiscard]] _CCCL_API constexpr bool is_strided() const
     noexcept(noexcept(::cuda::std::declval<const mapping_type&>().is_strided()))

cuda/cccl/headers/include/cuda/std/__tuple_dir/make_tuple_types.h

@@ -20,7 +20,9 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/__fwd/complex.h>
 #include <cuda/std/__fwd/array.h>
+#include <cuda/std/__fwd/complex.h>
 #include <cuda/std/__fwd/tuple.h>
 #include <cuda/std/__tuple_dir/tuple_element.h>
 #include <cuda/std/__tuple_dir/tuple_indices.h>
@@ -61,7 +63,27 @@ struct __make_tuple_types_flat<array<_Vt, _Np>, __tuple_indices<_Idx...>>
   template <size_t>
   using __value_type = _Vt;
   template <class _Tp, class _ApplyFn = __apply_cvref_fn<_Tp>>
-  using __apply_quals = __tuple_types<__type_call<_ApplyFn, __value_type<_Idx>>...>;
+  using __apply_quals _CCCL_NODEBUG_ALIAS = __tuple_types<__type_call<_ApplyFn, __value_type<_Idx>>...>;
+};
+
+template <class _Vt, size_t... _Idx>
+struct __make_tuple_types_flat<complex<_Vt>, __tuple_indices<_Idx...>>
+{
+  static_assert(sizeof...(_Idx) == 2, "__make_tuple_types: complex has only 2 members");
+  template <size_t>
+  using __value_type = _Vt;
+  template <class _Tp, class _ApplyFn = __apply_cvref_fn<_Tp>>
+  using __apply_quals _CCCL_NODEBUG_ALIAS = __tuple_types<__type_call<_ApplyFn, __value_type<_Idx>>...>;
+};
+
+template <class _Vt, size_t... _Idx>
+struct __make_tuple_types_flat<::cuda::complex<_Vt>, __tuple_indices<_Idx...>>
+{
+  static_assert(sizeof...(_Idx) == 2, "__make_tuple_types: complex has only 2 members");
+  template <size_t>
+  using __value_type = _Vt;
+  template <class _Tp, class _ApplyFn = __apply_cvref_fn<_Tp>>
+  using __apply_quals _CCCL_NODEBUG_ALIAS = __tuple_types<__type_call<_ApplyFn, __value_type<_Idx>>...>;
 };
 
 template <class _Tp,

cuda/cccl/headers/include/cuda/std/__tuple_dir/tuple_like.h

@@ -20,6 +20,7 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/__fwd/complex.h>
 #include <cuda/std/__concepts/concept_macros.h>
 #include <cuda/std/__fwd/array.h>
 #include <cuda/std/__fwd/complex.h>
@@ -58,6 +59,9 @@ inline constexpr bool __tuple_like_impl<array<_Tp, _Size>> = true;
 template <class _Tp>
 inline constexpr bool __tuple_like_impl<complex<_Tp>> = true;
 
+template <class _Tp>
+inline constexpr bool __tuple_like_impl<::cuda::complex<_Tp>> = true;
+
 template <class _Ip, class _Sp, ::cuda::std::ranges::subrange_kind _Kp>
 inline constexpr bool __tuple_like_impl<::cuda::std::ranges::subrange<_Ip, _Sp, _Kp>> = true;
 

cuda/cccl/headers/include/cuda/std/__tuple_dir/tuple_like_ext.h

@@ -20,6 +20,7 @@
 #  pragma system_header
 #endif // no system header
 
+#include <cuda/__fwd/complex.h>
 #include <cuda/std/__fwd/array.h>
 #include <cuda/std/__fwd/complex.h>
 #include <cuda/std/__fwd/pair.h>
@@ -54,6 +55,9 @@ inline constexpr bool __tuple_like_ext<array<_Tp, _Size>> = true;
 template <class _Tp>
 inline constexpr bool __tuple_like_ext<complex<_Tp>> = true;
 
+template <class _Tp>
+inline constexpr bool __tuple_like_ext<::cuda::complex<_Tp>> = true;
+
 template <class... _Tp>
 inline constexpr bool __tuple_like_ext<__tuple_types<_Tp...>> = true;
 

cuda/cccl/headers/include/cuda/std/__type_traits/promote.h

@@ -20,10 +20,8 @@
 #  pragma system_header
 #endif // no system header
 
-#include <cuda/std/__type_traits/integral_constant.h>
 #include <cuda/std/__type_traits/is_same.h>
 #include <cuda/std/__utility/declval.h>
-#include <cuda/std/cstddef>
 
 #include <cuda/std/__cccl/prologue.h>
 
@@ -49,6 +47,9 @@ struct __numeric_type
   _CCCL_API inline static double __test(unsigned long long);
   _CCCL_API inline static double __test(double);
   _CCCL_API inline static long double __test(long double);
+#if _CCCL_HAS_FLOAT128()
+  _CCCL_API inline static __float128 __test(__float128);
+#endif // _CCCL_HAS_FLOAT128()
 
   using type              = decltype(__test(declval<_Tp>()));
   static const bool value = !is_same_v<type, void>;

cuda/cccl/headers/include/cuda/std/string_view

@@ -57,7 +57,7 @@
 #include <cuda/std/version>
 
 #if !_CCCL_COMPILER(NVRTC)
-#  include <iosfwd>
+#  include <string_view>
 #endif // !_CCCL_COMPILER(NVRTC)
 
 #include <cuda/std/__cccl/prologue.h>
@@ -727,14 +727,21 @@ _CCCL_HOST_DEVICE basic_string_view(_Range&&) -> basic_string_view<::cuda::std::
 
 // operator <<
 
-#if 0 // todo: we need to implement char_traits stream types & functions
+#if !_CCCL_COMPILER(NVRTC)
+template <class _CharT>
+_CCCL_HOST_API ::std::basic_ostream<_CharT>&
+operator<<(::std::basic_ostream<_CharT>& __os, basic_string_view<_CharT> __str)
+{
+  return __os << ::std::basic_string_view<_CharT>{__str.data(), __str.size()};
+}
+
 template <class _CharT, class _Traits>
-_CCCL_API inline ::std::basic_ostream<_CharT, _Traits>&
+_CCCL_HOST_API ::std::basic_ostream<_CharT, _Traits>&
 operator<<(::std::basic_ostream<_CharT, _Traits>& __os, basic_string_view<_CharT, _Traits> __str)
 {
-  return __os.write(__str.data(), static_cast<::std::streamsize>(__str.size()));
+  return __os << ::std::basic_string_view<_CharT, _Traits>{__str.data(), __str.size()};
 }
-#endif // 0
+#endif // !_CCCL_COMPILER(NVRTC)
 
 // literals
 

cuda/cccl/headers/include/cuda/std/version

@@ -141,7 +141,7 @@
 // #   define __cccl_lib_shared_mutex                       201505L
 // # define __cccl_lib_shared_ptr_arrays                    201611L
 // # define __cccl_lib_shared_ptr_weak_type                 201606L
-// # define __cccl_lib_string_view                          201606L
+#define __cccl_lib_string_view 201803L
 // # define __cccl_lib_to_chars                             201611L
 // #  define __cccl_lib_uncaught_exceptions           201411L
 // #  define __cccl_lib_unordered_map_try_emplace     201411L
@@ -171,7 +171,6 @@
 // # define __cccl_lib_constexpr_misc                       201811L
 // # define __cccl_lib_constexpr_numeric                    201911L
 // # define __cccl_lib_constexpr_string                     201907L
-// # define __cccl_lib_constexpr_string_view                201811L
 // # define __cccl_lib_constexpr_swap_algorithms            201806L
 // # define __cccl_lib_constexpr_tuple                      201811L
 // # define __cccl_lib_constexpr_utility                    201811L
@@ -204,8 +203,6 @@
 // # define __cccl_lib_source_location                      201907L
 // # define __cccl_lib_ssize                                201902L
 // # define __cccl_lib_starts_ends_with                     201711L
-// # undef  __cccl_lib_string_view
-// # define __cccl_lib_string_view                          201803L
 // # define __cccl_lib_syncbuf                              201803L
 // # define __cccl_lib_three_way_comparison                 201907L
 #  define __cccl_lib_unwrap_ref 201811L

cuda/cccl/headers/include/thrust/detail/integer_math.h

@@ -27,6 +27,8 @@
 #endif // no system header
 #include <thrust/detail/type_deduction.h>
 
+#include <cuda/std/__bit/countl.h>
+#include <cuda/std/__type_traits/make_unsigned.h>
 #include <cuda/std/limits>
 #include <cuda/std/type_traits>
 
@@ -36,25 +38,6 @@ THRUST_NAMESPACE_BEGIN
 namespace detail
 {
 
-template <typename Integer>
-_CCCL_HOST_DEVICE _CCCL_FORCEINLINE Integer clz(Integer x)
-{
-  Integer result;
-
-  NV_IF_TARGET(NV_IS_DEVICE,
-               (result = ::__clz(x);),
-               (int num_bits = 8 * sizeof(Integer); int num_bits_minus_one = num_bits - 1; result = num_bits;
-                for (int i = num_bits_minus_one; i >= 0; --i) {
-                  if ((Integer(1) << i) & x)
-                  {
-                    result = num_bits_minus_one - i;
-                    break;
-                  }
-                }));
-
-  return result;
-}
-
 template <typename Integer>
 _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool is_power_of_2(Integer x)
 {
@@ -85,7 +68,7 @@ _CCCL_HOST_DEVICE _CCCL_FORCEINLINE Integer log2(Integer x)
   Integer num_bits           = 8 * sizeof(Integer);
   Integer num_bits_minus_one = num_bits - 1;
 
-  return num_bits_minus_one - clz(x);
+  return num_bits_minus_one - ::cuda::std::countl_zero(::cuda::std::__to_unsigned_like(x));
 }
 
 template <typename Integer>

cuda/cccl/headers/include/thrust/iterator/iterator_traits.h

@@ -316,6 +316,17 @@ struct iterator_traversal<::cuda::zip_iterator<Iterators...>>
   using type = detail::minimum_type<iterator_traversal_t<Iterators>...>;
 };
 
+template <class Fn, class... Iterators>
+struct iterator_system<::cuda::zip_transform_iterator<Fn, Iterators...>>
+{
+  using type = detail::minimum_system_t<iterator_system_t<Iterators>...>;
+};
+template <class Fn, class... Iterators>
+struct iterator_traversal<::cuda::zip_transform_iterator<Fn, Iterators...>>
+{
+  using type = detail::minimum_type<iterator_traversal_t<Iterators>...>;
+};
+
 //! \} // end iterator_traits
 
 THRUST_NAMESPACE_END

cuda/cccl/headers/include/thrust/system/cuda/detail/copy.h

@@ -48,6 +48,13 @@
 #include <thrust/system/cuda/detail/util.h>
 #include <thrust/type_traits/is_trivially_relocatable.h>
 
+#if _CCCL_HAS_CUDA_COMPILER()
+#  include <cub/device/dispatch/tuning/tuning_transform.cuh>
+#endif // _CCCL_HAS_CUDA_COMPILER()
+
+#include <cuda/__fwd/zip_iterator.h>
+#include <cuda/std/tuple>
+
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
@@ -61,6 +68,21 @@ template <class Derived, class InputIt, class OutputIt, class TransformOp>
 OutputIt _CCCL_API _CCCL_FORCEINLINE
 transform(execution_policy<Derived>& policy, InputIt first, InputIt last, OutputIt result, TransformOp transform_op);
 
+// Forward declare to work around a cyclic include, since "cuda/detail/transform.h" includes this header
+// We want this to unwrap zip_transform_iterator
+namespace __transform
+{
+_CCCL_EXEC_CHECK_DISABLE
+template <class Derived, class Offset, class... InputIts, class OutputIt, class TransformOp, class Predicate>
+OutputIt _CCCL_API _CCCL_FORCEINLINE cub_transform_many(
+  execution_policy<Derived>& policy,
+  ::cuda::std::tuple<InputIts...> firsts,
+  OutputIt result,
+  Offset num_items,
+  TransformOp transform_op,
+  Predicate pred);
+} // namespace __transform
+
 namespace __copy
 {
 template <class H, class D, class T, class Size>
@@ -190,6 +212,17 @@ device_to_device(execution_policy<Derived>& policy, InputIt first, InputIt last,
 
     return result + n;
   }
+  else if constexpr (::cuda::__is_zip_transform_iterator<InputIt>)
+  {
+    const auto n = ::cuda::std::distance(first, last);
+    return cuda_cub::__transform::cub_transform_many(
+      policy,
+      ::cuda::std::move(first).__base(),
+      result,
+      n,
+      ::cuda::std::move(first).__pred(),
+      cub::detail::transform::always_true_predicate{});
+  }
   else
   {
     return cuda_cub::transform(

cuda/cccl/headers/include/thrust/system/cuda/detail/tabulate.h

@@ -39,37 +39,23 @@
 #if _CCCL_HAS_CUDA_COMPILER()
 #  include <thrust/system/cuda/config.h>
 
-#  include <thrust/distance.h>
-#  include <thrust/system/cuda/detail/parallel_for.h>
+#  include <thrust/system/cuda/detail/transform.h>
 #  include <thrust/system/cuda/execution_policy.h>
 
+#  include <cuda/__functional/address_stability.h>
+#  include <cuda/std/iterator>
+
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-namespace __tabulate
-{
-template <class Iterator, class TabulateOp>
-struct functor
-{
-  Iterator items;
-  TabulateOp op;
-
-  template <typename Size>
-  void _CCCL_DEVICE operator()(Size idx)
-  {
-    items[idx] = op(idx);
-  }
-};
-} // namespace __tabulate
-
 template <class Derived, class Iterator, class TabulateOp>
 void _CCCL_HOST_DEVICE tabulate(execution_policy<Derived>& policy, Iterator first, Iterator last, TabulateOp tabulate_op)
 {
-  using size_type = thrust::detail::it_difference_t<Iterator>;
-  size_type count = ::cuda::std::distance(first, last);
-  cuda_cub::parallel_for(policy, __tabulate::functor<Iterator, TabulateOp>{first, tabulate_op}, count);
+  using size_type  = ::cuda::std::iter_difference_t<Iterator>;
+  const auto count = ::cuda::std::distance(first, last);
+  cuda_cub::transform_n(
+    policy, ::cuda::counting_iterator<size_type>{}, count, first, ::cuda::proclaim_copyable_arguments(tabulate_op));
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

cuda/cccl/headers/include/thrust/type_traits/unwrap_contiguous_iterator.h

@@ -25,72 +25,39 @@
 
 THRUST_NAMESPACE_BEGIN
 
-namespace detail
-{
-// Type traits for contiguous iterators:
-template <typename Iterator>
-struct contiguous_iterator_traits
-{
-  static_assert(thrust::is_contiguous_iterator_v<Iterator>,
-                "contiguous_iterator_traits requires a contiguous iterator.");
-
-  using raw_pointer =
-    typename thrust::detail::pointer_traits<decltype(&*::cuda::std::declval<Iterator>())>::raw_pointer;
-};
-} // namespace detail
-
-//! Converts a contiguous iterator type to its underlying raw pointer type.
-template <typename ContiguousIterator>
-using unwrap_contiguous_iterator_t = typename detail::contiguous_iterator_traits<ContiguousIterator>::raw_pointer;
-
 //! Converts a contiguous iterator to its underlying raw pointer.
+_CCCL_EXEC_CHECK_DISABLE
 template <typename ContiguousIterator>
 _CCCL_HOST_DEVICE auto unwrap_contiguous_iterator(ContiguousIterator it)
-  -> unwrap_contiguous_iterator_t<ContiguousIterator>
 {
   static_assert(thrust::is_contiguous_iterator_v<ContiguousIterator>,
                 "unwrap_contiguous_iterator called with non-contiguous iterator.");
   return thrust::raw_pointer_cast(&*it);
 }
 
-namespace detail
-{
-// Implementation for non-contiguous iterators -- passthrough.
-template <typename Iterator, bool IsContiguous = thrust::is_contiguous_iterator_v<Iterator>>
-struct try_unwrap_contiguous_iterator_impl
-{
-  using type = Iterator;
-
-  static _CCCL_HOST_DEVICE type get(Iterator it)
-  {
-    return it;
-  }
-};
+//! Converts a contiguous iterator type to its underlying raw pointer type.
+template <typename ContiguousIterator>
+using unwrap_contiguous_iterator_t = decltype(unwrap_contiguous_iterator(::cuda::std::declval<ContiguousIterator>()));
 
-// Implementation for contiguous iterators -- unwraps to raw pointer.
+//! Takes an iterator and, if it is contiguous, unwraps it to the raw pointer it represents. Otherwise returns the
+//! iterator unmodified.
+_CCCL_EXEC_CHECK_DISABLE
 template <typename Iterator>
-struct try_unwrap_contiguous_iterator_impl<Iterator, true /*is_contiguous*/>
+_CCCL_HOST_DEVICE auto try_unwrap_contiguous_iterator(Iterator it)
 {
-  using type = unwrap_contiguous_iterator_t<Iterator>;
-
-  static _CCCL_HOST_DEVICE type get(Iterator it)
+  if constexpr (thrust::is_contiguous_iterator_v<Iterator>)
   {
     return unwrap_contiguous_iterator(it);
   }
-};
-} // namespace detail
+  else
+  {
+    return it;
+  }
+}
 
 //! Takes an iterator type and, if it is contiguous, yields the raw pointer type it represents. Otherwise returns the
 //! iterator type unmodified.
 template <typename Iterator>
-using try_unwrap_contiguous_iterator_t = typename detail::try_unwrap_contiguous_iterator_impl<Iterator>::type;
-
-//! Takes an iterator and, if it is contiguous, unwraps it to the raw pointer it represents. Otherwise returns the
-//! iterator unmodified.
-template <typename Iterator>
-_CCCL_HOST_DEVICE auto try_unwrap_contiguous_iterator(Iterator it) -> try_unwrap_contiguous_iterator_t<Iterator>
-{
-  return detail::try_unwrap_contiguous_iterator_impl<Iterator>::get(it);
-}
+using try_unwrap_contiguous_iterator_t = decltype(try_unwrap_contiguous_iterator(::cuda::std::declval<Iterator>()));
 
 THRUST_NAMESPACE_END

cuda/cccl/parallel/experimental/__init__.py

@@ -1,73 +1,24 @@
-# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+# Copyright (c) 2025, NVIDIA CORPORATION.
 #
-# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 
-from .algorithms import (
-    DoubleBuffer,
-    SortOrder,
-    binary_transform,
-    exclusive_scan,
-    histogram_even,
-    inclusive_scan,
-    make_binary_transform,
-    make_exclusive_scan,
-    make_histogram_even,
-    make_inclusive_scan,
-    make_merge_sort,
-    make_radix_sort,
-    make_reduce_into,
-    make_segmented_reduce,
-    make_unary_transform,
-    make_unique_by_key,
-    merge_sort,
-    radix_sort,
-    reduce_into,
-    segmented_reduce,
-    unary_transform,
-    unique_by_key,
-)
-from .iterators import (
-    CacheModifiedInputIterator,
-    ConstantIterator,
-    CountingIterator,
-    ReverseIterator,
-    TransformIterator,
-    TransformOutputIterator,
-    ZipIterator,
-)
-from .op import OpKind
-from .struct import gpu_struct
+# alias for backwards compatibility
 
-__all__ = [
-    "binary_transform",
-    "CacheModifiedInputIterator",
-    "ConstantIterator",
-    "CountingIterator",
-    "DoubleBuffer",
-    "exclusive_scan",
-    "gpu_struct",
-    "histogram_even",
-    "inclusive_scan",
-    "make_binary_transform",
-    "make_exclusive_scan",
-    "make_histogram_even",
-    "make_inclusive_scan",
-    "make_merge_sort",
-    "make_radix_sort",
-    "make_reduce_into",
-    "make_segmented_reduce",
-    "make_unary_transform",
-    "make_unique_by_key",
-    "merge_sort",
-    "OpKind",
-    "radix_sort",
-    "reduce_into",
-    "ReverseIterator",
-    "segmented_reduce",
-    "SortOrder",
-    "TransformIterator",
-    "TransformOutputIterator",
-    "unary_transform",
-    "unique_by_key",
-    "ZipIterator",
-]
+from warnings import warn
+
+from cuda.compute import *  # noqa: F403
+
+warn(
+    "The module cuda.cccl.parallel.experimental is deprecated. Use cuda.compute instead.",
+    FutureWarning,
+)