cuda-cccl 0.3.2__cp313-cp313-manylinux_2_24_aarch64.whl → 0.3.4__cp313-cp313-manylinux_2_24_aarch64.whl

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

@@ -39,11 +39,12 @@
 #if _CCCL_HAS_CUDA_COMPILER()
 #  include <thrust/system/cuda/config.h>
 
-#  include <thrust/distance.h>
-#  include <thrust/iterator/zip_iterator.h>
 #  include <thrust/pair.h>
 #  include <thrust/system/cuda/detail/execution_policy.h>
-#  include <thrust/zip_function.h>
+
+#  include <cuda/__iterator/zip_function.h>
+#  include <cuda/__iterator/zip_iterator.h>
+#  include <cuda/std/__iterator/distance.h>
 
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
@@ -63,151 +64,16 @@ THRUST_NAMESPACE_END
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-namespace detail
-{
-template <class ValueType, class InputIt1, class InputIt2, class BinaryOp>
-struct transform_pair_of_input_iterators_t
-{
-  using self_t            = transform_pair_of_input_iterators_t;
-  using difference_type   = thrust::detail::it_difference_t<InputIt1>;
-  using value_type        = ValueType;
-  using pointer           = void;
-  using reference         = value_type;
-  using iterator_category = ::cuda::std::random_access_iterator_tag;
-
-  InputIt1 input1;
-  InputIt2 input2;
-  mutable BinaryOp op;
-
-  transform_pair_of_input_iterators_t() = default;
-
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE
-  transform_pair_of_input_iterators_t(InputIt1 input1_, InputIt2 input2_, BinaryOp op_)
-      : input1(input1_)
-      , input2(input2_)
-      , op(op_)
-  {}
-
-  transform_pair_of_input_iterators_t(const self_t&) = default;
-
-  // BinaryOp might not be copy assignable, such as when it is a lambda.
-  // Define an explicit copy assignment operator that doesn't try to assign it.
-  _CCCL_HOST_DEVICE self_t& operator=(const self_t& o)
-  {
-    input1 = o.input1;
-    input2 = o.input2;
-    return *this;
-  }
-
-  /// Postfix increment
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE self_t operator++(int)
-  {
-    self_t retval = *this;
-    ++input1;
-    ++input2;
-    return retval;
-  }
-
-  /// Prefix increment
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE self_t& operator++()
-  {
-    ++input1;
-    ++input2;
-    return *this;
-  }
-
-  /// Indirection
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE reference operator*() const
-  {
-    return op(*input1, *input2);
-  }
-  /// Indirection
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE reference operator*()
-  {
-    return op(*input1, *input2);
-  }
-
-  /// Addition
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE self_t operator+(difference_type n) const
-  {
-    return self_t(input1 + n, input2 + n, op);
-  }
-
-  /// Addition assignment
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE self_t& operator+=(difference_type n)
-  {
-    input1 += n;
-    input2 += n;
-    return *this;
-  }
-
-  /// Subtraction
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE self_t operator-(difference_type n) const
-  {
-    return self_t(input1 - n, input2 - n, op);
-  }
-
-  /// Subtraction assignment
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE self_t& operator-=(difference_type n)
-  {
-    input1 -= n;
-    input2 -= n;
-    return *this;
-  }
-
-  /// Distance
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE difference_type operator-(self_t other) const
-  {
-    return input1 - other.input1;
-  }
-
-  /// Array subscript
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE reference operator[](difference_type n) const
-  {
-    return op(input1[n], input2[n]);
-  }
-
-  /// Equal to
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool operator==(const self_t& rhs) const
-  {
-    return (input1 == rhs.input1) && (input2 == rhs.input2);
-  }
-
-  /// Not equal to
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool operator!=(const self_t& rhs) const
-  {
-    return (input1 != rhs.input1) || (input2 != rhs.input2);
-  }
-
-  _CCCL_HOST_DEVICE _CCCL_FORCEINLINE bool operator<(const self_t& rhs) const
-  {
-    return input1 < rhs.input1;
-  }
-}; // struct transform_pair_of_input_iterators_t
-} // namespace detail
-
 template <class Derived, class InputIt1, class InputIt2, class BinaryPred>
 pair<InputIt1, InputIt2> _CCCL_HOST_DEVICE
 mismatch(execution_policy<Derived>& policy, InputIt1 first1, InputIt1 last1, InputIt2 first2, BinaryPred binary_pred)
 {
-  const auto transform_first =
-    detail::transform_pair_of_input_iterators_t<bool, InputIt1, InputIt2, BinaryPred>(first1, first2, binary_pred);
-  const auto result = cuda_cub::find_if_not(
-    policy, transform_first, transform_first + ::cuda::std::distance(first1, last1), ::cuda::std::identity{});
-  return thrust::make_pair(first1 + ::cuda::std::distance(transform_first, result),
-                           first2 + ::cuda::std::distance(transform_first, result));
-
-  // FIXME(bgruber): the following code should be equivalent and not require a dedicated iterator. However, it
-  // additionally requires the value_type to constructible/destructible on the device, which should be fixed at some
-  // point. See also: https://github.com/NVIDIA/cccl/issues/3591
-#  if 0
   const auto n            = ::cuda::std::distance(first1, last1);
-  const auto first        = make_zip_iterator(first1, first2);
-  const auto last         = make_zip_iterator(last1, first2 + n);
-  const auto mismatch_pos = cuda_cub::find_if_not(policy, first, last, make_zip_function(binary_pred));
+  const auto first        = ::cuda::make_zip_iterator(first1, first2);
+  const auto last         = ::cuda::make_zip_iterator(last1, first2 + n);
+  const auto mismatch_pos = cuda_cub::find_if_not(policy, first, last, ::cuda::zip_function(binary_pred));
   const auto dist         = ::cuda::std::distance(first, mismatch_pos);
   return thrust::make_pair(first1 + dist, first2 + dist);
-#  endif
 }
 
 template <class Derived, class InputIt1, class InputIt2>
@@ -217,7 +83,6 @@ mismatch(execution_policy<Derived>& policy, InputIt1 first1, InputIt1 last1, Inp
   using InputType1 = thrust::detail::it_value_t<InputIt1>;
   return cuda_cub::mismatch(policy, first1, last1, first2, ::cuda::std::equal_to<InputType1>());
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -49,7 +49,6 @@ THRUST_NAMESPACE_BEGIN
 
 namespace cuda_cub
 {
-
 _CCCL_EXEC_CHECK_DISABLE
 template <class Derived, class F, class Size>
 void _CCCL_HOST_DEVICE parallel_for(execution_policy<Derived>& policy, F f, Size count)
@@ -74,7 +73,6 @@ void _CCCL_HOST_DEVICE parallel_for(execution_policy<Derived>& policy, F f, Size
   ));
   // clang-format on
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -60,10 +60,8 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 namespace detail
 {
-
 template <typename Derived, typename InputIt, typename StencilIt, typename OutputIt, typename Predicate, typename OffsetT>
 struct DispatchPartitionIf
 {
@@ -253,7 +251,6 @@ THRUST_RUNTIME_FUNCTION InputIt inplace_partition(
     partition(policy, tmp.data().get(), tmp.data().get() + num_items, stencil, first, predicate);
   return first + num_selected;
 }
-
 } // namespace detail
 
 //-------------------------
@@ -399,7 +396,6 @@ is_partitioned(execution_policy<Derived>& policy, ItemsIt first, ItemsIt last, P
   ItemsIt end      = cuda_cub::find_if(policy, boundary, last, predicate);
   return end == last;
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -51,7 +51,6 @@ THRUST_NAMESPACE_BEGIN
 
 namespace cuda_cub
 {
-
 template <typename MR, typename DerivedPolicy>
 _CCCL_HOST MR* get_per_device_resource(execution_policy<DerivedPolicy>&)
 {
@@ -64,7 +63,6 @@ _CCCL_HOST MR* get_per_device_resource(execution_policy<DerivedPolicy>&)
   std::lock_guard<std::mutex> lock{map_lock};
   return &device_id_to_resource[device_id];
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -82,10 +82,8 @@ void _CCCL_HOST_DEVICE reduce_into(
 
 namespace cuda_cub
 {
-
 namespace __reduce
 {
-
 template <bool>
 struct is_true : thrust::detail::false_type
 {};
@@ -610,7 +608,6 @@ struct DrainAgent
 
 namespace detail
 {
-
 template <typename Derived, typename InputIt, typename Size, typename T, typename BinaryOp>
 THRUST_RUNTIME_FUNCTION size_t get_reduce_n_temporary_storage_size(
   execution_policy<Derived>& policy, InputIt first, Size num_items, T init, BinaryOp binary_op)
@@ -700,7 +697,6 @@ THRUST_RUNTIME_FUNCTION void reduce_n_into_impl(
   status = cuda_cub::synchronize_optional(policy);
   cuda_cub::throw_on_error(status, "reduce failed to synchronize");
 }
-
 } // namespace detail
 
 //-------------------------
@@ -774,7 +770,6 @@ _CCCL_HOST_DEVICE void reduce_into(execution_policy<Derived>& policy, InputIt fi
   using value_type = thrust::detail::it_value_t<InputIt>;
   return cuda_cub::reduce_into(policy, first, last, output, value_type(0));
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -83,10 +83,8 @@ _CCCL_HOST_DEVICE thrust::pair<OutputIterator1, OutputIterator2> reduce_by_key(
 
 namespace cuda_cub
 {
-
 namespace __reduce_by_key
 {
-
 template <bool>
 struct is_true : thrust::detail::false_type
 {};
@@ -912,7 +910,6 @@ THRUST_RUNTIME_FUNCTION pair<KeysOutputIt, ValuesOutputIt> reduce_by_key(
 
   return result;
 }
-
 } // namespace __reduce_by_key
 
 //-------------------------
@@ -990,7 +987,6 @@ pair<KeyOutputIt, ValOutputIt> _CCCL_HOST_DEVICE reduce_by_key(
   return cuda_cub::reduce_by_key(
     policy, keys_first, keys_last, values_first, keys_output, values_output, ::cuda::std::equal_to<KeyT>());
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -42,7 +42,6 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 // in-place
 
 _CCCL_EXEC_CHECK_DISABLE
@@ -101,7 +100,6 @@ remove_copy(execution_policy<Derived>& policy, InputIt first, InputIt last, Outp
   thrust::detail::equal_to_value<T> pred{value};
   return cuda_cub::remove_copy_if(policy, first, last, result, pred);
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -116,7 +116,6 @@ OutputIt _CCCL_HOST_DEVICE replace_copy(
 {
   return cuda_cub::replace_copy_if(policy, first, last, result, thrust::detail::equal_to_value<T>{old_value}, new_value);
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -42,13 +42,11 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 template <class Derived, class ItemsIt, class ResultIt>
 ResultIt _CCCL_HOST_DEVICE reverse_copy(execution_policy<Derived>& policy, ItemsIt first, ItemsIt last, ResultIt result);
 
 template <class Derived, class ItemsIt>
 void _CCCL_HOST_DEVICE reverse(execution_policy<Derived>& policy, ItemsIt first, ItemsIt last);
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 
@@ -62,7 +60,6 @@ THRUST_NAMESPACE_END
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 template <class Derived, class ItemsIt, class ResultIt>
 ResultIt _CCCL_HOST_DEVICE reverse_copy(execution_policy<Derived>& policy, ItemsIt first, ItemsIt last, ResultIt result)
 {
@@ -81,7 +78,6 @@ void _CCCL_HOST_DEVICE reverse(execution_policy<Derived>& policy, ItemsIt first,
 
   cuda_cub::swap_ranges(policy, first, mid, ::cuda::std::make_reverse_iterator(last));
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -42,7 +42,6 @@
 
 #  include <cub/device/device_scan.cuh>
 
-#  include <thrust/detail/integer_math.h>
 #  include <thrust/detail/temporary_array.h>
 #  include <thrust/detail/type_traits.h>
 #  include <thrust/distance.h>
@@ -57,7 +56,6 @@ namespace cuda_cub
 {
 namespace detail
 {
-
 _CCCL_EXEC_CHECK_DISABLE
 template <typename Derived, typename InputIt, typename Size, typename OutputIt, typename ScanOp>
 _CCCL_HOST_DEVICE OutputIt inclusive_scan_n_impl(
@@ -225,7 +223,6 @@ _CCCL_HOST_DEVICE OutputIt exclusive_scan_n_impl(
 
   return result + num_items;
 }
-
 } // namespace detail
 
 //-------------------------
@@ -333,7 +330,6 @@ exclusive_scan(thrust::cuda_cub::execution_policy<Derived>& policy, InputIt firs
   using init_type = thrust::detail::it_value_t<InputIt>;
   return cuda_cub::exclusive_scan(policy, first, last, result, init_type{});
 };
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 

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

@@ -61,7 +61,6 @@ namespace cuda_cub
 {
 namespace detail
 {
-
 _CCCL_EXEC_CHECK_DISABLE
 template <typename Derived,
           typename KeysInIt,
@@ -278,7 +277,6 @@ _CCCL_HOST_DEVICE ValuesOutIt exclusive_scan_by_key_n(
 
   return result + num_items;
 }
-
 } // namespace detail
 
 //-------------------------
@@ -405,7 +403,6 @@ ValOutputIt _CCCL_HOST_DEVICE exclusive_scan_by_key(
   using value_type = thrust::detail::it_value_t<ValInputIt>;
   return cuda_cub::exclusive_scan_by_key(policy, key_first, key_last, value_first, value_result, value_type{});
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 

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

@@ -45,7 +45,6 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 template <class Derived, class ItemsIt, class MapIt, class ResultIt>
 void _CCCL_HOST_DEVICE
 scatter(execution_policy<Derived>& policy, ItemsIt first, ItemsIt last, MapIt map, ResultIt result)
@@ -85,7 +84,6 @@ void _CCCL_HOST_DEVICE scatter_if(
   cuda_cub::scatter_if(
     policy, first, last, map, stencil, result, ::cuda::proclaim_copyable_arguments(::cuda::std::identity{}));
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -55,16 +55,15 @@
 
 #  include <cuda/std/__algorithm/max.h>
 #  include <cuda/std/__algorithm/min.h>
+#  include <cuda/std/__bit/popcount.h>
 #  include <cuda/std/cstdint>
 
 THRUST_NAMESPACE_BEGIN
 
 namespace cuda_cub
 {
-
 namespace __set_operations
 {
-
 template <bool UpperBound, class IntT, class Size, class It, class T, class Comp>
 _CCCL_DEVICE_API _CCCL_FORCEINLINE void
 binary_search_iteration(It data, Size& begin, Size& end, T key, int shift, Comp comp)
@@ -560,7 +559,7 @@ struct SetOpAgent
       Size tile_output_count    = 0;
       Size thread_output_prefix = 0;
       Size tile_output_prefix   = 0;
-      Size thread_output_count  = static_cast<Size>(__popc(active_mask));
+      Size thread_output_count  = static_cast<Size>(::cuda::std::popcount(static_cast<unsigned>(active_mask)));
 
       if (tile_idx == 0) // first tile
       {
@@ -595,7 +594,7 @@ struct SetOpAgent
               tile_output_prefix,
               tile_output_count);
 
-      if (HAS_VALUES::value)
+      if constexpr (HAS_VALUES::value)
       {
         value_type values_loc[ITEMS_PER_THREAD];
         gmem_to_reg<!IS_LAST_TILE>(values_loc, values1_in + keys1_beg, values2_in + keys2_beg, num_keys1, num_keys2);
@@ -1728,7 +1727,6 @@ pair<KeysOutputIt, ItemsOutputIt> _CCCL_HOST_DEVICE set_union_by_key(
     items_result,
     ::cuda::std::less<value_type>());
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -43,7 +43,6 @@
 #  include <cub/device/device_radix_sort.cuh>
 
 #  include <thrust/detail/alignment.h>
-#  include <thrust/detail/integer_math.h>
 #  include <thrust/detail/temporary_array.h>
 #  include <thrust/detail/trivial_sequence.h>
 #  include <thrust/distance.h>
@@ -63,10 +62,8 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 namespace __merge_sort
 {
-
 template <class KeysIt, class ItemsIt, class Size, class CompareOp>
 THRUST_RUNTIME_FUNCTION cudaError_t doit_step(
   void* d_temp_storage,
@@ -172,7 +169,6 @@ THRUST_RUNTIME_FUNCTION void merge_sort(
 
 namespace __radix_sort
 {
-
 template <class SORT_ITEMS, class Comparator>
 struct dispatch;
 
@@ -304,7 +300,6 @@ THRUST_RUNTIME_FUNCTION void radix_sort(execution_policy<Derived>& policy, Key*
 
 namespace __smart_sort
 {
-
 template <class Key, class CompareOp>
 using can_use_primitive_sort = ::cuda::std::integral_constant<
   bool,
@@ -345,12 +340,16 @@ template <
   class CompareOp,
   ::cuda::std::enable_if_t<can_use_primitive_sort<thrust::detail::it_value_t<KeysIt>, CompareOp>::value, int> = 0>
 THRUST_RUNTIME_FUNCTION void smart_sort(
-  execution_policy<Policy>& policy, KeysIt keys_first, KeysIt keys_last, ItemsIt items_first, CompareOp compare_op)
+  execution_policy<Policy>& policy,
+  KeysIt keys_first,
+  KeysIt keys_last,
+  [[maybe_unused]] ItemsIt items_first,
+  CompareOp compare_op)
 {
   // ensure sequences have trivial iterators
   thrust::detail::trivial_sequence<KeysIt, Policy> keys(policy, keys_first, keys_last);
 
-  if (SORT_ITEMS::value)
+  if constexpr (SORT_ITEMS::value)
   {
     thrust::detail::trivial_sequence<ItemsIt, Policy> values(
       policy, items_first, items_first + (keys_last - keys_first));
@@ -362,7 +361,7 @@ THRUST_RUNTIME_FUNCTION void smart_sort(
       keys_last - keys_first,
       compare_op);
 
-    if (!is_contiguous_iterator_v<ItemsIt>)
+    if constexpr (!is_contiguous_iterator_v<ItemsIt>)
     {
       cuda_cub::copy(policy, values.begin(), values.end(), items_first);
     }
@@ -378,7 +377,7 @@ THRUST_RUNTIME_FUNCTION void smart_sort(
   }
 
   // copy results back, if necessary
-  if (!is_contiguous_iterator_v<KeysIt>)
+  if constexpr (!is_contiguous_iterator_v<KeysIt>)
   {
     cuda_cub::copy(policy, keys.begin(), keys.end(), keys_first);
   }
@@ -464,7 +463,6 @@ stable_sort_by_key(execution_policy<Derived>& policy, KeysIt keys_first, KeysIt
   using key_type = thrust::detail::it_value_t<KeysIt>;
   cuda_cub::stable_sort_by_key(policy, keys_first, keys_last, values, ::cuda::std::less<key_type>());
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -33,7 +33,6 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 // If par_nosync does not have a user provided allocator attached, these
 // overloads should be selected.
 
@@ -127,6 +126,5 @@ _CCCL_HOST void return_temporary_buffer(execute_on_stream_nosync& system, Pointe
     }
   }
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END

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

@@ -422,7 +422,6 @@ _CCCL_API _CCCL_FORCEINLINE OutputIt transform_if_n(
   return __transform::binary_if_with_stencil(
     policy, first1, first2, result, num_items, stencil, transform_op, predicate);
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -60,10 +60,8 @@
 THRUST_NAMESPACE_BEGIN
 namespace cuda_cub
 {
-
 namespace detail
 {
-
 template <typename Derived, typename InputIt, typename Size, typename UnaryOp, typename T, typename BinaryOp>
 THRUST_RUNTIME_FUNCTION T transform_reduce_n_impl(
   execution_policy<Derived>& policy, InputIt first, Size num_items, UnaryOp unary_op, T init, BinaryOp binary_op)
@@ -120,7 +118,6 @@ THRUST_RUNTIME_FUNCTION T transform_reduce_n_impl(
   // make this guarantee.
   return thrust::cuda_cub::get_value(policy, thrust::detail::aligned_reinterpret_cast<T*>(tmp.data().get()));
 }
-
 } // namespace detail
 
 _CCCL_EXEC_CHECK_DISABLE
@@ -137,7 +134,6 @@ T _CCCL_HOST_DEVICE transform_reduce(
        cvt_to_seq(derived_cast(policy)), first, first + num_items, transform_op, init, reduce_op);));
   return init;
 }
-
 } // namespace cuda_cub
 THRUST_NAMESPACE_END
 #endif

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

@@ -48,7 +48,6 @@ THRUST_NAMESPACE_BEGIN
 
 namespace cuda_cub
 {
-
 template <class Derived, class InputIt, class OutputIt, class TransformOp, class ScanOp>
 OutputIt _CCCL_HOST_DEVICE transform_inclusive_scan(
   execution_policy<Derived>& policy,
@@ -112,7 +111,6 @@ OutputIt _CCCL_HOST_DEVICE transform_exclusive_scan(
   return cuda_cub::exclusive_scan_n(
     policy, transformed_iterator_t(first, transform_op), num_items, result, init, scan_op);
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -43,6 +43,7 @@
 #  include <thrust/system/cuda/detail/util.h>
 #  include <thrust/type_traits/is_trivially_relocatable.h>
 
+#  include <cuda/std/__new/device_new.h>
 #  include <cuda/std/iterator>
 
 THRUST_NAMESPACE_BEGIN
@@ -70,13 +71,7 @@ struct functor
   {
     InputType const& in = raw_reference_cast(input[idx]);
     OutputType& out     = raw_reference_cast(output[idx]);
-
-#  if _CCCL_CUDA_COMPILER(CLANG)
-    // XXX unsafe, but clang is seemngly unable to call in-place new
-    out = in;
-#  else
     ::new (static_cast<void*>(&out)) OutputType(in);
-#  endif
   }
 };
 } // namespace __uninitialized_copy
@@ -110,7 +105,6 @@ uninitialized_copy(execution_policy<Derived>& policy, InputIt first, InputIt las
 {
   return cuda_cub::uninitialized_copy_n(policy, first, ::cuda::std::distance(first, last), result);
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END

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

@@ -43,6 +43,8 @@
 #  include <thrust/system/cuda/detail/parallel_for.h>
 #  include <thrust/system/cuda/detail/util.h>
 
+#  include <cuda/std/__new/device_new.h>
+
 THRUST_NAMESPACE_BEGIN
 
 namespace cuda_cub
@@ -61,13 +63,7 @@ struct functor
   void _CCCL_DEVICE_API _CCCL_FORCEINLINE operator()(Size idx)
   {
     value_type& out = raw_reference_cast(items[idx]);
-
-#  if _CCCL_CUDA_COMPILER(CLANG)
-    // XXX unsafe. cuda-clang is seemingly unable to call ::new in device code
-    out = value;
-#  else
     ::new (static_cast<void*>(&out)) value_type(value);
-#  endif
   }
 };
 } // namespace __uninitialized_fill
@@ -98,7 +94,6 @@ void _CCCL_HOST_DEVICE uninitialized_fill(execution_policy<Derived>& policy, Ite
 {
   cuda_cub::uninitialized_fill_n(policy, first, ::cuda::std::distance(first, last), x);
 }
-
 } // namespace cuda_cub
 
 THRUST_NAMESPACE_END