Use cuda::proclaim_return_type on device lambdas. (#14577)

This PR makes cudf compatible with Thrust 2 by adding `cuda::proclaim_return_type`, but does not upgrade the version of Thrust/CCCL just yet. Currently we use libcudacxx 2.1.0, which makes `cuda::proclaim_return_type` available, but we still use Thrust 1.17.2 which doesn't require device lambdas to have proclaimed return types. This diff is separated out from #14576 and reduces the diff we must carry in #14576 -- which should contain only packaging changes / version updates for the CCCL 2 migration. This PR is **nonbreaking**, while #14576 will be **breaking**.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Nghia Truong (https://github.com/ttnghia)
  - David Wendt (https://github.com/davidwendt)

URL: #14577

cpp/benchmarks/common/generate_input.cu

@@ -53,6 +53,8 @@
 #include <thrust/transform.h>
 #include <thrust/tuple.h>
 
+#include <cuda/functional>
+
 #include <algorithm>
 #include <cstdint>
 #include <memory>
@@ -247,12 +249,12 @@ struct random_value_fn<T, std::enable_if_t<cudf::is_chrono<T>()>> {
       sec.end(),
       ns.begin(),
       result.begin(),
-      [] __device__(int64_t sec_value, int64_t nanoseconds_value) {
+      cuda::proclaim_return_type<T>([] __device__(int64_t sec_value, int64_t nanoseconds_value) {
         auto const timestamp_ns =
           cudf::duration_s{sec_value} + cudf::duration_ns{nanoseconds_value};
         // Return value in the type's precision
         return T(cuda::std::chrono::duration_cast<typename T::duration>(timestamp_ns));
-      });
+      }));
     return result;
   }
 };
@@ -367,12 +369,13 @@ rmm::device_uvector<cudf::size_type> sample_indices_with_run_length(cudf::size_t
     // This is gather.
     auto avg_repeated_sample_indices_iterator = thrust::make_transform_iterator(
       thrust::make_counting_iterator(0),
-      [rb              = run_lens.begin(),
-       re              = run_lens.end(),
-       samples_indices = samples_indices.begin()] __device__(cudf::size_type i) {
-        auto sample_idx = thrust::upper_bound(thrust::seq, rb, re, i) - rb;
-        return samples_indices[sample_idx];
-      });
+      cuda::proclaim_return_type<cudf::size_type>(
+        [rb              = run_lens.begin(),
+         re              = run_lens.end(),
+         samples_indices = samples_indices.begin()] __device__(cudf::size_type i) {
+          auto sample_idx = thrust::upper_bound(thrust::seq, rb, re, i) - rb;
+          return samples_indices[sample_idx];
+        }));
     rmm::device_uvector<cudf::size_type> repeated_sample_indices(num_rows,
                                                                  cudf::get_default_stream());
     thrust::copy(thrust::device,
@@ -513,7 +516,7 @@ std::unique_ptr<cudf::column> create_random_utf8_string_column(data_profile cons
     lengths.end(),
     null_mask.begin(),
     lengths.begin(),
-    [] __device__(auto) { return 0; },
+    cuda::proclaim_return_type<cudf::size_type>([] __device__(auto) { return 0; }),
     thrust::logical_not<bool>{});
   auto valid_lengths = thrust::make_transform_iterator(
     thrust::make_zip_iterator(thrust::make_tuple(lengths.begin(), null_mask.begin())),

cpp/include/cudf/column/column_view.hpp

@@ -478,7 +478,10 @@ class mutable_column_view : public detail::column_view_base {
  public:
   mutable_column_view() = default;
 
-  ~mutable_column_view() = default;
+  ~mutable_column_view(){
+    // Needed so that the first instance of the implicit destructor for any TU isn't 'constructed'
+    // from a host+device function marking the implicit version also as host+device
+  };
 
   mutable_column_view(mutable_column_view const&) = default;  ///< Copy constructor
   mutable_column_view(mutable_column_view&&)      = default;  ///< Move constructor

cpp/include/cudf/detail/null_mask.cuh

@@ -37,6 +37,8 @@
 #include <thrust/transform.h>
 #include <thrust/tuple.h>
 
+#include <cuda/functional>
+
 #include <algorithm>
 #include <iterator>
 #include <optional>
@@ -330,20 +332,21 @@ rmm::device_uvector<size_type> segmented_count_bits(bitmask_type const* bitmask,
     // set bits from the length of the segment.
     auto segments_begin =
       thrust::make_zip_iterator(first_bit_indices_begin, last_bit_indices_begin);
-    auto segment_length_iterator =
-      thrust::transform_iterator(segments_begin, [] __device__(auto const& segment) {
+    auto segment_length_iterator = thrust::transform_iterator(
+      segments_begin, cuda::proclaim_return_type<size_type>([] __device__(auto const& segment) {
         auto const begin = thrust::get<0>(segment);
         auto const end   = thrust::get<1>(segment);
         return end - begin;
-      });
+      }));
     thrust::transform(rmm::exec_policy(stream),
                       segment_length_iterator,
                       segment_length_iterator + num_ranges,
                       d_bit_counts.data(),
                       d_bit_counts.data(),
-                      [] __device__(auto segment_size, auto segment_bit_count) {
-                        return segment_size - segment_bit_count;
-                      });
+                      cuda::proclaim_return_type<size_type>(
+                        [] __device__(auto segment_size, auto segment_bit_count) {
+                          return segment_size - segment_bit_count;
+                        }));
   }
 
   CUDF_CHECK_CUDA(stream.value());
@@ -541,12 +544,12 @@ std::pair<rmm::device_buffer, size_type> segmented_null_mask_reduction(
 {
   auto const segments_begin =
     thrust::make_zip_iterator(first_bit_indices_begin, last_bit_indices_begin);
-  auto const segment_length_iterator =
-    thrust::make_transform_iterator(segments_begin, [] __device__(auto const& segment) {
+  auto const segment_length_iterator = thrust::make_transform_iterator(
+    segments_begin, cuda::proclaim_return_type<size_type>([] __device__(auto const& segment) {
       auto const begin = thrust::get<0>(segment);
       auto const end   = thrust::get<1>(segment);
       return end - begin;
-    });
+    }));
 
   auto const num_segments =
     static_cast<size_type>(std::distance(first_bit_indices_begin, first_bit_indices_end));

cpp/include/cudf/detail/sizes_to_offsets_iterator.cuh

@@ -27,6 +27,8 @@
 #include <thrust/distance.h>
 #include <thrust/scan.h>
 
+#include <cuda/functional>
+
 #include <stdexcept>
 
 namespace cudf {
@@ -311,9 +313,10 @@ std::pair<std::unique_ptr<column>, size_type> make_offsets_child_column(
   // using exclusive-scan technically requires count+1 input values even though
   // the final input value is never used.
   // The input iterator is wrapped here to allow the last value to be safely read.
-  auto map_fn = [begin, count] __device__(size_type idx) -> size_type {
-    return idx < count ? static_cast<size_type>(begin[idx]) : size_type{0};
-  };
+  auto map_fn =
+    cuda::proclaim_return_type<size_type>([begin, count] __device__(size_type idx) -> size_type {
+      return idx < count ? static_cast<size_type>(begin[idx]) : size_type{0};
+    });
   auto input_itr = cudf::detail::make_counting_transform_iterator(0, map_fn);
   // Use the sizes-to-offsets iterator to compute the total number of elements
   auto const total_elements = sizes_to_offsets(input_itr, input_itr + count + 1, d_offsets, stream);

cpp/include/cudf/detail/utilities/cast_functor.cuh

@@ -0,0 +1,63 @@
+/*
+ * Copyright (c) 2023, NVIDIA CORPORATION.
+ *
+ * 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.
+ */
+
+#pragma once
+
+/**
+ * @brief A casting functor wrapping another functor.
+ * @file
+ */
+
+#include <cudf/types.hpp>
+
+#include <cuda/functional>
+
+#include <type_traits>
+#include <utility>
+
+namespace cudf {
+namespace detail {
+
+/**
+ * @brief Functor that casts another functor's result to a specified type.
+ *
+ * CUB 2.0.0 reductions require that the binary operator returns the same type
+ * as the initial value type, so we wrap binary operators with this when used
+ * by CUB.
+ */
+template <typename ResultType, typename F>
+struct cast_functor_fn {
+  F f;
+
+  template <typename... Ts>
+  CUDF_HOST_DEVICE inline ResultType operator()(Ts&&... args)
+  {
+    return static_cast<ResultType>(f(std::forward<Ts>(args)...));
+  }
+};
+
+/**
+ * @brief Function creating a casting functor.
+ */
+template <typename ResultType, typename F>
+inline cast_functor_fn<ResultType, std::decay_t<F>> cast_functor(F&& f)
+{
+  return cast_functor_fn<ResultType, std::decay_t<F>>{std::forward<F>(f)};
+}
+
+}  // namespace detail
+
+}  // namespace cudf

cpp/include/cudf/detail/utilities/element_argminmax.cuh

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022, NVIDIA CORPORATION.
+ * Copyright (c) 2022-2023, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -37,7 +37,7 @@ struct element_argminmax_fn {
   bool const has_nulls;
   bool const arg_min;
 
-  __device__ inline auto operator()(size_type const& lhs_idx, size_type const& rhs_idx) const
+  __device__ inline size_type operator()(size_type const& lhs_idx, size_type const& rhs_idx) const
   {
     // The extra bounds checking is due to issue github.com/rapidsai/cudf/9156 and
     // github.com/NVIDIA/thrust/issues/1525

cpp/include/cudf/lists/detail/gather.cuh

@@ -30,6 +30,8 @@
 #include <thrust/iterator/counting_iterator.h>
 #include <thrust/transform.h>
 
+#include <cuda/functional>
+
 namespace cudf {
 namespace lists {
 namespace detail {
@@ -83,12 +85,12 @@ gather_data make_gather_data(cudf::lists_column_view const& source_column,
 
   auto sizes_itr = cudf::detail::make_counting_transform_iterator(
     0,
-    [source_column_nullmask,
-     source_column_offset = source_column.offset(),
-     gather_map,
-     output_count,
-     src_offsets,
-     src_size] __device__(int32_t index) -> int32_t {
+    cuda::proclaim_return_type<int32_t>([source_column_nullmask,
+                                         source_column_offset = source_column.offset(),
+                                         gather_map,
+                                         output_count,
+                                         src_offsets,
+                                         src_size] __device__(int32_t index) -> int32_t {
       int32_t offset_index = index < output_count ? gather_map[index] : 0;
 
       // if this is an invalid index, this will be a NULL list
@@ -102,7 +104,7 @@ gather_data make_gather_data(cudf::lists_column_view const& source_column,
 
       // the length of this list
       return src_offsets[offset_index + 1] - src_offsets[offset_index];
-    });
+    }));
 
   auto [dst_offsets_c, map_size] =
     cudf::detail::make_offsets_child_column(sizes_itr, sizes_itr + output_count, stream, mr);

cpp/include/cudf/lists/detail/scatter.cuh

@@ -39,6 +39,8 @@
 #include <thrust/sequence.h>
 #include <thrust/transform.h>
 
+#include <cuda/functional>
+
 #include <cinttypes>
 
 namespace cudf {
@@ -62,9 +64,10 @@ rmm::device_uvector<unbound_list_view> list_vector_from_column(
                     index_begin,
                     index_end,
                     vector.begin(),
-                    [label, lists_column] __device__(size_type row_index) {
-                      return unbound_list_view{label, lists_column, row_index};
-                    });
+                    cuda::proclaim_return_type<unbound_list_view>(
+                      [label, lists_column] __device__(size_type row_index) {
+                        return unbound_list_view{label, lists_column, row_index};
+                      }));
 
   return vector;
 }
@@ -115,7 +118,8 @@ std::unique_ptr<column> scatter_impl(rmm::device_uvector<unbound_list_view> cons
     lists_column_view(target);  // Checks that target is a list column.
 
   auto list_size_begin = thrust::make_transform_iterator(
-    target_vector.begin(), [] __device__(unbound_list_view l) { return l.size(); });
+    target_vector.begin(),
+    cuda::proclaim_return_type<size_type>([] __device__(unbound_list_view l) { return l.size(); }));
   auto offsets_column = std::get<0>(cudf::detail::make_offsets_child_column(
     list_size_begin, list_size_begin + target.size(), stream, mr));
 

cpp/include/cudf/reduction/detail/reduction.cuh

@@ -19,6 +19,7 @@
 #include "reduction_operators.cuh"
 
 #include <cudf/column/column_factories.hpp>
+#include <cudf/detail/utilities/cast_functor.cuh>
 #include <cudf/utilities/type_dispatcher.hpp>
 
 #include <rmm/cuda_stream_view.hpp>
@@ -64,7 +65,7 @@ std::unique_ptr<scalar> reduce(InputIterator d_in,
                                rmm::cuda_stream_view stream,
                                rmm::mr::device_memory_resource* mr)
 {
-  auto const binary_op     = op.get_binary_op();
+  auto const binary_op     = cudf::detail::cast_functor<OutputType>(op.get_binary_op());
   auto const initial_value = init.value_or(op.template get_identity<OutputType>());
   auto dev_result          = rmm::device_scalar<OutputType>{initial_value, stream, mr};
 
@@ -124,7 +125,7 @@ std::unique_ptr<scalar> reduce(InputIterator d_in,
                                rmm::cuda_stream_view stream,
                                rmm::mr::device_memory_resource* mr)
 {
-  auto const binary_op     = op.get_binary_op();
+  auto const binary_op     = cudf::detail::cast_functor<OutputType>(op.get_binary_op());
   auto const initial_value = init.value_or(op.template get_identity<OutputType>());
   auto dev_result          = rmm::device_scalar<OutputType>{initial_value, stream};
 
@@ -190,7 +191,7 @@ std::unique_ptr<scalar> reduce(InputIterator d_in,
                                rmm::cuda_stream_view stream,
                                rmm::mr::device_memory_resource* mr)
 {
-  auto const binary_op     = op.get_binary_op();
+  auto const binary_op     = cudf::detail::cast_functor<IntermediateType>(op.get_binary_op());
   auto const initial_value = op.template get_identity<IntermediateType>();
 
   rmm::device_scalar<IntermediateType> intermediate_result{initial_value, stream};

cpp/include/cudf/reduction/detail/segmented_reduction.cuh

@@ -18,6 +18,8 @@
 
 #include "reduction_operators.cuh"
 
+#include <cudf/detail/utilities/cast_functor.cuh>
+
 #include <rmm/cuda_stream_view.hpp>
 #include <rmm/device_buffer.hpp>
 #include <rmm/exec_policy.hpp>
@@ -45,7 +47,7 @@ namespace detail {
  * @param d_offset_begin Begin iterator to segment indices
  * @param d_offset_end   End iterator to segment indices
  * @param d_out          Output data iterator
- * @param binary_op      The reduction operator
+ * @param op             The reduction operator
  * @param initial_value  Initial value of the reduction
  * @param stream         CUDA stream used for device memory operations and kernel launches
  *
@@ -61,12 +63,12 @@ void segmented_reduce(InputIterator d_in,
                       OffsetIterator d_offset_begin,
                       OffsetIterator d_offset_end,
                       OutputIterator d_out,
-                      BinaryOp binary_op,
+                      BinaryOp op,
                       OutputType initial_value,
                       rmm::cuda_stream_view stream)
 {
   auto const num_segments = static_cast<size_type>(std::distance(d_offset_begin, d_offset_end)) - 1;
-
+  auto const binary_op    = cudf::detail::cast_functor<OutputType>(op);
   // Allocate temporary storage
   size_t temp_storage_bytes = 0;
   cub::DeviceSegmentedReduce::Reduce(nullptr,
@@ -148,8 +150,8 @@ void segmented_reduce(InputIterator d_in,
   using OutputType       = typename thrust::iterator_value<OutputIterator>::type;
   using IntermediateType = typename thrust::iterator_value<InputIterator>::type;
   auto num_segments      = static_cast<size_type>(std::distance(d_offset_begin, d_offset_end)) - 1;
-  auto const binary_op   = op.get_binary_op();
   auto const initial_value = op.template get_identity<IntermediateType>();
+  auto const binary_op     = cudf::detail::cast_functor<IntermediateType>(op.get_binary_op());
 
   rmm::device_uvector<IntermediateType> intermediate_result{static_cast<std::size_t>(num_segments),
                                                             stream};

cpp/include/cudf/strings/detail/copy_if_else.cuh

@@ -29,6 +29,8 @@
 #include <thrust/iterator/transform_iterator.h>
 #include <thrust/optional.h>
 
+#include <cuda/functional>
+
 namespace cudf {
 namespace strings {
 namespace detail {
@@ -78,10 +80,11 @@ std::unique_ptr<cudf::column> copy_if_else(StringIterLeft lhs_begin,
   auto null_mask       = (null_count > 0) ? std::move(valid_mask.first) : rmm::device_buffer{};
 
   // build offsets column
-  auto offsets_transformer = [lhs_begin, rhs_begin, filter_fn] __device__(size_type idx) {
-    auto const result = filter_fn(idx) ? lhs_begin[idx] : rhs_begin[idx];
-    return result.has_value() ? result->size_bytes() : 0;
-  };
+  auto offsets_transformer = cuda::proclaim_return_type<size_type>(
+    [lhs_begin, rhs_begin, filter_fn] __device__(size_type idx) {
+      auto const result = filter_fn(idx) ? lhs_begin[idx] : rhs_begin[idx];
+      return result.has_value() ? result->size_bytes() : 0;
+    });
 
   auto offsets_transformer_itr = thrust::make_transform_iterator(
     thrust::make_counting_iterator<size_type>(0), offsets_transformer);