Move detail reduction functions to cudf::reduction::detail namespace (#…

…12971)

Moves internal/detail reduction (and segmented reduction) functions and other declarations from the `cudf::detail` namespace to the `cudf::reduction::detail` namespace. The detail function headers also moved from `cpp/include/cudf/detail` to `cpp/include/cudf/reduction/detail` directory.
The public header `cpp/include/cudf/reduction.hpp` was not changed or moved.

Authors:
  - David Wendt (https://github.com/davidwendt)

Approvers:
  - Vyas Ramasubramani (https://github.com/vyasr)
  - Robert Maynard (https://github.com/robertmaynard)
  - AJ Schmidt (https://github.com/ajschmidt8)

URL: #12971

conda/recipes/libcudf/meta.yaml

@@ -114,7 +114,6 @@ outputs:
         - test -f $PREFIX/include/cudf/detail/nvtx/nvtx3.hpp
         - test -f $PREFIX/include/cudf/detail/nvtx/ranges.hpp
         - test -f $PREFIX/include/cudf/detail/quantiles.hpp
-        - test -f $PREFIX/include/cudf/detail/reduction_functions.hpp
         - test -f $PREFIX/include/cudf/detail/repeat.hpp
         - test -f $PREFIX/include/cudf/detail/replace.hpp
         - test -f $PREFIX/include/cudf/detail/reshape.hpp
@@ -123,7 +122,6 @@ outputs:
         - test -f $PREFIX/include/cudf/detail/scan.hpp
         - test -f $PREFIX/include/cudf/detail/scatter.hpp
         - test -f $PREFIX/include/cudf/detail/search.hpp
-        - test -f $PREFIX/include/cudf/detail/segmented_reduction_functions.hpp
         - test -f $PREFIX/include/cudf/detail/sequence.hpp
         - test -f $PREFIX/include/cudf/detail/sorting.hpp
         - test -f $PREFIX/include/cudf/detail/stream_compaction.hpp
@@ -218,6 +216,8 @@ outputs:
         - test -f $PREFIX/include/cudf/partitioning.hpp
         - test -f $PREFIX/include/cudf/quantiles.hpp
         - test -f $PREFIX/include/cudf/reduction.hpp
+        - test -f $PREFIX/include/cudf/reduction/detail/reduction_functions.hpp
+        - test -f $PREFIX/include/cudf/reduction/detail/segmented_reduction_functions.hpp
         - test -f $PREFIX/include/cudf/replace.hpp
         - test -f $PREFIX/include/cudf/reshape.hpp
         - test -f $PREFIX/include/cudf/rolling.hpp

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

@@ -16,7 +16,7 @@
 
 #pragma once
 
-#include <cudf/detail/reduction_operators.cuh>
+#include "reduction_operators.cuh"
 
 #include <cudf/column/column_factories.hpp>
 #include <cudf/utilities/type_dispatcher.hpp>
@@ -31,6 +31,8 @@
 #include <thrust/for_each.h>
 #include <thrust/iterator/iterator_traits.h>
 
+#include <optional>
+
 namespace cudf {
 namespace reduction {
 namespace detail {

cpp/include/cudf/detail/reduction_functions.hpp → cpp/include/cudf/reduction/detail/reduction_functions.hpp

@@ -27,6 +27,7 @@
 
 namespace cudf {
 namespace reduction {
+namespace detail {
 /**
  * @brief Computes sum of elements in input column
  *
@@ -323,5 +324,6 @@ std::unique_ptr<scalar> merge_sets(lists_column_view const& col,
                                    rmm::cuda_stream_view stream,
                                    rmm::mr::device_memory_resource* mr);
 
+}  // namespace detail
 }  // namespace reduction
 }  // namespace cudf

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

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019, NVIDIA CORPORATION.
+ * Copyright (c) 2019-2023, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -26,6 +26,7 @@
 
 namespace cudf {
 namespace reduction {
+namespace detail {
 // intermediate data structure to compute `var`, `std`
 template <typename ResultType>
 struct var_std {
@@ -244,7 +245,7 @@ struct variance : public compound_op<variance> {
   using op = cudf::DeviceSum;
 
   template <typename ResultType>
-  using transformer = cudf::reduction::transformer_var_std<ResultType>;
+  using transformer = cudf::reduction::detail::transformer_var_std<ResultType>;
 
   template <typename ResultType>
   struct intermediate {
@@ -270,7 +271,7 @@ struct standard_deviation : public compound_op<standard_deviation> {
   using op = cudf::DeviceSum;
 
   template <typename ResultType>
-  using transformer = cudf::reduction::transformer_var_std<ResultType>;
+  using transformer = cudf::reduction::detail::transformer_var_std<ResultType>;
 
   template <typename ResultType>
   struct intermediate {
@@ -288,7 +289,7 @@ struct standard_deviation : public compound_op<standard_deviation> {
     };
   };
 };
-
 }  // namespace op
+}  // namespace detail
 }  // namespace reduction
 }  // namespace cudf

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

@@ -16,7 +16,7 @@
 
 #pragma once
 
-#include <cudf/detail/reduction_operators.cuh>
+#include "reduction_operators.cuh"
 
 #include <rmm/cuda_stream_view.hpp>
 #include <rmm/device_buffer.hpp>

cpp/include/cudf/detail/segmented_reduction_functions.hpp → cpp/include/cudf/reduction/detail/segmented_reduction_functions.hpp

@@ -27,6 +27,7 @@
 
 namespace cudf {
 namespace reduction {
+namespace detail {
 
 /**
  * @brief Compute sum of each segment in the input column
@@ -50,14 +51,13 @@ namespace reduction {
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Sums of segments as type `output_dtype`
  */
-std::unique_ptr<column> segmented_sum(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  std::optional<std::reference_wrapper<scalar const>> init,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_sum(column_view const& col,
+                                      device_span<size_type const> offsets,
+                                      data_type const output_dtype,
+                                      null_policy null_handling,
+                                      std::optional<std::reference_wrapper<scalar const>> init,
+                                      rmm::cuda_stream_view stream,
+                                      rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Computes product of each segment in the input column
@@ -81,14 +81,13 @@ std::unique_ptr<column> segmented_sum(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Product of segments as type `output_dtype`
  */
-std::unique_ptr<column> segmented_product(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  std::optional<std::reference_wrapper<scalar const>> init,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_product(column_view const& col,
+                                          device_span<size_type const> offsets,
+                                          data_type const output_dtype,
+                                          null_policy null_handling,
+                                          std::optional<std::reference_wrapper<scalar const>> init,
+                                          rmm::cuda_stream_view stream,
+                                          rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Compute minimum of each segment in the input column
@@ -111,14 +110,13 @@ std::unique_ptr<column> segmented_product(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Minimums of segments as type `output_dtype`
  */
-std::unique_ptr<column> segmented_min(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  std::optional<std::reference_wrapper<scalar const>> init,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_min(column_view const& col,
+                                      device_span<size_type const> offsets,
+                                      data_type const output_dtype,
+                                      null_policy null_handling,
+                                      std::optional<std::reference_wrapper<scalar const>> init,
+                                      rmm::cuda_stream_view stream,
+                                      rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Compute maximum of each segment in the input column
@@ -141,14 +139,13 @@ std::unique_ptr<column> segmented_min(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Maximums of segments as type `output_dtype`
  */
-std::unique_ptr<column> segmented_max(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  std::optional<std::reference_wrapper<scalar const>> init,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_max(column_view const& col,
+                                      device_span<size_type const> offsets,
+                                      data_type const output_dtype,
+                                      null_policy null_handling,
+                                      std::optional<std::reference_wrapper<scalar const>> init,
+                                      rmm::cuda_stream_view stream,
+                                      rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Compute if any of the values in the segment are true when typecasted to bool
@@ -172,14 +169,13 @@ std::unique_ptr<column> segmented_max(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Column of type BOOL8 for the results of the segments
  */
-std::unique_ptr<column> segmented_any(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  std::optional<std::reference_wrapper<scalar const>> init,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_any(column_view const& col,
+                                      device_span<size_type const> offsets,
+                                      data_type const output_dtype,
+                                      null_policy null_handling,
+                                      std::optional<std::reference_wrapper<scalar const>> init,
+                                      rmm::cuda_stream_view stream,
+                                      rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Compute if all of the values in the segment are true when typecasted to bool
@@ -203,14 +199,13 @@ std::unique_ptr<column> segmented_any(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Column of BOOL8 for the results of the segments
  */
-std::unique_ptr<column> segmented_all(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  std::optional<std::reference_wrapper<scalar const>> init,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_all(column_view const& col,
+                                      device_span<size_type const> offsets,
+                                      data_type const output_dtype,
+                                      null_policy null_handling,
+                                      std::optional<std::reference_wrapper<scalar const>> init,
+                                      rmm::cuda_stream_view stream,
+                                      rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Computes mean of elements of segments in the input column
@@ -233,13 +228,12 @@ std::unique_ptr<column> segmented_all(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Column of `output_dtype` for the reduction results of the segments
  */
-std::unique_ptr<column> segmented_mean(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_mean(column_view const& col,
+                                       device_span<size_type const> offsets,
+                                       data_type const output_dtype,
+                                       null_policy null_handling,
+                                       rmm::cuda_stream_view stream,
+                                       rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Computes sum of squares of elements of segments in the input column
@@ -262,13 +256,12 @@ std::unique_ptr<column> segmented_mean(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Column of `output_dtype` for the reduction results of the segments
  */
-std::unique_ptr<column> segmented_sum_of_squares(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_sum_of_squares(column_view const& col,
+                                                 device_span<size_type const> offsets,
+                                                 data_type const output_dtype,
+                                                 null_policy null_handling,
+                                                 rmm::cuda_stream_view stream,
+                                                 rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Computes the standard deviation of elements of segments in the input column
@@ -293,14 +286,13 @@ std::unique_ptr<column> segmented_sum_of_squares(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Column of `output_dtype` for the reduction results of the segments
  */
-std::unique_ptr<column> segmented_standard_deviation(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  size_type ddof,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_standard_deviation(column_view const& col,
+                                                     device_span<size_type const> offsets,
+                                                     data_type const output_dtype,
+                                                     null_policy null_handling,
+                                                     size_type ddof,
+                                                     rmm::cuda_stream_view stream,
+                                                     rmm::mr::device_memory_resource* mr);
 
 /**
  * @brief Computes the variance of elements of segments in the input column
@@ -325,14 +317,14 @@ std::unique_ptr<column> segmented_standard_deviation(
  * @param mr Device memory resource used to allocate the returned column's device memory
  * @return Column of `output_dtype` for the reduction results of the segments
  */
-std::unique_ptr<column> segmented_variance(
-  column_view const& col,
-  device_span<size_type const> offsets,
-  data_type const output_dtype,
-  null_policy null_handling,
-  size_type ddof,
-  rmm::cuda_stream_view stream,
-  rmm::mr::device_memory_resource* mr = rmm::mr::get_current_device_resource());
+std::unique_ptr<column> segmented_variance(column_view const& col,
+                                           device_span<size_type const> offsets,
+                                           data_type const output_dtype,
+                                           null_policy null_handling,
+                                           size_type ddof,
+                                           rmm::cuda_stream_view stream,
+                                           rmm::mr::device_memory_resource* mr);
 
+}  // namespace detail
 }  // namespace reduction
 }  // namespace cudf

cpp/src/lists/stream_compaction/apply_boolean_mask.cu

@@ -21,10 +21,10 @@
 #include <cudf/detail/null_mask.hpp>
 #include <cudf/detail/nvtx/ranges.hpp>
 #include <cudf/detail/replace.hpp>
-#include <cudf/detail/segmented_reduction_functions.hpp>
 #include <cudf/detail/stream_compaction.hpp>
 #include <cudf/lists/detail/stream_compaction.hpp>
 #include <cudf/lists/stream_compaction.hpp>
+#include <cudf/reduction/detail/segmented_reduction_functions.hpp>
 #include <cudf/utilities/bit.hpp>
 #include <cudf/utilities/default_stream.hpp>
 
@@ -65,12 +65,14 @@ std::unique_ptr<column> apply_boolean_mask(lists_column_view const& input,
       cudf::detail::slice(
         boolean_mask.offsets(), {boolean_mask.offset(), boolean_mask.size() + 1}, stream)
         .front();
-    auto const sizes       = cudf::reduction::segmented_sum(boolean_mask_sliced_child,
-                                                      boolean_mask_sliced_offsets,
-                                                      offset_data_type,
-                                                      null_policy::EXCLUDE,
-                                                      std::nullopt,
-                                                      stream);
+    auto const sizes =
+      cudf::reduction::detail::segmented_sum(boolean_mask_sliced_child,
+                                             boolean_mask_sliced_offsets,
+                                             offset_data_type,
+                                             null_policy::EXCLUDE,
+                                             std::nullopt,
+                                             stream,
+                                             rmm::mr::get_current_device_resource());
     auto const d_sizes     = column_device_view::create(*sizes, stream);
     auto const sizes_begin = cudf::detail::make_null_replacement_iterator(*d_sizes, offset_type{0});
     auto const sizes_end   = sizes_begin + sizes->size();