Refactor to try and reduce/parallelise compile times

cpp/CMakeLists.txt

@@ -643,6 +643,8 @@ add_library(
   src/reshape/interleave_columns.cu
   src/reshape/tile.cu
   src/rolling/detail/optimized_unbounded_window.cpp
+  src/rolling/detail/range_following.cu
+  src/rolling/detail/range_preceding.cu
   src/rolling/detail/rolling_collect_list.cu
   src/rolling/detail/rolling_fixed_window.cu
   src/rolling/detail/rolling_variable_window.cu

cpp/include/cudf/detail/rolling.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021-2024, NVIDIA CORPORATION.
+ * Copyright (c) 2021-2025, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -29,6 +29,26 @@
 namespace CUDF_EXPORT cudf {
 namespace detail {
 
+namespace rolling {
+/**
+ * @brief Direction tag for a range-based rolling window.
+ */
+enum class direction : bool {
+  PRECEDING,  ///< A preceding window.
+  FOLLOWING,  ///< A following window.
+};
+/**
+ * @brief Wrapper for preprocessed group information from sorted group keys.
+ */
+struct preprocessed_group_info {
+  rmm::device_uvector<size_type> const& labels;   ///< Mapping from row index to group label
+  rmm::device_uvector<size_type> const& offsets;  ///< Mapping from group label to row offsets
+  rmm::device_uvector<size_type> const&
+    nulls_per_group;  ///< Mapping from group label to null count in the group
+};
+
+}  // namespace rolling
+
 /**
  * @copydoc std::unique_ptr<column> rolling_window(
  *            column_view const& input,
@@ -48,5 +68,55 @@ std::unique_ptr<column> rolling_window(column_view const& input,
                                        rmm::cuda_stream_view stream,
                                        rmm::device_async_resource_ref mr);
 
+/**
+ * @brief Make a column representing the window offsets for a range-based window
+ *
+ * @tparam Direction Is this a preceding window or a following one.
+ *
+ * @param group_keys Table defining grouping of the windows. May be empty. If
+ * non-empty, group keys must be sorted.
+ * @param orderby Column use to define window ranges. If @p group_keys is empty,
+ * must be sorted. If
+ * @p group_keys is non-empty, must be sorted within each group. As well as
+ * being sorted, must be sorted consistently with the @p order and @p null_order
+ * parameters.
+ * @param order The sort order of the @p orderby column.
+ * @param null_order The sort order of nulls in the @p orderby column.
+ * @param row_delta Pointer to scalar providing the delta for the window range.
+ * May be null, but only if the @p window_type is @p CURRENT_ROW or @p
+ * UNBOUNDED. Note that @p row_delta is always added to the current row value.
+ * @param window_type The type of window we are computing bounds for.
+ * @param stream CUDA stream used for device memory operations and kernel
+ * launches.
+ * @param mr Device memory resource used for allocations.
+ */
+template <rolling::direction Direction>
+[[nodiscard]] std::unique_ptr<column> make_range_window_bound(
+  column_view const& orderby,
+  std::optional<rolling::preprocessed_group_info> const& grouping,
+  order order,
+  null_order null_order,
+  range_window_type window,
+  rmm::cuda_stream_view stream,
+  rmm::device_async_resource_ref mr);
+
+[[nodiscard]] std::unique_ptr<column> make_preceding_range_window_bound(
+  column_view const& orderby,
+  std::optional<rolling::preprocessed_group_info> const& grouping,
+  order order,
+  null_order null_order,
+  range_window_type window,
+  rmm::cuda_stream_view stream,
+  rmm::device_async_resource_ref mr);
+
+[[nodiscard]] std::unique_ptr<column> make_following_range_window_bound(
+  column_view const& orderby,
+  std::optional<rolling::preprocessed_group_info> const& grouping,
+  order order,
+  null_order null_order,
+  range_window_type window,
+  rmm::cuda_stream_view stream,
+  rmm::device_async_resource_ref mr);
+
 }  // namespace detail
 }  // namespace CUDF_EXPORT cudf

cpp/src/rolling/detail/range_following.cu

@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2025, 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.
+ */
+
+#include "range_utils.cuh"
+
+#include <cudf/column/column.hpp>
+#include <cudf/column/column_view.hpp>
+#include <cudf/detail/rolling.hpp>
+#include <cudf/rolling.hpp>
+#include <cudf/types.hpp>
+
+#include <rmm/cuda_stream_view.hpp>
+#include <rmm/resource_ref.hpp>
+
+#include <optional>
+
+namespace CUDF_EXPORT cudf {
+namespace detail {
+
+[[nodiscard]] std::unique_ptr<column> make_following_range_window_bound(
+  column_view const& orderby,
+  std::optional<rolling::preprocessed_group_info> const& grouping,
+  order order,
+  null_order null_order,
+  range_window_type window,
+  rmm::cuda_stream_view stream,
+  rmm::device_async_resource_ref mr)
+{
+  return make_range_window_bound<rolling::direction::FOLLOWING>(
+    orderby, grouping, order, null_order, window, stream, mr);
+}
+
+}  // namespace detail
+}  // namespace CUDF_EXPORT cudf

cpp/src/rolling/detail/range_preceding.cu

@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2025, 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.
+ */
+
+#include "range_utils.cuh"
+
+#include <cudf/column/column.hpp>
+#include <cudf/column/column_view.hpp>
+#include <cudf/detail/rolling.hpp>
+#include <cudf/rolling.hpp>
+#include <cudf/types.hpp>
+
+#include <rmm/cuda_stream_view.hpp>
+#include <rmm/resource_ref.hpp>
+
+#include <optional>
+
+namespace CUDF_EXPORT cudf {
+namespace detail {
+
+[[nodiscard]] std::unique_ptr<column> make_preceding_range_window_bound(
+  column_view const& orderby,
+  std::optional<rolling::preprocessed_group_info> const& grouping,
+  order order,
+  null_order null_order,
+  range_window_type window,
+  rmm::cuda_stream_view stream,
+  rmm::device_async_resource_ref mr)
+{
+  return make_range_window_bound<rolling::direction::PRECEDING>(
+    orderby, grouping, order, null_order, window, stream, mr);
+}
+
+}  // namespace detail
+}  // namespace CUDF_EXPORT cudf