Add a shortcut for when the input clusters are all empty for the tdigest merge

cpp/include/cudf/detail/tdigest/tdigest.hpp

@@ -143,28 +143,29 @@ std::unique_ptr<column> make_tdigest_column(size_type num_rows,
                                             rmm::device_async_resource_ref mr);
 
 /**
- * @brief Create an empty tdigest column.
+ * @brief Create a tdigest column of empty clusters.
  *
- * An empty tdigest column contains a single row of length 0
+ * The column created contains the specified number of rows of empty clusters.
  *
  * @param stream CUDA stream used for device memory operations and kernel launches.
  * @param mr Device memory resource used to allocate the returned column's device memory.
  *
- * @returns An empty tdigest column.
+ * @returns A tdigest column of empty clusters.
  */
 CUDF_EXPORT
-std::unique_ptr<column> make_empty_tdigest_column(rmm::cuda_stream_view stream,
-                                                  rmm::device_async_resource_ref mr);
+std::unique_ptr<column> make_tdigest_column_of_empty_clusters(size_type num_rows,
+                                                              rmm::cuda_stream_view stream,
+                                                              rmm::device_async_resource_ref mr);
 
 /**
- * @brief Create an empty tdigest scalar.
+ * @brief Create a scalar of an empty tdigest cluster.
  *
- * An empty tdigest scalar is a struct_scalar that contains a single row of length 0
+ * The returned scalar is a struct_scalar that contains a single row of an empty cluster.
  *
  * @param stream CUDA stream used for device memory operations and kernel launches.
  * @param mr Device memory resource used to allocate the returned column's device memory.
  *
- * @returns An empty tdigest scalar.
+ * @returns A scalar of an empty tdigest cluster.
  */
 std::unique_ptr<scalar> make_empty_tdigest_scalar(rmm::cuda_stream_view stream,
                                                   rmm::device_async_resource_ref mr);

cpp/include/cudf_test/tdigest_utilities.cuh

@@ -270,8 +270,8 @@ void tdigest_simple_all_nulls_aggregation(Func op)
     static_cast<column_view>(values).type(), tdigest_gen{}, op, values, delta);
 
   // NOTE: an empty tdigest column still has 1 row.
-  auto expected = cudf::tdigest::detail::make_empty_tdigest_column(
-    cudf::get_default_stream(), cudf::get_current_device_resource_ref());
+  auto expected = cudf::tdigest::detail::make_tdigest_column_of_empty_clusters(
+    1, cudf::get_default_stream(), cudf::get_current_device_resource_ref());
 
   CUDF_TEST_EXPECT_COLUMNS_EQUAL(*result, *expected);
 }
@@ -562,12 +562,12 @@ template <typename MergeFunc>
 void tdigest_merge_empty(MergeFunc merge_op)
 {
   // 3 empty tdigests all in the same group
-  auto a = cudf::tdigest::detail::make_empty_tdigest_column(
-    cudf::get_default_stream(), cudf::get_current_device_resource_ref());
-  auto b = cudf::tdigest::detail::make_empty_tdigest_column(
-    cudf::get_default_stream(), cudf::get_current_device_resource_ref());
-  auto c = cudf::tdigest::detail::make_empty_tdigest_column(
-    cudf::get_default_stream(), cudf::get_current_device_resource_ref());
+  auto a = cudf::tdigest::detail::make_tdigest_column_of_empty_clusters(
+    1, cudf::get_default_stream(), cudf::get_current_device_resource_ref());
+  auto b = cudf::tdigest::detail::make_tdigest_column_of_empty_clusters(
+    1, cudf::get_default_stream(), cudf::get_current_device_resource_ref());
+  auto c = cudf::tdigest::detail::make_tdigest_column_of_empty_clusters(
+    1, cudf::get_default_stream(), cudf::get_current_device_resource_ref());
   std::vector<column_view> cols;
   cols.push_back(*a);
   cols.push_back(*b);
@@ -577,8 +577,8 @@ void tdigest_merge_empty(MergeFunc merge_op)
   auto const delta = 1000;
   auto result      = merge_op(*values, delta);
 
-  auto expected = cudf::tdigest::detail::make_empty_tdigest_column(
-    cudf::get_default_stream(), cudf::get_current_device_resource_ref());
+  auto expected = cudf::tdigest::detail::make_tdigest_column_of_empty_clusters(
+    1, cudf::get_default_stream(), cudf::get_current_device_resource_ref());
 
   CUDF_TEST_EXPECT_COLUMNS_EQUAL(*expected, *result);
 }

cpp/src/quantiles/tdigest/tdigest.cu

@@ -292,32 +292,33 @@ std::unique_ptr<column> make_tdigest_column(size_type num_rows,
   return make_structs_column(num_rows, std::move(children), 0, {}, stream, mr);
 }
 
-std::unique_ptr<column> make_empty_tdigest_column(rmm::cuda_stream_view stream,
-                                                  rmm::device_async_resource_ref mr)
+std::unique_ptr<column> make_tdigest_column_of_empty_clusters(size_type num_rows,
+                                                              rmm::cuda_stream_view stream,
+                                                              rmm::device_async_resource_ref mr)
 {
   auto offsets = cudf::make_fixed_width_column(
-    data_type(type_id::INT32), 2, mask_state::UNALLOCATED, stream, mr);
+    data_type(type_id::INT32), num_rows + 1, mask_state::UNALLOCATED, stream, mr);
   thrust::fill(rmm::exec_policy(stream),
                offsets->mutable_view().begin<size_type>(),
                offsets->mutable_view().end<size_type>(),
                0);
 
-  auto min_col =
-    cudf::make_numeric_column(data_type(type_id::FLOAT64), 1, mask_state::UNALLOCATED, stream, mr);
+  auto min_col = cudf::make_numeric_column(
+    data_type(type_id::FLOAT64), num_rows, mask_state::UNALLOCATED, stream, mr);
   thrust::fill(rmm::exec_policy(stream),
                min_col->mutable_view().begin<double>(),
                min_col->mutable_view().end<double>(),
                0);
-  auto max_col =
-    cudf::make_numeric_column(data_type(type_id::FLOAT64), 1, mask_state::UNALLOCATED, stream, mr);
+  auto max_col = cudf::make_numeric_column(
+    data_type(type_id::FLOAT64), num_rows, mask_state::UNALLOCATED, stream, mr);
   thrust::fill(rmm::exec_policy(stream),
                max_col->mutable_view().begin<double>(),
                max_col->mutable_view().end<double>(),
                0);
 
-  return make_tdigest_column(1,
-                             make_empty_column(type_id::FLOAT64),
-                             make_empty_column(type_id::FLOAT64),
+  return make_tdigest_column(num_rows,
+                             cudf::make_empty_column(type_id::FLOAT64),
+                             cudf::make_empty_column(type_id::FLOAT64),
                              std::move(offsets),
                              std::move(min_col),
                              std::move(max_col),
@@ -338,7 +339,7 @@ std::unique_ptr<column> make_empty_tdigest_column(rmm::cuda_stream_view stream,
 std::unique_ptr<scalar> make_empty_tdigest_scalar(rmm::cuda_stream_view stream,
                                                   rmm::device_async_resource_ref mr)
 {
-  auto contents = make_empty_tdigest_column(stream, mr)->release();
+  auto contents = make_tdigest_column_of_empty_clusters(1, stream, mr)->release();
   return std::make_unique<struct_scalar>(
     std::move(*std::make_unique<table>(std::move(contents.children))), true, stream, mr);
 }