Reduce peak memory requirement in graph creation (part 1/2)

cpp/include/cugraph/detail/shuffle_wrappers.hpp

@@ -76,8 +76,8 @@ rmm::device_uvector<vertex_t> shuffle_vertices_by_gpu_id(
  * @param[in/out] d_edgelist_minors Vertex IDs for columns (if the graph adjacency matrix is stored
  * as is) or rows (if the graph adjacency matrix is stored transposed)
  * @param[in/out] d_edgelist_weights Optional edge weights
- * @param[in] groupby_and_count_local_partition If set to true, groupby and count edges based on
- * (local partition ID, GPU ID) pairs (where GPU IDs are computed by applying the
+ * @param[in] groupby_and_count_local_partition_by_minor If set to true, groupby and count edges
+ * based on (local partition ID, GPU ID) pairs (where GPU IDs are computed by applying the
  * compute_gpu_id_from_vertex_t function to the minor vertex ID). If set to false, groupby and count
  * edges by just local partition ID.
  *
@@ -91,7 +91,7 @@ rmm::device_uvector<size_t> groupby_and_count_edgelist_by_local_partition_id(
   rmm::device_uvector<vertex_t>& d_edgelist_majors,
   rmm::device_uvector<vertex_t>& d_edgelist_minors,
   std::optional<rmm::device_uvector<weight_t>>& d_edgelist_weights,
-  bool groupby_and_count_local_partition = false);
+  bool groupby_and_count_local_partition_by_minor = false);
 
 }  // namespace detail
 }  // namespace cugraph

cpp/include/cugraph/graph_functions.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -37,9 +37,6 @@ struct renumber_meta_t<vertex_t, edge_t, multi_gpu, std::enable_if_t<multi_gpu>>
   edge_t number_of_edges{};
   partition_t<vertex_t> partition{};
   std::vector<vertex_t> segment_offsets{};
-
-  vertex_t num_local_unique_edge_majors{};
-  vertex_t num_local_unique_edge_minors{};
 };
 
 template <typename vertex_t, typename edge_t, bool multi_gpu>

cpp/include/cugraph/prims/copy_v_transform_reduce_key_aggregated_out_nbr.cuh

@@ -482,7 +482,7 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
       rmm::device_uvector<weight_t> rx_key_aggregated_edge_weights(0, handle.get_stream());
       std::forward_as_tuple(
         std::tie(rx_major_vertices, rx_minor_keys, rx_key_aggregated_edge_weights), std::ignore) =
-        groupby_gpuid_and_shuffle_values(
+        groupby_gpu_id_and_shuffle_values(
           col_comm,
           triplet_first,
           triplet_first + tmp_major_vertices.size(),

cpp/include/cugraph/prims/transform_reduce_by_adj_matrix_row_col_key_e.cuh

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -505,7 +505,7 @@ transform_reduce_by_adj_matrix_row_col_key_e(
       rmm::device_uvector<vertex_t> rx_unique_keys(0, handle.get_stream());
       auto rx_value_for_unique_key_buffer = allocate_dataframe_buffer<T>(0, handle.get_stream());
       std::tie(rx_unique_keys, rx_value_for_unique_key_buffer, std::ignore) =
-        groupby_gpuid_and_shuffle_kv_pairs(
+        groupby_gpu_id_and_shuffle_kv_pairs(
           comm,
           tmp_keys.begin(),
           tmp_keys.end(),

cpp/include/cugraph/utilities/collect_comm.cuh

@@ -103,7 +103,7 @@ collect_values_for_keys(raft::comms::comms_t const& comm,
   {
     rmm::device_uvector<vertex_t> rx_unique_keys(0, stream_view);
     std::vector<size_t> rx_value_counts{};
-    std::tie(rx_unique_keys, rx_value_counts) = groupby_gpuid_and_shuffle_values(
+    std::tie(rx_unique_keys, rx_value_counts) = groupby_gpu_id_and_shuffle_values(
       comm,
       unique_keys.begin(),
       unique_keys.end(),
@@ -228,7 +228,7 @@ collect_values_for_unique_keys(raft::comms::comms_t const& comm,
   {
     rmm::device_uvector<vertex_t> rx_unique_keys(0, stream_view);
     std::vector<size_t> rx_value_counts{};
-    std::tie(rx_unique_keys, rx_value_counts) = groupby_gpuid_and_shuffle_values(
+    std::tie(rx_unique_keys, rx_value_counts) = groupby_gpu_id_and_shuffle_values(
       comm,
       unique_keys.begin(),
       unique_keys.end(),

cpp/include/cugraph/utilities/cython.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -588,7 +588,7 @@ template <typename vertex_t, typename edge_t, typename weight_t>
 std::unique_ptr<major_minor_weights_t<vertex_t, edge_t, weight_t>> call_shuffle(
   raft::handle_t const& handle,
   vertex_t*
-    edgelist_major_vertices,  // [IN / OUT]: groupby_gpuid_and_shuffle_values() sorts in-place
+    edgelist_major_vertices,  // [IN / OUT]: groupby_gpu_id_and_shuffle_values() sorts in-place
   vertex_t* edgelist_minor_vertices,  // [IN / OUT]
   weight_t* edgelist_weights,         // [IN / OUT]
   edge_t num_edgelist_edges);