Rename sort_and_shuffle to groupby_gpuid_and_shuffle (#1392)

Rename to better reflect what this function should do than how it is currently implemented (which can change in the future for better performance).

Authors:
  - Seunghwa Kang (@seunghwak)

Approvers:
  - Chuck Hastings (@ChuckHastings)
  - Alex Fender (@afender)

URL: #1392

cpp/include/patterns/copy_v_transform_reduce_key_aggregated_out_nbr.cuh

@@ -234,7 +234,7 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
 
     rmm::device_uvector<vertex_t> rx_unique_keys(0, handle.get_stream());
     std::vector<size_t> rx_value_counts{};
-    std::tie(rx_unique_keys, rx_value_counts) = sort_and_shuffle_values(
+    std::tie(rx_unique_keys, rx_value_counts) = groupby_gpuid_and_shuffle_values(
       comm,
       unique_keys.begin(),
       unique_keys.end(),
@@ -372,7 +372,7 @@ 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) =
-        sort_and_shuffle_values(
+        groupby_gpuid_and_shuffle_values(
           sub_comm,
           triplet_first,
           triplet_first + tmp_major_vertices.size(),

cpp/include/patterns/transform_reduce_by_adj_matrix_row_col_key_e.cuh

@@ -241,7 +241,7 @@ 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) =
-        sort_and_shuffle_kv_pairs(
+        groupby_gpuid_and_shuffle_kv_pairs(
           comm,
           unique_keys.begin(),
           unique_keys.end(),

cpp/include/utilities/cython.hpp

@@ -451,7 +451,8 @@ std::unique_ptr<cy_multi_edgelists_t> call_egonet(raft::handle_t const& handle,
 template <typename vertex_t, typename edge_t, typename weight_t>
 std::unique_ptr<major_minor_weights_t<vertex_t, weight_t>> call_shuffle(
   raft::handle_t const& handle,
-  vertex_t* edgelist_major_vertices,  // [IN / OUT]: sort_and_shuffle_values() sorts in-place
+  vertex_t*
+    edgelist_major_vertices,  // [IN / OUT]: groupby_gpuid_and_shuffle_values() sorts in-place
   vertex_t* edgelist_minor_vertices,  // [IN / OUT]
   weight_t* edgelist_weights,         // [IN / OUT]
   edge_t num_edgelist_edges,

cpp/include/utilities/shuffle_comm.cuh

@@ -232,11 +232,11 @@ auto shuffle_values(raft::comms::comms_t const &comm,
 }
 
 template <typename ValueIterator, typename ValueToGPUIdOp>
-auto sort_and_shuffle_values(raft::comms::comms_t const &comm,
-                             ValueIterator tx_value_first /* [INOUT */,
-                             ValueIterator tx_value_last /* [INOUT */,
-                             ValueToGPUIdOp value_to_gpu_id_op,
-                             cudaStream_t stream)
+auto groupby_gpuid_and_shuffle_values(raft::comms::comms_t const &comm,
+                                      ValueIterator tx_value_first /* [INOUT */,
+                                      ValueIterator tx_value_last /* [INOUT */,
+                                      ValueToGPUIdOp value_to_gpu_id_op,
+                                      cudaStream_t stream)
 {
   auto const comm_size = comm.get_size();
 
@@ -275,12 +275,12 @@ auto sort_and_shuffle_values(raft::comms::comms_t const &comm,
 }
 
 template <typename VertexIterator, typename ValueIterator, typename KeyToGPUIdOp>
-auto sort_and_shuffle_kv_pairs(raft::comms::comms_t const &comm,
-                               VertexIterator tx_key_first /* [INOUT */,
-                               VertexIterator tx_key_last /* [INOUT */,
-                               ValueIterator tx_value_first /* [INOUT */,
-                               KeyToGPUIdOp key_to_gpu_id_op,
-                               cudaStream_t stream)
+auto groupby_gpuid_and_shuffle_kv_pairs(raft::comms::comms_t const &comm,
+                                        VertexIterator tx_key_first /* [INOUT */,
+                                        VertexIterator tx_key_last /* [INOUT */,
+                                        ValueIterator tx_value_first /* [INOUT */,
+                                        KeyToGPUIdOp key_to_gpu_id_op,
+                                        cudaStream_t stream)
 {
   auto d_tx_value_counts = detail::sort_and_count(
     comm, tx_key_first, tx_key_last, tx_value_first, key_to_gpu_id_op, stream);

cpp/src/experimental/coarsen_graph.cu

@@ -329,7 +329,7 @@ coarsen_graph(
     std::forward_as_tuple(
       std::tie(rx_edgelist_major_vertices, rx_edgelist_minor_vertices, rx_edgelist_weights),
       std::ignore) =
-      sort_and_shuffle_values(
+      groupby_gpuid_and_shuffle_values(
         handle.get_comms(),
         edge_first,
         edge_first + coarsened_edgelist_major_vertices.size(),
@@ -371,7 +371,7 @@ coarsen_graph(
     handle.get_stream());
 
   rmm::device_uvector<vertex_t> rx_unique_labels(0, handle.get_stream());
-  std::tie(rx_unique_labels, std::ignore) = sort_and_shuffle_values(
+  std::tie(rx_unique_labels, std::ignore) = groupby_gpuid_and_shuffle_values(
     handle.get_comms(),
     unique_labels.begin(),
     unique_labels.end(),

cpp/src/experimental/relabel.cu

@@ -106,7 +106,7 @@ void relabel(raft::handle_t const& handle,
           thrust::make_tuple(label_pair_old_labels.begin(), label_pair_new_labels.begin()));
         std::forward_as_tuple(std::tie(rx_label_pair_old_labels, rx_label_pair_new_labels),
                               std::ignore) =
-          sort_and_shuffle_values(
+          groupby_gpuid_and_shuffle_values(
             handle.get_comms(),
             pair_first,
             pair_first + num_label_pairs,
@@ -142,7 +142,7 @@ void relabel(raft::handle_t const& handle,
       {
         rmm::device_uvector<vertex_t> rx_unique_old_labels(0, handle.get_stream());
         std::vector<size_t> rx_value_counts{};
-        std::tie(rx_unique_old_labels, rx_value_counts) = sort_and_shuffle_values(
+        std::tie(rx_unique_old_labels, rx_value_counts) = groupby_gpuid_and_shuffle_values(
           handle.get_comms(),
           unique_old_labels.begin(),
           unique_old_labels.end(),

cpp/src/experimental/renumber_edgelist.cu

@@ -151,14 +151,14 @@ rmm::device_uvector<vertex_t> compute_renumber_map(
     auto pair_first = thrust::make_zip_iterator(thrust::make_tuple(labels.begin(), counts.begin()));
     rmm::device_uvector<vertex_t> rx_labels(0, handle.get_stream());
     rmm::device_uvector<edge_t> rx_counts(0, handle.get_stream());
-    std::forward_as_tuple(std::tie(rx_labels, rx_counts), std::ignore) = sort_and_shuffle_values(
-      comm,
-      pair_first,
-      pair_first + labels.size(),
-      [key_func = detail::compute_gpu_id_from_vertex_t<vertex_t>{comm_size}] __device__(auto val) {
-        return key_func(thrust::get<0>(val));
-      },
-      handle.get_stream());
+    std::forward_as_tuple(std::tie(rx_labels, rx_counts), std::ignore) =
+      groupby_gpuid_and_shuffle_values(
+        comm,
+        pair_first,
+        pair_first + labels.size(),
+        [key_func = detail::compute_gpu_id_from_vertex_t<vertex_t>{comm_size}] __device__(
+          auto val) { return key_func(thrust::get<0>(val)); },
+        handle.get_stream());
 
     labels.resize(rx_labels.size(), handle.get_stream());
     counts.resize(labels.size(), handle.get_stream());
@@ -309,7 +309,7 @@ void expensive_check_edgelist(
         handle.get_stream());
 
       rmm::device_uvector<vertex_t> rx_unique_edge_vertices(0, handle.get_stream());
-      std::tie(rx_unique_edge_vertices, std::ignore) = sort_and_shuffle_values(
+      std::tie(rx_unique_edge_vertices, std::ignore) = groupby_gpuid_and_shuffle_values(
         handle.get_comms(),
         unique_edge_vertices.begin(),
         unique_edge_vertices.end(),

cpp/src/utilities/cython.cu

@@ -749,7 +749,8 @@ void call_sssp(raft::handle_t const& handle,
 template <typename vertex_t, typename edge_t, typename weight_t>
 std::unique_ptr<major_minor_weights_t<vertex_t, weight_t>> call_shuffle(
   raft::handle_t const& handle,
-  vertex_t* edgelist_major_vertices,  // [IN / OUT]: sort_and_shuffle_values() sorts in-place
+  vertex_t*
+    edgelist_major_vertices,  // [IN / OUT]: groupby_gpuid_and_shuffle_values() sorts in-place
   vertex_t* edgelist_minor_vertices,  // [IN / OUT]
   weight_t* edgelist_weights,         // [IN / OUT]
   edge_t num_edgelist_edges,
@@ -770,7 +771,7 @@ std::unique_ptr<major_minor_weights_t<vertex_t, weight_t>> call_shuffle(
   std::forward_as_tuple(
     std::tie(ptr_ret->get_major(), ptr_ret->get_minor(), ptr_ret->get_weights()),
     std::ignore) =
-    cugraph::experimental::sort_and_shuffle_values(
+    cugraph::experimental::groupby_gpuid_and_shuffle_values(
       comm,  // handle.get_comms(),
       zip_edge,
       zip_edge + num_edgelist_edges,

python/cugraph/structure/graph_primtypes.pxd

@@ -200,7 +200,7 @@ cdef extern from "utilities/cython.hpp" namespace "cugraph::cython":
         vertex_t get_part_matrix_partition_minor_first()
         vertex_t get_part_matrix_partition_minor_last()        
 
-# 4. `sort_and_shuffle_values()` wrapper:
+# 4. `groupby_gpuid_and_shuffle_values()` wrapper:
 #
 cdef extern from "utilities/cython.hpp" namespace "cugraph::cython":