Optimize K-Truss (#4375)

This PR leverages our existing edge triangle count to implement both SG and MG K-Truss as our initial version.
This PR also:

- Exposes the `rx count` in several as our shuffling functions
- Add C and C++ tests for MG K-Truss

Closes #4500

Authors:
  - Joseph Nke (https://github.com/jnke2016)
  - Brad Rees (https://github.com/BradReesWork)

Approvers:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)
  - Seunghwa Kang (https://github.com/seunghwak)
  - Chuck Hastings (https://github.com/ChuckHastings)

URL: #4375

Author: jnke2016

cpp/CMakeLists.txt

@@ -284,6 +284,9 @@ set(CUGRAPH_SOURCES
     src/community/k_truss_sg_v64_e64.cu
     src/community/k_truss_sg_v32_e32.cu
     src/community/k_truss_sg_v32_e64.cu
+    src/community/k_truss_mg_v64_e64.cu
+    src/community/k_truss_mg_v32_e32.cu
+    src/community/k_truss_mg_v32_e64.cu
     src/lookup/lookup_src_dst_mg_v32_e32.cu
     src/lookup/lookup_src_dst_mg_v32_e64.cu
     src/lookup/lookup_src_dst_mg_v64_e64.cu

cpp/include/cugraph/detail/shuffle_wrappers.hpp

@@ -53,7 +53,8 @@ std::tuple<rmm::device_uvector<vertex_t>,
            rmm::device_uvector<vertex_t>,
            std::optional<rmm::device_uvector<weight_t>>,
            std::optional<rmm::device_uvector<edge_t>>,
-           std::optional<rmm::device_uvector<edge_type_id_t>>>
+           std::optional<rmm::device_uvector<edge_type_id_t>>,
+           std::vector<size_t>>
 shuffle_ext_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning(
   raft::handle_t const& handle,
   rmm::device_uvector<vertex_t>&& majors,
@@ -86,14 +87,15 @@ shuffle_ext_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning(
  * (exclusive) vertex ID.
  *
  * @return Tuple of vectors storing shuffled major vertices, minor vertices and optional weights,
- * edge ids and edge types
+ * edge ids and edge types and rx counts
  */
 template <typename vertex_t, typename edge_t, typename weight_t, typename edge_type_id_t>
 std::tuple<rmm::device_uvector<vertex_t>,
            rmm::device_uvector<vertex_t>,
            std::optional<rmm::device_uvector<weight_t>>,
            std::optional<rmm::device_uvector<edge_t>>,
-           std::optional<rmm::device_uvector<edge_type_id_t>>>
+           std::optional<rmm::device_uvector<edge_type_id_t>>,
+           std::vector<size_t>>
 shuffle_int_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning(
   raft::handle_t const& handle,
   rmm::device_uvector<vertex_t>&& majors,

cpp/include/cugraph/mtmg/detail/per_device_edgelist.hpp

@@ -251,7 +251,8 @@ class per_device_edgelist_t {
              store_transposed ? src_[0] : dst_[0],
              tmp_wgt,
              tmp_edge_id,
-             tmp_edge_type) =
+             tmp_edge_type,
+             std::ignore) =
       cugraph::detail::shuffle_ext_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning(
         handle.raft_handle(),
         store_transposed ? std::move(dst_[0]) : std::move(src_[0]),

cpp/src/c_api/graph_functions.cpp

@@ -72,7 +72,7 @@ struct create_vertex_pairs_functor : public cugraph::c_api::abstract_functor {
         second_copy.data(), second_->as_type<vertex_t>(), second_->size_, handle_.get_stream());
 
       if constexpr (multi_gpu) {
-        std::tie(first_copy, second_copy, std::ignore, std::ignore, std::ignore) =
+        std::tie(first_copy, second_copy, std::ignore, std::ignore, std::ignore, std::ignore) =
           cugraph::detail::shuffle_ext_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning<
             vertex_t,
             edge_t,

cpp/src/c_api/graph_mg.cpp

@@ -167,7 +167,8 @@ struct create_graph_functor : public cugraph::c_api::abstract_functor {
                store_transposed ? edgelist_srcs : edgelist_dsts,
                edgelist_weights,
                edgelist_edge_ids,
-               edgelist_edge_types) =
+               edgelist_edge_types,
+               std::ignore) =
         cugraph::detail::shuffle_ext_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning(
           handle_,
           std::move(store_transposed ? edgelist_dsts : edgelist_srcs),

cpp/src/c_api/k_truss.cpp

@@ -60,10 +60,7 @@ struct k_truss_functor : public cugraph::c_api::abstract_functor {
   {
     if constexpr (!cugraph::is_candidate<vertex_t, edge_t, weight_t>::value) {
       unsupported();
-    } else if constexpr (multi_gpu) {
-      unsupported();
     } else {
-      // k_truss expects store_transposed == false
       if constexpr (store_transposed) {
         error_code_ = cugraph::c_api::
           transpose_storage<vertex_t, edge_t, weight_t, store_transposed, multi_gpu>(

cpp/src/community/detail/refine_impl.cuh

@@ -627,6 +627,7 @@ refine_clustering(
                store_transposed ? d_srcs : d_dsts,
                d_weights,
                std::ignore,
+               std::ignore,
                std::ignore) =
         cugraph::detail::shuffle_ext_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning<
           vertex_t,

cpp/src/community/edge_triangle_count_impl.cuh

@@ -250,7 +250,7 @@ edge_property_t<graph_view_t<vertex_t, edge_t, false, multi_gpu>, edge_t> edge_t
                          handle.get_stream());
 
       // There are still multiple copies here but is it worth sorting and reducing again?
-      std::tie(pair_srcs, pair_dsts, std::ignore, pair_count, std::ignore) =
+      std::tie(pair_srcs, pair_dsts, std::ignore, pair_count, std::ignore, std::ignore) =
         shuffle_int_vertex_pairs_with_values_to_local_gpu_by_edge_partitioning<vertex_t,
                                                                                edge_t,
                                                                                weight_t,