Ktruss implementation (#4059)

Implements SG and MG ktruss using graph primitives and drop `cuHornet`. Closes #3447 Closes #3448 Closes #3449 Closes #3450 Closes #3451 Closes #3452 Closes #3453 Authors: - Joseph Nke (https://github.com/jnke2016) - Seunghwa Kang (https://github.com/seunghwak) Approvers: - Brad Rees (https://github.com/BradReesWork) - Chuck Hastings (https://github.com/ChuckHastings) - Seunghwa Kang (https://github.com/seunghwak) - Rick Ratzel (https://github.com/rlratzel) URL: #4059
rapidsai · Mar 26, 2024 · f753e51 · f753e51
1 parent 8706115
commit f753e51
Show file tree

Hide file tree

Showing 15 changed files with 1,693 additions and 334 deletions.
diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
@@ -37,30 +37,6 @@ if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND
     message(FATAL_ERROR "GCC compiler must be at least 9.3")
 endif()
 
-# Remove the following archs from CMAKE_CUDA_ARCHITECTURES that
-# cuhornet currently doesn't support
-#
-# >= 90
-set(supported_archs "70" "72" "75" "80" "86" "89" "90")
-foreach( arch IN LISTS CMAKE_CUDA_ARCHITECTURES)
-    string(REPLACE "-real" "" arch ${arch})
-    if( arch IN_LIST supported_archs )
-        list(APPEND usable_arch_values ${arch})
-    endif()
-endforeach()
-# Make sure everything but the 'newest' arch
-# is marked as `-real` so we only generate PTX for
-# arch > 86
-list(POP_BACK usable_arch_values latest_arch)
-list(TRANSFORM usable_arch_values APPEND "-real")
-if (usable_arch_values)
-    list(APPEND usable_arch_values ${latest_arch})
-else()
-    list(APPEND usable_arch_values ${latest_arch}-real)
-endif()
-
-set(CMAKE_CUDA_ARCHITECTURES ${usable_arch_values})
-
 # Write the version header
 rapids_cmake_write_version_file(include/cugraph/version_config.hpp)
 rapids_cmake_write_version_file(include/cugraph_c/version_config.hpp)
@@ -168,7 +144,6 @@ if(USE_CUGRAPH_OPS)
 endif()
 
 include(cmake/thirdparty/get_nccl.cmake)
-include(cmake/thirdparty/get_cuhornet.cmake)
 
 if (BUILD_CUGRAPH_MTMG_TESTS)
   include(cmake/thirdparty/get_ucp.cmake)
@@ -197,6 +172,7 @@ set(CUGRAPH_SOURCES
     src/community/detail/common_methods_sg.cu
     src/community/detail/refine_sg.cu
     src/community/detail/refine_mg.cu
+    src/community/edge_triangle_count_sg.cu
     src/community/detail/maximal_independent_moves_sg.cu
     src/community/detail/maximal_independent_moves_mg.cu
     src/detail/utility_wrappers.cu
@@ -222,10 +198,10 @@ set(CUGRAPH_SOURCES
     src/community/ecg_sg.cu
     src/community/ecg_mg.cu
     src/community/legacy/louvain.cu
-    src/community/legacy/ktruss.cu
     src/community/legacy/ecg.cu
     src/community/egonet_sg.cu
     src/community/egonet_mg.cu
+    src/community/k_truss_sg.cu
     src/sampling/random_walks.cu
     src/sampling/random_walks_sg.cu
     src/sampling/detail/prepare_next_frontier_sg.cu
@@ -391,7 +367,6 @@ if (USE_CUGRAPH_OPS)
             $<$<BOOL:${CUDA_STATIC_RUNTIME}>:raft::raft>
             $<$<BOOL:${CUDA_STATIC_RUNTIME}>:${COMPILED_RAFT_LIB}>
             cuco::cuco
-            cugraph::cuHornet
             NCCL::NCCL
     )
 else()
@@ -404,7 +379,6 @@ else()
             $<$<BOOL:${CUDA_STATIC_RUNTIME}>:raft::raft>
             $<$<BOOL:${CUDA_STATIC_RUNTIME}>:${COMPILED_RAFT_LIB}>
             cuco::cuco
-            cugraph::cuHornet
             NCCL::NCCL
     )
 endif()
@@ -427,6 +401,7 @@ add_library(cugraph_c
         src/c_api/eigenvector_centrality.cpp
         src/c_api/betweenness_centrality.cpp
         src/c_api/core_number.cpp
+        src/c_api/k_truss.cpp
         src/c_api/core_result.cpp
         src/c_api/extract_ego.cpp
         src/c_api/ecg.cpp
@@ -455,7 +430,6 @@ add_library(cugraph_c
         src/c_api/weakly_connected_components.cpp
         src/c_api/strongly_connected_components.cpp
         src/c_api/allgather.cpp
-        src/c_api/legacy_k_truss.cpp
         )
 add_library(cugraph::cugraph_c ALIAS cugraph_c)
 

diff --git a/cpp/cmake/thirdparty/get_cuhornet.cmake b/cpp/cmake/thirdparty/get_cuhornet.cmake
diff --git a/cpp/include/cugraph/algorithms.hpp b/cpp/include/cugraph/algorithms.hpp
@@ -427,41 +427,6 @@ void connected_components(legacy::GraphCSRView<VT, ET, WT> const& graph,
                           cugraph_cc_t connectivity_type,
                           VT* labels);
 
-/**
- * @brief     Compute k truss for a graph  ** temporary
- *
- * K Truss is the maximal subgraph of a graph which contains at least three
- * vertices where every edge is incident to at least k-2 triangles.
- *
- * This version is a temporary solution to clean up python integration through the C API.
- *
- * This version is only supported SG.
- *
- * @throws                  cugraph::logic_error with a custom message when an error
- * occurs.
- *
- * @tparam vertex_t         Type of vertex identifiers. Supported value : int (signed, 32-bit)
- * @tparam weight_t         Type of edge weights. Supported values : float or double.
- *
- * @param[in] handle        Library handle (RAFT).
- * @param[in] src           Source vertices from COO
- * @param[in] dst           Destination vertices from COO
- * @param[in] wgt           Optional edge weights from COO
- * @param[in] k             The order of the truss
- * @return                  Tuple containing extracted src, dst and optional weights for the
- * subgraph
- */
-template <typename vertex_t, typename weight_t>
-std::tuple<rmm::device_uvector<vertex_t>,
-           rmm::device_uvector<vertex_t>,
-           std::optional<rmm::device_uvector<weight_t>>>
-k_truss_subgraph(raft::handle_t const& handle,
-                 raft::device_span<vertex_t> src,
-                 raft::device_span<vertex_t> dst,
-                 std::optional<raft::device_span<weight_t>> wgt,
-                 size_t number_of_vertices,
-                 int k);
-
 /**
  * @brief      Compute Hungarian algorithm on a weighted bipartite graph
  *
@@ -1842,7 +1807,7 @@ void weakly_connected_components(raft::handle_t const& handle,
 enum class k_core_degree_type_t { IN = 0, OUT = 1, INOUT = 2 };
 
 /**
- * @brief   Compute core numbers of individual vertices from K-core decomposition.
+ * @brief   Compute core numbers of individual vertices from K-Core decomposition.
  *
  * The input graph should not have self-loops nor multi-edges. Currently, only undirected graphs are
  * supported.
@@ -1855,11 +1820,11 @@ enum class k_core_degree_type_t { IN = 0, OUT = 1, INOUT = 2 };
  * handles to various CUDA libraries) to run graph algorithms.
  * @param graph_view Graph view object.
  * @param core_numbers Pointer to the output core number array.
- * @param degree_type Dictate whether to compute the K-core decomposition based on in-degrees,
+ * @param degree_type Dictate whether to compute the K-Core decomposition based on in-degrees,
  * out-degrees, or in-degrees + out_degrees.
- * @param k_first Find K-cores from K = k_first. Any vertices that do not belong to k_first-core
+ * @param k_first Find K-Cores from K = k_first. Any vertices that do not belong to k_first-core
  * will have core numbers of 0.
- * @param k_last Find K-cores to K = k_last. Any vertices that belong to (k_last)-core will have
+ * @param k_last Find K-Cores to K = k_last. Any vertices that belong to (k_last)-core will have
  * their core numbers set to their degrees on k_last-core.
  * @param do_expensive_check A flag to run expensive checks for input arguments (if set to `true`).
  */
@@ -1873,7 +1838,7 @@ void core_number(raft::handle_t const& handle,
                  bool do_expensive_check = false);
 
 /**
- * @brief   Extract K Core of a graph
+ * @brief   Extract K-Core of a graph
  *
  * @throws     cugraph::logic_error when an error occurs.
  *
@@ -1884,7 +1849,7 @@ void core_number(raft::handle_t const& handle,
  * @param  graph_view      Graph view object.
  * @param edge_weight_view Optional view object holding edge weights for @p graph_view.
  * @param  k               Order of the core. This value must not be negative.
- * @param degree_type Optional parameter to dictate whether to compute the K-core decomposition
+ * @param degree_type Optional parameter to dictate whether to compute the K-Core decomposition
  *                    based on in-degrees, out-degrees, or in-degrees + out_degrees.  One of @p
  *                    degree_type and @p core_numbers must be specified.
  * @param  core_numbers    Optional output from core_number algorithm.  If not specified then
@@ -2040,6 +2005,32 @@ void triangle_count(raft::handle_t const& handle,
                     raft::device_span<edge_t> counts,
                     bool do_expensive_check = false);
 
+/*
+ * @brief Compute K-Truss.
+ *
+ * Extract the K-Truss subgraph of a graph
+ *
+ * @tparam vertex_t Type of vertex identifiers. Needs to be an integral type.
+ * @tparam edge_t Type of edge identifiers. Needs to be an integral type.
+ * @tparam multi_gpu Flag indicating whether template instantiation should target single-GPU (false)
+ * @param handle RAFT handle object to encapsulate resources (e.g. CUDA stream, communicator, and
+ * handles to various CUDA libraries) to run graph algorithms.
+ * @param graph_view Graph view object.
+ * @param edge_weight_view Optional view object holding edge weights for @p graph_view.
+ * @param k The desired k to be used for extracting the K-Truss subgraph
+ * @param do_expensive_check A flag to run expensive checks for input arguments (if set to `true`).
+ * @return edge list of the K-Truss subgraph
+ */
+template <typename vertex_t, typename edge_t, typename weight_t, bool multi_gpu>
+std::tuple<rmm::device_uvector<vertex_t>,
+           rmm::device_uvector<vertex_t>,
+           std::optional<rmm::device_uvector<weight_t>>>
+k_truss(raft::handle_t const& handle,
+        graph_view_t<vertex_t, edge_t, false, multi_gpu> const& graph_view,
+        std::optional<edge_property_view_t<edge_t, weight_t const*>> edge_weight_view,
+        edge_t k,
+        bool do_expensive_check = false);
+
 /**
  * @brief     Compute Jaccard similarity coefficient
  *

diff --git a/cpp/include/cugraph/utilities/graph_traits.hpp b/cpp/include/cugraph/utilities/graph_traits.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021-2023, NVIDIA CORPORATION.
+ * Copyright (c) 2021-2024, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -67,8 +67,8 @@ struct is_candidate {
 };
 
 // meta-function that constrains
-// all 3 template param candidates where vertex_t and edge_t
-// are restricted to int32_t:
+// vertex_t and edge_t are restricted to int32_t:
+// FIXME: Drop this functor as it was only used by legacy K-Truss
 //
 template <typename vertex_t, typename edge_t, typename weight_t>
 struct is_candidate_legacy {

diff --git a/cpp/src/c_api/legacy_k_truss.cpp → cpp/src/c_api/k_truss.cpp b/cpp/src/c_api/legacy_k_truss.cpp → cpp/src/c_api/k_truss.cpp
@@ -58,7 +58,7 @@ struct k_truss_functor : public cugraph::c_api::abstract_functor {
             bool multi_gpu>
   void operator()()
   {
-    if constexpr (!cugraph::is_candidate_legacy<vertex_t, edge_t, weight_t>::value) {
+    if constexpr (!cugraph::is_candidate<vertex_t, edge_t, weight_t>::value) {
       unsupported();
     } else if constexpr (multi_gpu) {
       unsupported();
@@ -81,26 +81,14 @@ struct k_truss_functor : public cugraph::c_api::abstract_functor {
       auto number_map = reinterpret_cast<rmm::device_uvector<vertex_t>*>(graph_->number_map_);
 
       auto graph_view = graph->view();
-      rmm::device_uvector<vertex_t> src(0, handle_.get_stream());
-      rmm::device_uvector<vertex_t> dst(0, handle_.get_stream());
-      std::optional<rmm::device_uvector<weight_t>> wgt{std::nullopt};
 
-      std::tie(src, dst, wgt, std::ignore) = cugraph::decompress_to_edgelist(
-        handle_,
-        graph_view,
-        edge_weights ? std::make_optional(edge_weights->view()) : std::nullopt,
-        std::optional<cugraph::edge_property_view_t<edge_t, edge_t const*>>{std::nullopt},
-        std::optional<raft::device_span<vertex_t const>>(std::nullopt),
-        do_expensive_check_);
-
-      auto [result_src, result_dst, result_wgt] = cugraph::k_truss_subgraph(
-        handle_,
-        raft::device_span<vertex_t>(src.data(), src.size()),
-        raft::device_span<vertex_t>(dst.data(), dst.size()),
-        wgt ? std::make_optional(raft::device_span<weight_t>(wgt->data(), wgt->size()))
-            : std::nullopt,
-        graph_view.number_of_vertices(),
-        k_);
+      auto [result_src, result_dst, result_wgt] =
+        cugraph::k_truss<vertex_t, edge_t, weight_t, multi_gpu>(
+          handle_,
+          graph_view,
+          edge_weights ? std::make_optional(edge_weights->view()) : std::nullopt,
+          k_,
+          do_expensive_check_);
 
       cugraph::unrenumber_int_vertices<vertex_t, multi_gpu>(
         handle_,
@@ -127,9 +115,9 @@ struct k_truss_functor : public cugraph::c_api::abstract_functor {
       result_ = new cugraph::c_api::cugraph_induced_subgraph_result_t{
         new cugraph::c_api::cugraph_type_erased_device_array_t(result_src, graph_->vertex_type_),
         new cugraph::c_api::cugraph_type_erased_device_array_t(result_dst, graph_->vertex_type_),
-        wgt ? new cugraph::c_api::cugraph_type_erased_device_array_t(*result_wgt,
-                                                                     graph_->weight_type_)
-            : NULL,
+        result_wgt ? new cugraph::c_api::cugraph_type_erased_device_array_t(*result_wgt,
+                                                                            graph_->weight_type_)
+                   : NULL,
         NULL,
         NULL,
         new cugraph::c_api::cugraph_type_erased_device_array_t(edge_offsets,