update forked from release branch-0.19

CHANGELOG.md

@@ -6,9 +6,60 @@
 
 ## Bug Fixes
 
-# 0.18.0
-
-Please see https://github.com/rapidsai/cugraph/releases/tag/branch-0.18-latest for the latest changes to this development branch.
+# cuGraph 0.18.0 (24 Feb 2021)
+
+## Bug Fixes 🐛
+
+- Fixed TSP returned routes (#1412) @hlinsen
+- Updated CI scripts to use a different error handling convention, updated LD_LIBRARY_PATH for project flash runs (#1386) @rlratzel
+- Bug fixes for MNMG coarsen_graph, renumber_edgelist, relabel (#1364) @seunghwak
+- Set a specific known working commit hash for gunrock instead of "dev" (#1336) @rlratzel
+- Updated git utils used by copyright.py for compatibility with current CI env (#1325) @rlratzel
+- Fix MNMG Louvain tests on Pascal architecture (#1322) @ChuckHastings
+- FIX Set bash trap after PATH is updated (#1321) @dillon-cullinan
+- Fix graph nodes function and renumbering from series (#1319) @Iroy30
+- Fix Branch 0.18 merge 0.17 (#1314) @BradReesWork
+- Fix EXPERIMENTAL_LOUVAIN_TEST on Pascal (#1312) @ChuckHastings
+- Updated cuxfilter to 0.18, removed datashader indirect dependency in conda dev .yml files (#1311) @rlratzel
+- Update SG PageRank C++ tests (#1307) @seunghwak
+
+## Documentation 📖
+
+- Enabled MultiGraph class and tests, updated SOURCEBUILD.md to include the latest build.sh options (#1351) @rlratzel
+
+## New Features 🚀
+
+- New EgoNet extractor (#1365) @afender
+- Implement induced subgraph extraction primitive (SG C++) (#1354) @seunghwak
+
+## Improvements 🛠️
+
+- Update stale GHA with exemptions & new labels (#1413) @mike-wendt
+- Add GHA to mark issues/prs as stale/rotten (#1408) @Ethyling
+- update subgraph tests and remove legacy pagerank (#1378) @Iroy30
+- Update the conda environments and README file (#1369) @BradReesWork
+- Prepare Changelog for Automation (#1368) @ajschmidt8
+- Update CMakeLists.txt files for consistency with RAPIDS and to support cugraph as an external project and other tech debt removal (#1367) @rlratzel
+- Use new coarsen_graph primitive in Louvain (#1362) @ChuckHastings
+- Added initial infrastructure for MG C++ testing and a Pagerank MG test using it (#1361) @rlratzel
+- Add SG TSP (#1360) @hlinsen
+- Build a Dendrogram class, adapt Louvain/Leiden/ECG to use it (#1359) @ChuckHastings
+- Auto-label PRs based on their content (#1358) @jolorunyomi
+- Implement MNMG Renumber (#1355) @aschaffer
+- Enabling pytest code coverage output by default (#1352) @jnke2016
+- Added configuration for new cugraph-doc-codeowners review group (#1344) @rlratzel
+- API update to match RAFT PR #120 (#1343) @drobison00
+- Pin gunrock to v1.2 for version 0.18 (#1342) @ChuckHastings
+- Fix #1340 - Use generic from_edgelist() methods (#1341) @miguelusque
+- Using RAPIDS_DATASET_ROOT_DIR env var in place of absolute path to datasets in tests (#1337) @jnke2016
+- Expose dense implementation of Hungarian algorithm (#1333) @ChuckHastings
+- SG Pagerank transition (#1332) @Iroy30
+- improving error checking and docs (#1327) @BradReesWork
+- Fix MNMG cleanup exceptions (#1326) @Iroy30
+- Create labeler.yml (#1318) @jolorunyomi
+- Updates to support nightly MG test automation (#1308) @rlratzel
+- Add C++ graph functions (coarsen_grpah, renumber_edgelist, relabel) and primitvies (transform_reduce_by_adj_matrix_row_key, transform_reduce_by_adj_matrix_col_key, copy_v_transform_reduce_key_aggregated_out_nbr) (#1257) @seunghwak
+>>>>>>> upstream/branch-0.18
 
 # cuGraph 0.17.0 (10 Dec 2020)
 ## New Features

cpp/CMakeLists.txt

@@ -298,7 +298,7 @@ else(DEFINED ENV{RAFT_PATH})
   FetchContent_Declare(
     raft
     GIT_REPOSITORY    https://github.com/rapidsai/raft.git
-    GIT_TAG           4a79adcb0c0e87964dcdc9b9122f242b5235b702
+    GIT_TAG           a3461b201ea1c9f61571f1927274f739e775d2d2
     SOURCE_SUBDIR     raft
   )
 

cpp/include/experimental/detail/graph_utils.cuh

@@ -25,6 +25,7 @@
 #include <rmm/device_uvector.hpp>
 
 #include <thrust/sort.h>
+#include <thrust/tabulate.h>
 #include <thrust/transform.h>
 #include <cuco/detail/hash_functions.cuh>
 
@@ -39,7 +40,7 @@ namespace detail {
 // compute the numbers of nonzeros in rows (of the graph adjacency matrix, if store_transposed =
 // false) or columns (of the graph adjacency matrix, if store_transposed = true)
 template <typename vertex_t, typename edge_t>
-rmm::device_uvector<edge_t> compute_major_degree(
+rmm::device_uvector<edge_t> compute_major_degrees(
   raft::handle_t const &handle,
   std::vector<edge_t const *> const &adj_matrix_partition_offsets,
   partition_t<vertex_t> const &partition)
@@ -120,7 +121,7 @@ rmm::device_uvector<edge_t> compute_major_degree(
 // compute the numbers of nonzeros in rows (of the graph adjacency matrix, if store_transposed =
 // false) or columns (of the graph adjacency matrix, if store_transposed = true)
 template <typename vertex_t, typename edge_t>
-rmm::device_uvector<edge_t> compute_major_degree(
+rmm::device_uvector<edge_t> compute_major_degrees(
   raft::handle_t const &handle,
   std::vector<rmm::device_uvector<edge_t>> const &adj_matrix_partition_offsets,
   partition_t<vertex_t> const &partition)
@@ -131,7 +132,22 @@ rmm::device_uvector<edge_t> compute_major_degree(
                  adj_matrix_partition_offsets.end(),
                  tmp_offsets.begin(),
                  [](auto const &offsets) { return offsets.data(); });
-  return compute_major_degree(handle, tmp_offsets, partition);
+  return compute_major_degrees(handle, tmp_offsets, partition);
+}
+
+// compute the numbers of nonzeros in rows (of the graph adjacency matrix, if store_transposed =
+// false) or columns (of the graph adjacency matrix, if store_transposed = true)
+template <typename vertex_t, typename edge_t>
+rmm::device_uvector<edge_t> compute_major_degrees(raft::handle_t const &handle,
+                                                  edge_t const *offsets,
+                                                  vertex_t number_of_vertices)
+{
+  rmm::device_uvector<edge_t> degrees(number_of_vertices, handle.get_stream());
+  thrust::tabulate(rmm::exec_policy(handle.get_stream())->on(handle.get_stream()),
+                   degrees.begin(),
+                   degrees.end(),
+                   [offsets] __device__(auto i) { return offsets[i + 1] - offsets[i]; });
+  return degrees;
 }
 
 template <typename vertex_t, typename edge_t>

cpp/include/experimental/graph_view.hpp

@@ -494,6 +494,12 @@ class graph_view_t<vertex_t,
              : static_cast<weight_t const*>(nullptr);
   }
 
+  rmm::device_uvector<edge_t> compute_in_degrees(raft::handle_t const& handle) const;
+  rmm::device_uvector<edge_t> compute_out_degrees(raft::handle_t const& handle) const;
+
+  rmm::device_uvector<weight_t> compute_in_weight_sums(raft::handle_t const& handle) const;
+  rmm::device_uvector<weight_t> compute_out_weight_sums(raft::handle_t const& handle) const;
+
  private:
   std::vector<edge_t const*> adj_matrix_partition_offsets_{};
   std::vector<vertex_t const*> adj_matrix_partition_indices_{};
@@ -638,6 +644,12 @@ class graph_view_t<vertex_t,
   // private.
   weight_t const* weights() const { return weights_; }
 
+  rmm::device_uvector<edge_t> compute_in_degrees(raft::handle_t const& handle) const;
+  rmm::device_uvector<edge_t> compute_out_degrees(raft::handle_t const& handle) const;
+
+  rmm::device_uvector<weight_t> compute_in_weight_sums(raft::handle_t const& handle) const;
+  rmm::device_uvector<weight_t> compute_out_weight_sums(raft::handle_t const& handle) const;
+
  private:
   edge_t const* offsets_{nullptr};
   vertex_t const* indices_{nullptr};

cpp/include/patterns/copy_v_transform_reduce_key_aggregated_out_nbr.cuh

@@ -18,8 +18,10 @@
 #include <experimental/detail/graph_utils.cuh>
 #include <experimental/graph.hpp>
 #include <experimental/graph_view.hpp>
+#include <matrix_partition_device.cuh>
 #include <utilities/dataframe_buffer.cuh>
 #include <utilities/error.hpp>
+#include <utilities/host_scalar_comm.cuh>
 #include <utilities/shuffle_comm.cuh>
 #include <vertex_partition_device.cuh>
 
@@ -100,10 +102,10 @@ __global__ void for_all_major_for_all_nbr_low_degree(
       }
       thrust::fill(thrust::seq,
                    major_vertices + local_offset,
-                   major_vertices + local_offset + key_idx,
+                   major_vertices + local_offset + key_idx + 1,
                    matrix_partition.get_major_from_major_offset_nocheck(major_offset));
       thrust::fill(thrust::seq,
-                   major_vertices + local_offset + key_idx,
+                   major_vertices + local_offset + key_idx + 1,
                    major_vertices + local_offset + local_degree,
                    invalid_vertex);
     }
@@ -159,8 +161,7 @@ __global__ void for_all_major_for_all_nbr_low_degree(
  * pairs provided by @p map_key_first, @p map_key_last, and @p map_value_first (aggregated over the
  * entire set of processes in multi-GPU).
  * @param reduce_op Binary operator takes two input arguments and reduce the two variables to one.
- * @param init Initial value to be added to the reduced @p key_aggregated_e_op return values for
- * each vertex.
+ * @param init Initial value to be added to the reduced @p reduce_op return values for each vertex.
  * @param vertex_value_output_first Iterator pointing to the vertex property variables for the
  * first (inclusive) vertex (assigned to tihs process in multi-GPU). `vertex_value_output_last`
  * (exclusive) is deduced as @p vertex_value_output_first + @p
@@ -191,6 +192,7 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
                 "GraphViewType should support the push model.");
   static_assert(std::is_same<typename std::iterator_traits<VertexIterator>::value_type,
                              typename GraphViewType::vertex_type>::value);
+  static_assert(is_arithmetic_or_thrust_tuple_of_arithmetic<T>::value);
 
   using vertex_t = typename GraphViewType::vertex_type;
   using edge_t   = typename GraphViewType::edge_type;
@@ -393,7 +395,7 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
       tmp_major_vertices.begin(), tmp_minor_keys.begin(), tmp_key_aggregated_edge_weights.begin()));
     thrust::transform(rmm::exec_policy(handle.get_stream())->on(handle.get_stream()),
                       triplet_first,
-                      triplet_first + major_vertices.size(),
+                      triplet_first + tmp_major_vertices.size(),
                       tmp_e_op_result_buffer_first,
                       [adj_matrix_row_value_input_first,
                        key_aggregated_e_op,
@@ -408,7 +410,7 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
                           w,
                           *(adj_matrix_row_value_input_first +
                             matrix_partition.get_major_offset_from_major_nocheck(major)),
-                          kv_map.find(key)->second);
+                          kv_map.find(key)->second.load(cuda::std::memory_order_relaxed));
                       });
     tmp_minor_keys.resize(0, handle.get_stream());
     tmp_key_aggregated_edge_weights.resize(0, handle.get_stream());
@@ -488,11 +490,12 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
   auto major_vertex_first = thrust::make_transform_iterator(
     thrust::make_counting_iterator(size_t{0}),
     [major_vertices = major_vertices.data()] __device__(auto i) {
-      return ((i == 0) || (major_vertices[i] == major_vertices[i - 1]))
+      return ((i == 0) || (major_vertices[i] != major_vertices[i - 1]))
                ? major_vertices[i]
                : invalid_vertex_id<vertex_t>::value;
     });
   thrust::copy_if(
+    rmm::exec_policy(handle.get_stream())->on(handle.get_stream()),
     major_vertex_first,
     major_vertex_first + major_vertices.size(),
     unique_major_vertices.begin(),
@@ -506,9 +509,10 @@ void copy_v_transform_reduce_key_aggregated_out_nbr(
     thrust::make_permutation_iterator(
       vertex_value_output_first,
       thrust::make_transform_iterator(
-        major_vertices.begin(),
+        unique_major_vertices.begin(),
         [vertex_partition = vertex_partition_device_t<GraphViewType>(graph_view)] __device__(
           auto v) { return vertex_partition.get_local_vertex_offset_from_vertex_nocheck(v); })),
+    thrust::equal_to<vertex_t>{},
     reduce_op);
 
   thrust::transform(rmm::exec_policy(handle.get_stream())->on(handle.get_stream()),