Merge pull request rapidsai#840 from afender/opg_degree

[REVIEW] OPG degree

CHANGELOG.md

@@ -2,6 +2,7 @@
 
 ## New Features
 - PR #822 Added new functions in python graph class, similar to networkx
+- PR #840 OPG degree
 
 ## Improvements
 - PR #764 Updated sssp and bfs with GraphCSR, removed gdf_column, added nullptr weights test for sssp

cpp/CMakeLists.txt

@@ -290,6 +290,7 @@ link_directories(
     "${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES}")
 
 add_library(cugraph SHARED
+    src/comms/mpi/comms_mpi.cpp
     src/ktruss/ktruss.cu
     src/db/db_object.cu
     src/db/db_parser_integration_test.cu
@@ -352,6 +353,10 @@ add_library(cugraph SHARED
 #
 add_dependencies(cugraph cugunrock)
 
+if (BUILD_MPI)
+    add_compile_definitions(ENABLE_OPG=1)
+endif (BUILD_MPI)
+
 ###################################################################################################
 # - include paths ---------------------------------------------------------------------------------
 target_include_directories(cugraph

cpp/include/comms_mpi.hpp

@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2020, NVIDIA CORPORATION.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#pragma once
+#if ENABLE_OPG
+#include <mpi.h>
+#include <nccl.h>
+#endif
+#include <cstddef>
+namespace cugraph {
+namespace experimental {
+
+enum class ReduceOp { SUM, MAX, MIN };
+
+// basic info about the snmg env setup
+class Comm {
+ private:
+  int _p{0};
+  int _rank{0};
+  bool _finalize_mpi{false};
+  bool _finalize_nccl{false};
+
+  int _device_id{0};
+  int _device_count{0};
+
+  int _sm_count_per_device{0};
+  int _max_grid_dim_1D{0};
+  int _max_block_dim_1D{0};
+  int _l2_cache_size{0};
+  int _shared_memory_size_per_sm{0};
+
+#if ENABLE_OPG
+  MPI_Comm _mpi_comm{};
+  ncclComm_t _nccl_comm{};
+#endif
+
+ public:
+  Comm(){};
+  Comm(int p);
+#if ENABLE_OPG
+  Comm(ncclComm_t comm, int size, int rank);
+#endif
+  ~Comm();
+  int get_rank() const { return _rank; }
+  int get_p() const { return _p; }
+  int get_dev() const { return _device_id; }
+  int get_dev_count() const { return _device_count; }
+  int get_sm_count() const { return _sm_count_per_device; }
+  bool is_master() const { return (_rank == 0) ? true : false; }
+
+  void barrier();
+
+  template <typename value_t>
+  void allgather(size_t size, value_t *sendbuff, value_t *recvbuff) const;
+
+  template <typename value_t>
+  void allreduce(size_t size, value_t *sendbuff, value_t *recvbuff, ReduceOp reduce_op) const;
+};
+
+}  // namespace experimental
+}  // namespace cugraph

cpp/include/graph.hpp

@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 #pragma once
-
+#include <comms_mpi.hpp>
 namespace cugraph {
 namespace experimental {
 
@@ -47,6 +47,7 @@ enum class DegreeDirection {
 template <typename VT, typename ET, typename WT>
 class GraphBase {
  public:
+  Comm comm;
   WT *edge_data;  ///< edge weight
 
   GraphProperties prop;
@@ -57,12 +58,16 @@ class GraphBase {
   /**
    * @brief      Fill the identifiers array with the vertex identifiers.
    *
-   * @param[out]    identifier      Pointer to device memory to store the vertex identifiers
+   * @param[out]    identifier      Pointer to device memory to store the vertex
+   * identifiers
    */
   void get_vertex_identifiers(VT *identifiers) const;
 
+  void set_communicator(Comm &comm_) { comm = comm_; }
+
   GraphBase(WT *edge_data_, VT number_of_vertices_, ET number_of_edges_)
     : edge_data(edge_data_),
+      comm(),
       prop(),
       number_of_vertices(number_of_vertices_),
       number_of_edges(number_of_edges_)
@@ -102,8 +107,8 @@ class GraphCOO : public GraphBase<VT, ET, WT> {
   /**
    * @brief      Wrap existing arrays representing an edge list in a Graph.
    *
-   *             GraphCOO does not own the memory used to represent this graph. This
-   *             function does not allocate memory.
+   *             GraphCOO does not own the memory used to represent this graph.
+   * This function does not allocate memory.
    *
    * @param  source_indices        This array of size E (number of edges) contains the index of the
    * source for each edge. Indices must be in the range [0, V-1].
@@ -138,9 +143,11 @@ class GraphCompressedSparseBase : public GraphBase<VT, ET, WT> {
   VT *indices{nullptr};  ///< CSR indices
 
   /**
-   * @brief      Fill the identifiers in the array with the source vertex identifiers
+   * @brief      Fill the identifiers in the array with the source vertex
+   * identifiers
    *
-   * @param[out]    src_indices      Pointer to device memory to store the source vertex identifiers
+   * @param[out]    src_indices      Pointer to device memory to store the
+   * source vertex identifiers
    */
   void get_source_indices(VT *src_indices) const;
 
@@ -160,8 +167,8 @@ class GraphCompressedSparseBase : public GraphBase<VT, ET, WT> {
 
   /**
    * @brief      Wrap existing arrays representing adjacency lists in a Graph.
-   *             GraphCSR does not own the memory used to represent this graph. This
-   *             function does not allocate memory.
+   *             GraphCSR does not own the memory used to represent this graph.
+   * This function does not allocate memory.
    *
    * @param  offsets               This array of size V+1 (V is number of vertices) contains the
    * offset of adjacency lists of every vertex. Offsets must be in the range [0, E] (number of
@@ -199,8 +206,8 @@ class GraphCSR : public GraphCompressedSparseBase<VT, ET, WT> {
 
   /**
    * @brief      Wrap existing arrays representing adjacency lists in a Graph.
-   *             GraphCSR does not own the memory used to represent this graph. This
-   *             function does not allocate memory.
+   *             GraphCSR does not own the memory used to represent this graph.
+   * This function does not allocate memory.
    *
    * @param  offsets               This array of size V+1 (V is number of vertices) contains the
    * offset of adjacency lists of every vertex. Offsets must be in the range [0, E] (number of
@@ -235,9 +242,9 @@ class GraphCSC : public GraphCompressedSparseBase<VT, ET, WT> {
   GraphCSC() : GraphCompressedSparseBase<VT, ET, WT>(nullptr, nullptr, nullptr, 0, 0) {}
 
   /**
-   * @brief      Wrap existing arrays representing transposed adjacency lists in a Graph.
-   *             GraphCSC does not own the memory used to represent this graph. This
-   *             function does not allocate memory.
+   * @brief      Wrap existing arrays representing transposed adjacency lists in
+   * a Graph. GraphCSC does not own the memory used to represent this graph.
+   * This function does not allocate memory.
    *
    * @param  offsets               This array of size V+1 (V is number of vertices) contains the
    * offset of adjacency lists of every vertex. Offsets must be in the range [0, E] (number of