Merge remote-tracking branch 'upstream/branch-0.8' into branch-0.8

oorliu · May 24, 2019 · c2d5fb8 · c2d5fb8
2 parents d96db05 + d488887
commit c2d5fb8
Show file tree

Hide file tree

Showing 107 changed files with 14,957 additions and 13,794 deletions.
diff --git a/.gitmodules b/.gitmodules
@@ -1,7 +1,3 @@
-[submodule "cpp/nvgraph/cpp/thirdparty/cnmem"]
-	path = cpp/nvgraph/cpp/thirdparty/cnmem
-	url = https://github.com/NVIDIA/cnmem.git
-	branch = master
 [submodule "cpp/nvgraph/cpp/thirdparty/cub"]
 	path = cpp/nvgraph/cpp/thirdparty/cub
 	url = https://github.com/NVlabs/cub.git

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,3 +1,19 @@
+# cuGraph 0.8.0 (Date TBD)
+
+## New Features
+- PR #287 SNMG power iteration step1 
+- PR #297 SNMG degree calculation 
+
+## Improvements
+- PR #291 nvGraph is updated to use RMM instead of directly invoking cnmem functions.
+- PR #286 Reorganized cugraph source directory
+
+
+## Bug Fixes
+- PR #283 Automerge fix
+- PR #291 Fixed a RMM memory allocation failure due to duplicate copies of cnmem.o
+- PR #291 Fixed a cub CsrMV call error when RMM pool allocator is used.
+
 # cuGraph 0.7.0 (Date TBD)
 
 ## New Features
@@ -54,6 +70,7 @@
 - PR #262 Removed networkx conda dependency for both build and runtime
 - PR #271 Removed nvgraph conda dependency
 - PR #276 Removed libgdf_cffi import from bindings
+- PR #288 Add boost as a conda dependency
 
 # cuGraph 0.6.0 (22 Mar 2019)
 

diff --git a/README.md b/README.md
@@ -68,7 +68,7 @@ These limitations are being addressed and will be fixed future versions.
 
 ## Getting cuGraph
 ### Intro
-There are 4 ways to get cuGraph :
+There are 3 ways to get cuGraph :
 1. [Quick start with Docker Demo Repo](#quick)
 1. [Conda Installation](#conda)
 1. [Build from Source](#source)
@@ -133,5 +133,4 @@ The RAPIDS suite of open source software libraries aim to enable execution of en
 
 ### Apache Arrow on GPU
 
-The GPU version of [Apache Arrow](https://arrow.apache.org/) is a common API that enables efficient interchange of tabular data between processes running on the GPU. End-to-end computation on the GPU avoids unnecessary copying and converting of data off the GPU, reducing compute time and cost for high-performance analytics common in artificial intelligence workloads. As the name implies, cuDF uses the Apache Arrow columnar data format on the GPU. Currently, a subset of the features in Apache Arrow are supported.
-
+The GPU version of [Apache Arrow](https://arrow.apache.org/) is a common API that enables efficient interchange of tabular data between processes running on the GPU. End-to-end computation on the GPU avoids unnecessary copying and converting of data off the GPU, reducing compute time and cost for high-performance analytics common in artificial intelligence workloads. As the name implies, cuDF uses the Apache Arrow columnar data format on the GPU. Currently, a subset of the features in Apache Arrow are supported.
diff --git a/conda/recipes/cugraph/meta.yaml b/conda/recipes/cugraph/meta.yaml
@@ -26,11 +26,11 @@ requirements:
   build:
     - python x.x
     - libcugraph={{ version }}
-    - cudf=0.7*
+    - cudf=0.8*
   run:
     - python x.x
     - libcugraph={{ version }}
-    - cudf=0.7*
+    - cudf=0.8*
 
 #test:
 #  commands:

diff --git a/conda/recipes/libcugraph/meta.yaml b/conda/recipes/libcugraph/meta.yaml
@@ -25,15 +25,15 @@ build:
 requirements:
   build:
     - cmake>=3.12.4
-    - libcudf=0.7*
+    - libcudf=0.8*
     - cython
     - cudatoolkit {{ cuda_version }}.*
-    - boost-cpp
+    - boost
   run:
-    - libcudf=0.7*
+    - libcudf=0.8*
     - cython
     - {{ pin_compatible('cudatoolkit', max_pin='x.x') }}
-    - boost-cpp
+    - boost
 
 #test:
 #  commands:

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
@@ -16,7 +16,7 @@
 #=============================================================================
 cmake_minimum_required(VERSION 3.12 FATAL_ERROR)
 
-project(cuGraph VERSION 0.6.0 LANGUAGES C CXX CUDA)
+project(cuGraph VERSION 0.8.0 LANGUAGES C CXX CUDA)
 
 ###################################################################################################
 # - cmake modules ---------------------------------------------------------------------------------
@@ -136,7 +136,7 @@ include(ConfigureNvgraph)
 
 ###################################################################################################
 # - Find and add different modules and supporting repos -------------------------------------------
-find_package(Boost 1.45.0 COMPONENTS system)
+find_package(Boost REQUIRED)
 
 find_package(OpenMP)
 if (OPENMP_FOUND)
@@ -229,14 +229,21 @@ link_directories("${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES}" # CMAKE_CUDA_IMPLICIT
 ###################################################################################################
 # - library targets -------------------------------------------------------------------------------
 add_library(cugraph SHARED
-    src/grmat.cu
-    src/cugraph.cu
-    src/pagerank.cu
-    src/bfs.cu
-    src/jaccard.cu
-    src/overlap.cu
-    src/nvgraph_gdf.cu
-    src/two_hop_neighbors.cu
+    src/utilities/grmat.cu
+    src/utilities/degree.cu
+    src/structure/cugraph.cu
+    src/link_analysis/pagerank.cu
+    src/traversal/bfs.cu
+    src/link_prediction/jaccard.cu
+    src/link_prediction/overlap.cu
+    src/converters/nvgraph.cu
+    src/converters/renumber.cu
+    src/community/nvgraph_gdf.cu
+    src/traversal/nvgraph_sssp.cu
+    src/traversal/two_hop_neighbors.cu
+    src/snmg/blas/spmv.cu
+    src/snmg/degree/degree.cu
+    src/snmg/utils.cu
     ${CMAKE_CURRENT_BINARY_DIR}/gunrock/gunrock/util/test_utils.cu
     ${CMAKE_CURRENT_BINARY_DIR}/gunrock/gunrock/util/error_utils.cu
     ${CMAKE_CURRENT_BINARY_DIR}/gunrock/gunrock/util/misc_utils.cu

diff --git a/cpp/include/algorithms.h b/cpp/include/algorithms.h
@@ -170,3 +170,19 @@ gdf_error gdf_louvain(gdf_graph *graph,
                       void *final_modularity,
                       void *num_level,
                       gdf_column *louvain_parts);
+
+/**
+ * Computes the in-degree, out-degree, or the sum of both (determined by x) for the given graph. This is
+ * a multi-gpu operation operating on a partitioned graph.
+ * @param x 0 for in+out, 1 for in, 2 for out
+ * @param part_offsets Contains the start/end of each partitions vertex id range
+ * @param off The local partition offsets
+ * @param ind The local partition indices
+ * @param x_cols The results (located on each GPU)
+ * @return Error code
+ */
+gdf_error gdf_snmg_degree(int x,
+                          size_t* part_offsets,
+                          gdf_column* off,
+                          gdf_column* ind,
+                          gdf_column** x_cols);
diff --git a/cpp/include/rmm_utils.h b/cpp/include/rmm_utils.h
@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 #pragma once
-///#define DEBUG_NO_RMM
 
 #include <sstream>
 #include <stdexcept>
@@ -27,68 +26,17 @@
       throw std::runtime_error(ss.str());               \
     }
 
-#ifdef DEBUG_NO_RMM
-
-#include <thrust/device_malloc_allocator.h>
-#include <thrust/system_error.h>
-#include <thrust/system/cuda/error.h>
-#include <thrust/execution_policy.h>
-
-template<typename T>
-//using rmm_allocator = thrust::device_malloc_allocator<T>;
-class rmm_allocator : public thrust::device_malloc_allocator<T>
-{
-  public:
-    using value_type = T;
-
-    rmm_allocator(cudaStream_t stream = 0) : stream(stream) {}
-    ~rmm_allocator() {}
-
-private:
-    cudaStream_t stream;
-};
-
-using rmm_temp_allocator = rmm_allocator<char>; // Use this alias for thrust::cuda::par(allocator).on(stream)
-
-#define ALLOC_TRY(ptr, sz, stream){            \
-    if (stream == nullptr) ;                      \
-    cudaMalloc((ptr), (sz));                   \
-}
-
-#define ALLOC_MANAGED_TRY(ptr, sz, stream){    \
-    if (stream == nullptr) ;                      \
-    cudaMallocManaged((ptr), (sz));            \
-}
-
-  //#define REALLOC_TRY(ptr, new_sz, stream)
-
-#define ALLOC_FREE_TRY(ptr, stream){                \
-    if (stream == nullptr) ;                      \
-    cudaFree( (ptr) );                              \
-}
-#else
-
 #include <rmm/rmm.h>
 #include <rmm/thrust_rmm_allocator.h>
 
-using rmm_temp_allocator = rmm_allocator<char>;
-
-#define ALLOC_TRY( ptr, sz, stream ){                   \
-      RMM_TRY_THROW( RMM_ALLOC((ptr), (sz), (stream)) ) \
-    }
-
-//TODO: change this when RMM alloc managed will be available !!!!!
-#define ALLOC_MANAGED_TRY(ptr, sz, stream){         \
+#define ALLOC_TRY( ptr, sz, stream ){               \
   RMM_TRY_THROW( RMM_ALLOC((ptr), (sz), (stream)) ) \
 }
 
 #define REALLOC_TRY(ptr, new_sz, stream){             \
   RMM_TRY_THROW( RMM_REALLOC((ptr), (sz), (stream)) ) \
 }
 
-#define ALLOC_FREE_TRY(ptr, stream){                \
+#define ALLOC_FREE_TRY(ptr, stream){            \
   RMM_TRY_THROW( RMM_FREE( (ptr), (stream) ) )  \
 }
-
-#endif
-
diff --git a/cpp/nvgraph/cpp/CMakeLists.txt b/cpp/nvgraph/cpp/CMakeLists.txt
@@ -20,12 +20,12 @@ project(NV_GRAPH VERSION 0.4.0 LANGUAGES C CXX CUDA)
 ###################################################################################################
 # - compiler options ------------------------------------------------------------------------------
 
-set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD 14)
 set(CMAKE_C_COMPILER $ENV{CC})
 set(CMAKE_CXX_COMPILER $ENV{CXX})
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
-set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_STANDARD 14)
 set(CMAKE_CUDA_STANDARD_REQUIRED ON)
 
 if(CMAKE_COMPILER_IS_GNUCXX)
@@ -47,7 +47,7 @@ option(BUILD_TESTS "Configure CMake to build tests"
 if(CMAKE_COMPILER_IS_GNUCXX)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror")
 
-    option(CMAKE_CXX11_ABI "Enable the GLIBCXX11 ABI" OFF)
+    option(CMAKE_CXX11_ABI "Enable the GLIBCXX11 ABI" ON)
     if(CMAKE_CXX11_ABI)
         message(STATUS "nvGraph: Enabling the GLIBCXX11 ABI")
     else()
@@ -67,6 +67,25 @@ include(FeatureSummary)
 include(CheckIncludeFiles)
 include(CheckLibraryExists)
 
+###################################################################################################
+# - add rmm --------------------------------------------------------------------------------------
+find_path(RMM_INCLUDE "rmm"
+          HINTS "$ENV{RMM_ROOT}/include"
+                "$ENV{CONDA_PREFIX}/include/rmm"
+                "$ENV{CONDA_PREFIX}/include")
+
+find_library(RMM_LIBRARY "rmm"
+             HINTS "$ENV{RMM_ROOT}/lib"
+                   "$ENV{CONDA_PREFIX}/lib")
+
+message(STATUS "RMM: RMM_LIBRARY set to ${RMM_LIBRARY}")
+message(STATUS "RMM: RMM_INCLUDE set to ${RMM_INCLUDE}")
+
+add_library(rmm SHARED IMPORTED ${RMM_LIBRARY})
+if (RMM_INCLUDE AND RMM_LIBRARY)
+    set_target_properties(rmm PROPERTIES IMPORTED_LOCATION ${RMM_LIBRARY})
+endif (RMM_INCLUDE AND RMM_LIBRARY)
+
 ###################################################################################################
 # - add gtest -------------------------------------------------------------------------------------
 
@@ -90,9 +109,9 @@ include_directories(
                     "${CMAKE_BINARY_DIR}/include"
                     "${CMAKE_SOURCE_DIR}/include"
                     "${CMAKE_SOURCE_DIR}/thirdparty/cub"
-                    "${CMAKE_SOURCE_DIR}/thirdparty/cnmem/include"
                     "${CMAKE_SOURCE_DIR}/../external"
                     "${CMAKE_SOURCE_DIR}/../external/cusp"
+                    "${RMM_INCLUDE}"
                     "${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES}"
                    )
 
@@ -101,13 +120,13 @@ include_directories(
 
 link_directories("${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES}" # CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES is an undocumented/unsupported variable containing the link directories for nvcc
                  "${CMAKE_BINARY_DIR}/lib"
-                 "${GTEST_LIBRARY_DIR}")
+                 "${GTEST_LIBRARY_DIR}"
+                 "${RMM_LIBRARY}")
 
 ###################################################################################################
 # - library targets -------------------------------------------------------------------------------
 if(NVGRAPH_LIGHT MATCHES True)
     add_library(nvgraph_rapids SHARED
-                thirdparty/cnmem/src/cnmem.cpp
                 src/arnoldi.cu
                 src/bfs.cu
                 src/bfs2d.cu
@@ -141,7 +160,6 @@ if(NVGRAPH_LIGHT MATCHES True)
                )
 else(NVGRAPH_LIGHT MATCHES True)
         add_library(nvgraph_rapids SHARED
-                thirdparty/cnmem/src/cnmem.cpp
                 src/arnoldi.cu
                 src/bfs.cu
                 src/bfs2d.cu
@@ -203,7 +221,7 @@ endif(NVGRAPH_LIGHT MATCHES True)
 ###################################################################################################
 # - link libraries --------------------------------------------------------------------------------
 
-target_link_libraries(nvgraph_rapids cublas cusparse curand cusolver cudart )
+target_link_libraries(nvgraph_rapids cublas cusparse curand cusolver rmm cudart cuda)
 
 ###################################################################################################
 # - install targets -------------------------------------------------------------------------------