[CUDA] Update CCCL to 2.3.0 (#7171)

CMakeLists.txt

@@ -139,8 +139,6 @@ endif()
 if(USE_CUDA)
   message(STATUS "Build with CUDA support")
   project(dgl C CXX)
-  # see https://github.com/NVIDIA/thrust/issues/1401
-  add_definitions(-DTHRUST_CUB_WRAPPED_NAMESPACE=dgl)
   include(cmake/modules/CUDA.cmake)
   message(STATUS "Use external CCCL library for a consistent API and performance.")
   cuda_include_directories(BEFORE "${CMAKE_SOURCE_DIR}/third_party/cccl/thrust")

graphbolt/CMakeLists.txt

@@ -70,8 +70,6 @@ target_link_libraries(${LIB_GRAPHBOLT_NAME} "${TORCH_LIBRARIES}")
 if(USE_CUDA)
   set_target_properties(${LIB_GRAPHBOLT_NAME} PROPERTIES CUDA_STANDARD 17)
   message(STATUS "Use external CCCL library for a consistent API and performance for graphbolt.")
-  target_compile_definitions(${LIB_GRAPHBOLT_NAME} PRIVATE CUB_WRAPPED_NAMESPACE=graphbolt)
-  target_compile_definitions(${LIB_GRAPHBOLT_NAME} PRIVATE THRUST_NS_QUALIFIER=thrust)
   target_include_directories(${LIB_GRAPHBOLT_NAME} PRIVATE
                              "../third_party/cccl/thrust"
                              "../third_party/cccl/cub"

src/array/cuda/array_cumsum.cu

@@ -5,8 +5,9 @@
  */
 #include <dgl/array.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/array_nonzero.cu

@@ -6,8 +6,9 @@
 
 #include <dgl/array.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/array_sort.cu

@@ -5,8 +5,9 @@
  */
 #include <dgl/array.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/csr2coo.cu

@@ -8,8 +8,9 @@
 #include <thrust/iterator/counting_iterator.h>
 #include <thrust/iterator/transform_iterator.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/csr_sort.cu

@@ -5,8 +5,9 @@
  */
 #include <dgl/array.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/cuda_filter.cu

@@ -6,10 +6,11 @@
 
 #include <dgl/runtime/device_api.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
 #include "../../runtime/cuda/cuda_hashtable.cuh"
 #include "../filter.h"
-#include "./dgl_cub.cuh"
 
 using namespace dgl::runtime::cuda;
 

src/array/cuda/dgl_cub.cuh

-/**
- *  Copyright (c) 2021 by Contributors
- * @file cuda_common.h
- * @brief Wrapper to place cub in dgl namespace.
- */
-
-#ifndef DGL_ARRAY_CUDA_DGL_CUB_CUH_
-#define DGL_ARRAY_CUDA_DGL_CUB_CUH_
-
-// This should be defined in CMakeLists.txt
-#ifndef THRUST_CUB_WRAPPED_NAMESPACE
-static_assert(false, "THRUST_CUB_WRAPPED_NAMESPACE must be defined for DGL.");
-#endif
-
-#include "cub/cub.cuh"
-
-#endif  // DGL_ARRAY_CUDA_DGL_CUB_CUH_

src/array/cuda/labor_sampling.cu

@@ -34,6 +34,7 @@
 #include <thrust/zip_function.h>
 
 #include <algorithm>
+#include <cub/cub.cuh>  // NOLINT
 #include <limits>
 #include <numeric>
 #include <type_traits>
@@ -44,7 +45,6 @@
 #include "../../graph/transform/cuda/cuda_map_edges.cuh"
 #include "../../random/continuous_seed.h"
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./functor.cuh"
 #include "./spmm.cuh"
 

src/array/cuda/negative_sampling.cu

@@ -9,8 +9,9 @@
 #include <dgl/array_iterator.h>
 #include <dgl/random.h>
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 using namespace dgl::runtime;

src/array/cuda/rowwise_sampling.cu

@@ -9,11 +9,11 @@
 #include <dgl/runtime/device_api.h>
 #include <dgl/runtime/tensordispatch.h>
 
+#include <cub/cub.cuh>
 #include <numeric>
 
 #include "../../array/cuda/atomic.cuh"
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 using namespace dgl::cuda;

src/array/cuda/rowwise_sampling_prob.cu

@@ -10,11 +10,11 @@
 #include <dgl/random.h>
 #include <dgl/runtime/device_api.h>
 
+#include <cub/cub.cuh>
 #include <numeric>
 
 #include "../../array/cuda/atomic.cuh"
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 // require CUB 1.17 to use DeviceSegmentedSort

src/array/cuda/spmat_op_impl_csr.cu

@@ -7,13 +7,13 @@
 #include <thrust/execution_policy.h>
 #include <thrust/for_each.h>
 
+#include <cub/cub.cuh>
 #include <numeric>
 #include <unordered_set>
 #include <vector>
 
 #include "../../runtime/cuda/cuda_common.h"
 #include "./atomic.cuh"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/utils.cu

@@ -4,8 +4,9 @@
  * @brief Utilities for CUDA kernels.
  */
 
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
-#include "./dgl_cub.cuh"
 #include "./utils.h"
 
 namespace dgl {

src/array/cuda/utils.h

@@ -11,10 +11,10 @@
 #include <dgl/runtime/ndarray.h>
 #include <dmlc/logging.h>
 
+#include <cub/cub.cuh>
 #include <type_traits>
 
 #include "../../runtime/cuda/cuda_common.h"
-#include "dgl_cub.cuh"
 
 namespace dgl {
 namespace cuda {

src/graph/sampling/randomwalks/frequency_hashmap.cu

@@ -5,11 +5,11 @@
  */
 
 #include <algorithm>
+#include <cub/cub.cuh>  // NOLINT
 #include <tuple>
 #include <utility>
 
 #include "../../../array/cuda/atomic.cuh"
-#include "../../../array/cuda/dgl_cub.cuh"
 #include "../../../runtime/cuda/cuda_common.h"
 #include "frequency_hashmap.cuh"
 

src/graph/sampling/randomwalks/randomwalk_gpu.cu

@@ -10,11 +10,11 @@
 #include <dgl/random.h>
 #include <dgl/runtime/device_api.h>
 
+#include <cub/cub.cuh>
 #include <tuple>
 #include <utility>
 #include <vector>
 
-#include "../../../array/cuda/dgl_cub.cuh"
 #include "../../../runtime/cuda/cuda_common.h"
 #include "frequency_hashmap.cuh"
 

src/graph/transform/cuda/knn.cu

@@ -10,12 +10,12 @@
 #include <dgl/runtime/device_api.h>
 
 #include <algorithm>
+#include <cub/cub.cuh>  // NOLINT
 #include <limits>
 #include <string>
 #include <type_traits>
 #include <vector>
 
-#include "../../../array/cuda/dgl_cub.cuh"
 #include "../../../array/cuda/utils.h"
 #include "../../../runtime/cuda/cuda_common.h"
 #include "../knn.h"

src/partition/cuda/partition_op.cu

@@ -6,7 +6,8 @@
 
 #include <dgl/runtime/device_api.h>
 
-#include "../../array/cuda/dgl_cub.cuh"
+#include <cub/cub.cuh>
+
 #include "../../runtime/cuda/cuda_common.h"
 #include "../../runtime/workspace.h"
 #include "../partition_op.h"