[Bug] Multiple fixes for CUDA 11 support (#2333)

* multiple fixes

* fix CI

* fiddle

* revert stubs

* remove stubs

* poke

* remove linking of driver library

* minor
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

CMakeLists.txt

@@ -46,8 +46,8 @@ if(USE_CUDA)
     #   is fixed by https://github.com/NVIDIA/cub/commit/9143e47e048641aa0e6ddfd645bcd54ff1059939
     #   in 11.1.
     message(STATUS "Detected CUDA of version ${CUDA_VERSION}. Use external CUB/Thrust library.")
-    list(INSERT CUDA_INCLUDE_DIRS 0 "${CMAKE_SOURCE_DIR}/third_party/thrust")
-    list(INSERT CUDA_INCLUDE_DIRS 0 "${CMAKE_SOURCE_DIR}/third_party/cub")
+    cuda_include_directories(BEFORE "${CMAKE_SOURCE_DIR}/third_party/thrust")
+    cuda_include_directories(BEFORE "${CMAKE_SOURCE_DIR}/third_party/cub")
   endif()
 endif(USE_CUDA)
 
@@ -92,6 +92,7 @@ else(MSVC)
   check_cxx_compiler_flag("-std=c++14"    SUPPORT_CXX14)
   set(CMAKE_C_FLAGS "-O2 -Wall -fPIC ${CMAKE_C_FLAGS}")
   set(CMAKE_CXX_FLAGS "-O2 -Wall -fPIC -std=c++14 ${CMAKE_CXX_FLAGS}")
+  set(CMAKE_SHARED_LINKER_FLAGS "-Wl,--warn-common ${CMAKE_SHARED_LINKER_FLAGS}")
 endif(MSVC)
 
 if(USE_OPENMP)

Jenkinsfile

@@ -30,6 +30,7 @@ def unpack_lib(name, libs) {
 def build_dgl_linux(dev) {
   init_git()
   sh "bash tests/scripts/build_dgl.sh ${dev}"
+  sh "ls -lh /usr/lib/x86_64-linux-gnu/"
   pack_lib("dgl-${dev}-linux", dgl_linux_libs)
 }
 

cmake/modules/CUDA.cmake

@@ -253,8 +253,9 @@ macro(dgl_config_cuda out_variable)
   list(APPEND CMAKE_CUDA_FLAGS "${NVCC_FLAGS_EXTRA}")
 
   list(APPEND DGL_LINKER_LIBS
-    ${CUDA_CUDA_LIBRARY} ${CUDA_CUDART_LIBRARY}
-    ${CUDA_CUBLAS_LIBRARIES} ${CUDA_cusparse_LIBRARY})
+    ${CUDA_CUDART_LIBRARY}
+    ${CUDA_CUBLAS_LIBRARIES}
+    ${CUDA_cusparse_LIBRARY})
 
   set(${out_variable} ${DGL_CUDA_SRC})
 endmacro()

cmake/util/FindCUDA.cmake

@@ -41,28 +41,18 @@ macro(find_cuda use_cuda)
         ${CUDA_TOOLKIT_ROOT_DIR}/lib/x64
         ${CUDA_TOOLKIT_ROOT_DIR}/lib/Win32)
     else(MSVC)
-      find_library(_CUDA_CUDA_LIBRARY cuda
-        PATHS ${CUDA_TOOLKIT_ROOT_DIR}
-        PATH_SUFFIXES lib lib64 targets/x86_64-linux/lib targets/x86_64-linux/lib/stubs lib64/stubs
-        NO_DEFAULT_PATH)
-      if(_CUDA_CUDA_LIBRARY)
-        set(CUDA_CUDA_LIBRARY ${_CUDA_CUDA_LIBRARY})
-      endif()
-      find_library(CUDA_NVRTC_LIBRARY nvrtc
-        PATHS ${CUDA_TOOLKIT_ROOT_DIR}
-        PATH_SUFFIXES lib lib64 targets/x86_64-linux/lib targets/x86_64-linux/lib/stubs lib64/stubs lib/x86_64-linux-gnu
-        NO_DEFAULT_PATH)
-      #find_library(CUDA_CUDNN_LIBRARY cudnn
-      #  ${CUDA_TOOLKIT_ROOT_DIR}/lib64
-      #  ${CUDA_TOOLKIT_ROOT_DIR}/lib)
+      #find_library(CUDA_CUDA_LIBRARY cuda
+      #  PATHS ${CUDA_TOOLKIT_ROOT_DIR}
+      #  PATH_SUFFIXES lib lib64 targets/x86_64-linux/lib
+      #  NO_DEFAULT_PATH)
       find_library(CUDA_CUBLAS_LIBRARY cublas
         ${CUDA_TOOLKIT_ROOT_DIR}/lib64
         ${CUDA_TOOLKIT_ROOT_DIR}/lib)
     endif(MSVC)
     message(STATUS "Found CUDA_TOOLKIT_ROOT_DIR=" ${CUDA_TOOLKIT_ROOT_DIR})
-    message(STATUS "Found CUDA_CUDA_LIBRARY=" ${CUDA_CUDA_LIBRARY})
+    #message(STATUS "Found CUDA_CUDA_LIBRARY=" ${CUDA_CUDA_LIBRARY})
     message(STATUS "Found CUDA_CUDART_LIBRARY=" ${CUDA_CUDART_LIBRARY})
-    message(STATUS "Found CUDA_NVRTC_LIBRARY=" ${CUDA_NVRTC_LIBRARY})
+    #message(STATUS "Found CUDA_NVRTC_LIBRARY=" ${CUDA_NVRTC_LIBRARY})
     #message(STATUS "Found CUDA_CUDNN_LIBRARY=" ${CUDA_CUDNN_LIBRARY})
     message(STATUS "Found CUDA_CUBLAS_LIBRARY=" ${CUDA_CUBLAS_LIBRARY})
   endif(CUDA_FOUND)

src/array/cpu/array_op_impl.cc

@@ -222,26 +222,6 @@ IdArray Relabel_(const std::vector<IdArray>& arrays) {
 template IdArray Relabel_<kDLCPU, int32_t>(const std::vector<IdArray>& arrays);
 template IdArray Relabel_<kDLCPU, int64_t>(const std::vector<IdArray>& arrays);
 
-
-///////////////////////////// NonZero /////////////////////////////
-
-template <DLDeviceType XPU, typename IdType>
-IdArray NonZero(BoolArray bool_arr) {
-  const IdType* bool_data = static_cast<IdType*>(bool_arr->data);
-  CHECK(bool_arr->ndim == 1) << "NonZero only supports 1D array";
-  std::vector<IdType> nonzero_indices;
-  for (int64_t i = 0; i < bool_arr->shape[0]; i++) {
-    if ((bool_data[i]) != 0) {
-      nonzero_indices.push_back(i);
-    }
-  }
-  return VecToIdArray(nonzero_indices, sizeof(IdType) * 8);
-}
-
-// TODO(Allen): Implement GPU version
-template IdArray NonZero<kDLCPU, int32_t>(BoolArray bool_arr);
-template IdArray NonZero<kDLCPU, int64_t>(BoolArray bool_arr);
-
 }  // namespace impl
 }  // namespace aten
 }  // namespace dgl