Add CUDA_NVCC_THREADS_NUMBER (#6017)

Co-authored-by: Noneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>

CMakeLists.txt

@@ -149,6 +149,10 @@ if (BUILD_CUDA)
   message(STATUS "CUDA_VERSION: ${CUDA_VERSION}")
   message(STATUS "CUDA_TOOLKIT_ROOT_DIR: ${CUDA_TOOLKIT_ROOT_DIR}")
   set(CUDA_SEPARABLE_COMPILATION OFF)
+  if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "11.2")
+    set(CUDA_NVCC_THREADS_NUMBER "1" CACHE STRING "")
+    list(APPEND CUDA_NVCC_FLAGS -t ${CUDA_NVCC_THREADS_NUMBER})
+  endif()
   list(APPEND CUDA_NVCC_FLAGS -w -Werror cross-execution-space-call -Wno-deprecated-gpu-targets)
   #  half is not fully supported when __CUDA_ARCH__ < 530
   #  list(APPEND DEFAULT_CUDA_NVCC_GENCODES "arch=compute_30,code=sm_30")