CMake管理CUDA并使用cuSOLVER等

一、出现问题

我在使用官方案例的时候，使用VS2022+CMake管理编译的时候出现如下的错误：

在这里插入图片描述

官方CMakeLists.txt：


cmake_minimum_required(VERSION 3.9)set(ROUTINE bicgstab)project("${ROUTINE}_example"DESCRIPTION  "GPU-Accelerated Sparse Linear Algebra"HOMEPAGE_URL "https://docs.nvidia.com/cuda/cusparse/index.html"LANGUAGES    C CXX CUDA)find_package(CUDA REQUIRED)add_executable(${ROUTINE}_example)target_sources(${ROUTINE}_examplePUBLIC ${PROJECT_SOURCE_DIR}/${ROUTINE}_example.c
)target_include_directories(${ROUTINE}_examplePUBLIC ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES}
)link_directories(${CUDA_LIB_PATH})
message(STATUS ${CUDA_LIB_PATH})
target_link_libraries(${ROUTINE}_examplePUBLIC cudart cusparse cublas
)

二、解决问题

开始我觉得是不是find_package(CUDA REQUIRED)这里有一些bug，不能将CUDA一些库的目录正确设置，所以我又添加了:

link_directories("C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.3\lib\x64")

然后无效!

我在互联网上查找原因，发现可能是由于目录有空格导致的，需要在空格前使用转移字符：\

link_directories("C:\Program/ Files\NVIDIA/ GPU/ Computing/ Toolkit\CUDA\v12.3\lib\x64")

然后仍然无效!！

关键方案

然后查找官方文档：（有这么一段）链接
CUDA Runtime Library
The CUDA Runtime library (cudart) are what most applications will typically need to link against to make any calls such as cudaMalloc, and cudaFree.
Targets Created:
CUDA::cudart
CUDA::cudart_static

于是最后我的CMakeLists.txt改写成如下：


cmake_minimum_required(VERSION 3.9)set(ROUTINE bicgstab)project("${ROUTINE}_example"DESCRIPTION  "GPU-Accelerated Sparse Linear Algebra"HOMEPAGE_URL "https://docs.nvidia.com/cuda/cusparse/index.html"LANGUAGES    C CXX CUDA)find_package(CUDAToolkit REQUIRED)add_executable(${ROUTINE}_example)target_sources(${ROUTINE}_examplePUBLIC ${PROJECT_SOURCE_DIR}/${ROUTINE}_example.c
)target_include_directories(${ROUTINE}_examplePUBLIC ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES}
)link_directories(${CUDA_LIB_PATH})
message(STATUS ${CUDA_LIB_PATH})
target_link_libraries(${ROUTINE}_examplePUBLIC CUDA::cudart CUDA::cusparse CUDA::cublas
)