求助：Debian系统手动安装CUDA后的链接与编译问题

我在Debian系统通过.run文件安装了CUDA 11.7和NVIDIA驱动515，目前遇到两个核心问题：

• autoconf脚本检测cudart和curand库始终失败
• 跳过检测后编译C++代码，出现cuda.h缺失的错误

当前环境信息 #

nvcc版本输出 #

执行nvcc --version得到：

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Jun__8_16:49:14_PDT_2022
Cuda compilation tools, release 11.7, V11.7.99
Build cuda_11.7.r11.7/compiler.31442593_0

已配置的环境变量 #

~/.bashrc中已添加：

export PATH=${PATH}:/usr/local/cuda-11.7/bin
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda-11.7/lib64

ldconfig相关输出 #

将/usr/local/cuda-11.7/lib64添加到ld.so.conf后，执行sudo ldconfig -v | grep cuda输出：

ldconfig: Can't stat /usr/local/cuda-11/targets/x86_64-linux/lib: No such file or directory
ldconfig: Path `/usr/local/cuda-11.7/targets/x86_64-linux/lib' given more than once
ldconfig: Can't stat /usr/local/cuda-11.4/targets/x86_64-linux/lib: No such file or directory
ldconfig: Can't stat /usr/local/lib/x86_64-linux-gnu: No such file or directory
ldconfig: Path `/usr/lib/x86_64-linux-gnu' given more than once
ldconfig: Path `/usr/lib32' given more than once
ldconfig: Path `/usr/local/cuda-11.7/lib64' given more than once
ldconfig: Path `/lib/x86_64-linux-gnu' given more than once
ldconfig: Path `/usr/lib/x86_64-linux-gnu' given more than once
ldconfig: Path `/usr/lib' given more than once
ldconfig: /lib/x86_64-linux-gnu/ld-2.31.so is the dynamic linker, ignoring

/usr/local/cuda/targets/x86_64-linux/lib:
        libcudart.so.11.0 -> libcudart.so.11.7.99
        libicudata.so.67 -> libicudata.so.67.1
        libcuda.so.1 -> libcuda.so.515.65.01
ldconfig: /lib32/ld-2.31.so is the dynamic linker, ignoring

        libcuda.so.1 -> libcuda.so.515.65.01

执行sudo ldconfig -p | grep cuda输出：

libnvrtc.so.11.2 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvrtc.so.11.2
    libnvrtc.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvrtc.so
    libnvrtc-builtins.so.11.7 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvrtc-builtins.so.11.7
    libnvrtc-builtins.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvrtc-builtins.so
    libnvjpeg.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvjpeg.so.11
    libnvjpeg.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvjpeg.so
    libnvblas.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvblas.so.11
    libnvblas.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvblas.so
    libnvToolsExt.so.1 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvToolsExt.so.1
    libnvToolsExt.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnvToolsExt.so
    libnpps.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnpps.so.11
    libnpps.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnpps.so
    libnppitc.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppitc.so.11
    libnppitc.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppitc.so
    libnppisu.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppisu.so.11
    libnppisu.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppisu.so
    libnppist.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppist.so.11
    libnppist.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppist.so
    libnppim.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppim.so.11
    libnppim.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppim.so
    libnppig.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppig.so.11
    libnppig.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppig.so
    libnppif.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppif.so.11
    libnppif.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppif.so
    libnppidei.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppidei.so.11
    libnppidei.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppidei.so
    libnppicc.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppicc.so.11
    libnppicc.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppicc.so
    libnppial.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppial.so.11
    libnppial.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppial.so
    libnppc.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppc.so.11
    libnppc.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libnppc.so
    libicudata.so.67 (libc6,x86-64) => /lib/x86_64-linux-gnu/libicudata.so.67
    libicudata.so (libc6,x86-64) => /lib/x86_64-linux-gnu/libicudata.so
    libcusparse.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcusparse.so.11
    libcusparse.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcusparse.so
    libcusolverMg.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcusolverMg.so.11
    libcusolverMg.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcusolverMg.so
    libcusolver.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcusolver.so.11
    libcusolver.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcusolver.so
    libcurand.so.10 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcurand.so.10
    libcurand.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcurand.so
    libcuinj64.so.11.7 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcuinj64.so.11.7
    libcuinj64.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcuinj64.so
    libcufile_rdma.so.1 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufile_rdma.so.1
    libcufile_rdma.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufile_rdma.so
    libcufile.so.0 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufile.so.0
    libcufile.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufile.so
    libcufftw.so.10 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufftw.so.10
    libcufftw.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufftw.so
    libcufft.so.10 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufft.so.10
    libcufft.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcufft.so
    libcudart.so.11.0 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcudart.so.11.0
    libcudart.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcudart.so
    libcuda.so.1 (libc6,x86-64) => /lib/x86_64-linux-gnu/libcuda.so.1
    libcuda.so.1 (libc6) => /lib32/libcuda.so.1
    libcuda.so (libc6,x86-64) => /lib/x86_64-linux-gnu/libcuda.so
    libcuda.so (libc6) => /lib32/libcuda.so
    libcublasLt.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcublasLt.so.11
    libcublasLt.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcublasLt.so
    libcublas.so.11 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcublas.so.11
    libcublas.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libcublas.so
    libaccinj64.so.11.7 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libaccinj64.so.11.7
    libaccinj64.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libaccinj64.so
    libOpenCL.so.1 (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libOpenCL.so.1
    libOpenCL.so (libc6,x86-64) => /usr/local/cuda/targets/x86_64-linux/lib/libOpenCL.so

注：通过apt安装旧版CUDA时编译正常，但需要使用11.7版本。

解决方案 #

1. 补充CUDA头文件环境变量 #

cuda.h缺失是因为编译器找不到CUDA的头文件目录，需要添加头文件搜索路径到~/.bashrc：

export C_INCLUDE_PATH=${C_INCLUDE_PATH}:/usr/local/cuda-11.7/include
export CPLUS_INCLUDE_PATH=${CPLUS_INCLUDE_PATH}:/usr/local/cuda-11.7/include

添加后执行source ~/.bashrc使配置立即生效。

2. 修复autoconf库检测问题 #

autoconf脚本无法自动识别CUDA库路径时，需手动指定编译和链接参数：

• 运行autoconf相关脚本前，先设置以下环境变量：

export LDFLAGS="-L/usr/local/cuda-11.7/lib64 -L/usr/local/cuda/targets/x86_64-linux/lib"
export CFLAGS="-I/usr/local/cuda-11.7/include"
export CXXFLAGS="-I/usr/local/cuda-11.7/include"

• 再执行./configure等autoconf脚本，即可正常检测到cudart和curand库。

3. 统一CUDA路径软链接（可选） #

从ldconfig输出可见，系统默认通过/usr/local/cuda软链接访问CUDA资源，确保它指向正确的11.7版本：

sudo rm -f /usr/local/cuda
sudo ln -s /usr/local/cuda-11.7 /usr/local/cuda

此操作可避免后续工具因版本路径混乱导致的问题。

4. 验证配置有效性 #

创建测试文件test_cuda.cu：

#include <cuda.h>
#include <stdio.h>

int main() {
    int deviceCount;
    cudaGetDeviceCount(&deviceCount);
    printf("Detected %d CUDA devices\n", deviceCount);
    return 0;
}

用nvcc编译并运行：

nvcc test_cuda.cu -o test_cuda
./test_cuda

若正常输出设备数量，说明CUDA环境配置完成。

内容的提问来源于stack exchange，提问作者CGYak