0

我正在尝试从页面中给出的源文件安装 CUDF(https://github.com/rapidsai/cudf/blob/branch-0.15/CONTRIBUTING.md#setting-up-your-build-environment

在以下几个步骤之后,我无法通过执行 cmake 来完成它

在其依赖项之后构建并安装 libcudf。CMake 取决于 nvcc 可执行文件在您的路径上或在 $CUDACXX 中定义。

$ cd $CUDF_HOME/cpp  
$ mkdir build
$ cd build 

然后它说

-DCMAKE_INSTALL_PREFIX 设置为库的安装路径或 $CONDA_PREFIX 如果您使用 Anaconda,即 -DCMAKE_INSTALL_PREFIX=/install/path 或 -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX# -DCMAKE_CXX11_ABI 设置为 ON 或 OFF,具体取决于您想要的 ABI 版本,默认为开。打开时,使用 C++11 的 ABI 兼容性。关闭时,使用 C++11 之前的 ABI 兼容性。

cmake .. -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX -DCMAKE_CXX11_ABI=ON  

然后它显示如下错误

cmake: /usr/local/lib/libcurl.so.4: no version information available (required by cmake)
CMake Error at /usr/share/cmake-3.16/Modules/CMakeDetermineCompilerId.cmake:219 (file):
  file failed to open for writing (Permission denied):

    /home/CUDF/cudf/cpp/build/CMakeFiles/3.16.3/CompilerIdCUDA/CMakeCUDACompilerId.cu
Call Stack (most recent call first):
  /usr/share/cmake-3.16/Modules/CMakeDetermineCompilerId.cmake:230 (CMAKE_DETERMINE_COMPILER_ID_WRITE)
  /usr/share/cmake-3.16/Modules/CMakeDetermineCompilerId.cmake:32 (CMAKE_DETERMINE_COMPILER_ID_BUILD)
  /usr/share/cmake-3.16/Modules/CMakeDetermineCUDACompiler.cmake:72 (CMAKE_DETERMINE_COMPILER_ID)
  CMakeLists.txt:18 (project)


CMake Error at /usr/share/cmake-3.16/Modules/CMakeDetermineCompilerId.cmake:219 (file):
  file failed to open for writing (Permission denied):

    /home/CUDF/cudf/cpp/build/CMakeFiles/3.16.3/CompilerIdCUDA/CMakeCUDACompilerId.cu
Call Stack (most recent call first):
  /usr/share/cmake-3.16/Modules/CMakeDetermineCompilerId.cmake:230 (CMAKE_DETERMINE_COMPILER_ID_WRITE)
  /usr/share/cmake-3.16/Modules/CMakeDetermineCompilerId.cmake:32 (CMAKE_DETERMINE_COMPILER_ID_BUILD)
  /usr/share/cmake-3.16/Modules/CMakeDetermineCUDACompiler.cmake:72 (CMAKE_DETERMINE_COMPILER_ID)
  CMakeLists.txt:18 (project)


-- The CUDA compiler identification is unknown
-- Check for working CUDA compiler: /home/anaconda3/envs/cudf_dev/
-- Check for working CUDA compiler: /home/anaconda3/envs/cudf_dev/ -- broken
CMake Error at /usr/share/cmake-3.16/Modules/CMakeTestCUDACompiler.cmake:46 (message):
  The CUDA compiler

    "/home/anaconda3/envs/cudf_dev/"

  is not able to compile a simple test program.

  It fails with the following output:

    Change Dir: /home/CUDF/cudf/cpp/build/CMakeFiles/CMakeTmp
    
    Run Build Command(s):/usr/bin/make cmTC_9f254/fast && /usr/bin/make -f CMakeFiles/cmTC_9f254.dir/build.make CMakeFiles/cmTC_9f254.dir/build
    make[1]: Entering directory '/home/CUDF/cudf/cpp/build/CMakeFiles/CMakeTmp'
    /usr/bin/cmake: /usr/local/lib/libcurl.so.4: no version information available (required by /usr/bin/cmake)
    Building CUDA object CMakeFiles/cmTC_9f254.dir/main.cu.o
    /home/anaconda3/envs/cudf_dev/     -x cu -c /home/CUDF/cudf/cpp/build/CMakeFiles/CMakeTmp/main.cu -o CMakeFiles/cmTC_9f254.dir/main.cu.o
    make[1]: execvp: /home/anaconda3/envs/cudf_dev/: Permission denied
    make[1]: *** [CMakeFiles/cmTC_9f254.dir/build.make:66: CMakeFiles/cmTC_9f254.dir/main.cu.o] Error 127
    make[1]: Leaving directory '/home/CUDF/cudf/cpp/build/CMakeFiles/CMakeTmp'
    make: *** [Makefile:121: cmTC_9f254/fast] Error 2
    
    

  

  CMake will not be able to correctly generate this project.
Call Stack (most recent call first):
  CMakeLists.txt:18 (project)


-- Configuring incomplete, errors occurred!
See also "/home/CUDF/cudf/cpp/build/CMakeFiles/CMakeOutput.log".
See also "/home/CUDF/cudf/cpp/build/CMakeFiles/CMakeError.log".
k@k-HP:~/CUDF/cudf/cpp/build$ 

这一步我该怎么办?

Ps:我的 anaconda 路径就像 /home/anaconda3/ 和 cudf 环境路径就像 /home/anaconda3/envs/cudf_dev

4

1 回答 1

1

请注意,cudf 0.15 不支持 CUDA 10.0 和 python 3.6。

这似乎是一个 CUDA/权限问题。

您是否以 root 身份安装/构建了您的 conda env,现在您正尝试以用户身份运行构建?

你的输出是为了nvidia-smi什么?

此外,如果您安装了 CUDA 工具包并且 env 的 cuda 版本与 nvidia-smi 中显示的 CUDA 版本匹配。然后我建议您尝试从文件夹运行./build.sh命令。跑步对我有用。/home/CUDF/cudf/./build.sh

如果上述解决方案不起作用,您可以随时在 cudf repo ( https://github.com/rapidsai/cudf/issues ) 中提出问题,并提供文件的输出print_env.sh

您还可以使用以下命令安装每晚版本的 cudf-0.15:

conda install -c rapidsai-nightly -c nvidia -c conda-forge -c defaults cudf=0.15 python=3.7 cudatoolkit=10.1

您需要cudatoolkit根据系统中安装的 CUDA 工具包版本更新版本。

于 2020-07-28T15:48:06.440 回答