CMAKE_CUDA_COMPILER 尽管找到了 cuda，但标志是假的

Question

我的 cmake 似乎无法找到 cuda 编译器。将 cuda 作为包查找成功，但是 CMAKE_CUDA_COMPILER 设置为 false。以下输出由 cmake 生成：

-- The C compiler identification is GNU 9.4.0
-- The CXX compiler identification is GNU 9.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Build type not specified, using Release
-- Looking for a CUDA compiler
-- Looking for a CUDA compiler - NOTFOUND
-- CUDA Support disabled.
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE  
-- Found CUDA: /usr/local/cuda-10.2 (found suitable exact version "10.2") 
-- Found GDAL: /usr/lib/libgdal.so (found version "2.2.3") 
-- Found Boost: /usr/include (found version "1.65.1") found components:  filesystem system 
-- Found EXIV2: /usr/lib/x86_64-linux-gnu/libexiv2.so  
-- Checking for module 'eigen3'
--   Found eigen3, version 3.3.9
-- Found eigen: /usr/local/include/eigen3  
-- Found Boost: /usr/include (found suitable version "1.65.1", minimum required is "1.40.0") found components:  system filesystem thread date_time iostreams serialization chrono atomic regex 
-- Checking for module 'libopenni'
--   Found libopenni, version 1.5.4.0
-- Found openni: /usr/lib/libOpenNI.so  
-- Checking for module 'libopenni2'
--   Found libopenni2, version 2.2.0.3
-- Found OpenNI2: /usr/lib/libOpenNI2.so  
-- Could NOT find ensenso (missing: ENSENSO_LIBRARY ENSENSO_INCLUDE_DIR) 
** WARNING ** io features related to ensenso will be disabled
-- Could NOT find DAVIDSDK (missing: DAVIDSDK_LIBRARY DAVIDSDK_INCLUDE_DIR) 
** WARNING ** io features related to davidSDK will be disabled
-- Could NOT find DSSDK (missing: _DSSDK_LIBRARIES) 
** WARNING ** io features related to dssdk will be disabled
** WARNING ** io features related to pcap will be disabled
** WARNING ** io features related to png will be disabled
-- Found Boost: /usr/include (found version "1.65.1") found components:  system filesystem 
-- CUDA not found. Skipping PSL package...
-- Found OpenCV: /usr/local (found version "3.3.1") 
-- Found Eigen3: /usr/local/include/eigen3 (Required is at least version "2.91.0") 
-- Found OpenGL: /usr/lib/x86_64-linux-gnu/libOpenGL.so   
-- Found OpenMP_C: -fopenmp (found version "4.5") 
-- Found OpenMP_CXX: -fopenmp (found version "4.5") 
-- Found OpenMP: TRUE (found version "4.5")  
-- Configuring done
-- Generating done

相关行是：

-- Looking for a CUDA compiler - NOTFOUND
-- CUDA Support disabled.
-- Found CUDA: /usr/local/cuda-10.2 (found suitable exact version "10.2")

我在 CMake 3.15.7 中用于检查 cuda 的代码：

check_language(CUDA)
if(CMAKE_CUDA_COMPILER)
    enable_language(CUDA)
    message(STATUS "CUDA Support enabled.")

    include(FindCUDA)
    set(CUDA_ARCH_LIST Auto CACHE STRING  "List of CUDA architectures (e.g. Pascal, Volta, etc) or \
                                           compute capability versions (6.1, 7.0, etc) to generate code for. \
                                           Set to Auto for automatic detection (default).")
    cuda_select_nvcc_arch_flags(CUDA_ARCH_FLAGS ${CUDA_ARCH_LIST})
    list(APPEND CUDA_NVCC_FLAGS ${CUDA_ARCH_FLAGS})
else()
    message(STATUS "CUDA Support disabled.")
endif()

这是怎么回事？我发誓它在某些时候起作用了。但现在已经不是了。 nvcc --version 和 nvidia-smi 给出了合理的输出。我的 .bashrc 看起来像这样：

export CPATH=/usr/local/cuda-10.2/include:$CPATH
export PATH=/usr/local/cuda-10.2/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

Answer 1

好的，我找到问题了。每个 CUDA 版本只支持特定的 gcc 版本。兼容性可以在这里找到：

在安装 CUDA 之后，我更改了我的默认 gcc 版本，导致尽管 nvcc 和 nvidia-smi 正常工作，但 cmake 的这种奇怪行为无法检测到 cuda。我完全清除了 CUDA 和 Nvidia 驱动程序，设置了有效的 gcc 版本并重新安装了 CUDA + 驱动程序。它现在似乎工作正常。

CMAKE_CUDA_COMPILER 尽管找到了 cuda，但标志是假的

CMAKE_CUDA_COMPILER flag is false despite cuda being found

c++

cuda

cmake