CompileWithCUDA.md

Failed to parse CUDA nvcc implicit link information

CMake Error at /usr/share/cmake-3.16/Modules/CMakeDetermineCUDACompiler.cmake:187 (message):
    Failed to extract nvcc implicit link line.


Failed to parse CUDA nvcc implicit link information:
    no "PATH=" string found in "nvcc" output:
Failed to parse CUDA nvcc implicit link information:
    no "LIBRARIES=" string found in "nvcc" output:

solution

Add in CMakeLists.txt:

set(CMAKE_CUDA_COMPILER /usr/local/cuda-11.6/bin/nvcc)
find_package(CUDAToolkit)

CUDA_STANDARD is set to invalid value '17'

When set(CMAKE_CUDA_STANDARD 17) in CMakeLists and error occurs:

CUDA_STANDARD is set to invalid value '17'

Upgrade the cmake, the cmake installed by 'apt-get install' is old, upgrade it to 3.22.4 or later and the problem will be solved.

undefined symbol

ImportError: /home1/miniconda3/envs/dpvo/lib/python3.10/site-packages/dpviewerx.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN8pangolin7DisplayERKSs

check

readelf -d /home1/miniconda3/envs/dpvo/lib/python3.10/site-packages/dpviewerx.cpython-310-x86_64-linux-gnu.so

c++filt _ZN8pangolin7DisplayERKSs
nm -gDC /usr/local/lib/libpango_display.so | grep -i Display

solution

add compile definition in Pangolin CMakeLists.txt

add_compile_definitions(_GLIBCXX_USE_CXX11_ABI=0)