Fails to build with CUDA enabled

[moooV252]

System information (version)

• OpenCV => 4.3.0-527-ga711e2aa41 (master latest)
• Operating System / Platform => Windows 64 Bit
• Compiler => Visual Studio 2019

Detailed description

Have cloned latest master branch, as well as opencv-contrib master branch to separate folders.
Cuda version: 11.0RC (latest)

Build fails with an error:

Building NVCC (Device) object modules/world/CMakeFiles/cuda_compile_1.dir/__/core/src/cuda/Release/cuda_compile_1_generated_gpu_mat.cu.obj
nvcc fatal   : A single input file is required for a non-link phase when an outputfile is specified
CMake Error at cuda_compile_1_generated_gpu_mat.cu.obj.Release.cmake:220 (message):
Error generating
D:/_UTILS/_deps/opencv_build/modules/world/CMakeFiles/cuda_compile_1.dir/__/core/src/cuda/Release/cuda_compile_1_generated_gpu_mat.cu.obj

The same exact building process worked fine with opencv-4.2.0 and cuda 10.2

Steps to reproduce

1. Clone opencv and opencv-contrib from git
2. init vars for MKL and TBB:

"C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\tbb\bin\tbbvars.bat" intel64
"C:\Program Files (x86)\IntelSWTools\compilers_and_libraries\windows\mkl\bin\mklvars.bat" intel64

3. Cmake it:

cmake ..\opencv -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=C:\opencv -DOPENCV_ENABLE_NONFREE:BOOL=ON -DOPENCV_EXTRA_MODULES_PATH=..\opencv_contrib\modules -DBUILD_opencv_python2:BOOL=OFF -DBUILD_opencv_python3:BOOL=ON -DWITH_NVCUVID=OFF -DWITH_CUDA=ON -DCUDA_FAST_MATH=ON -DWITH_CUBLAS=ON -DWITH_MKL=ON -DMKL_USE_MULTITHREAD=ON -DMKL_WITH_TBB=ON -DWITH_TBB=ON -DINSTALL_TESTS=OFF -DINSTALL_C_EXAMPLES=OFF -DBUILD_TESTS=OFF -DBUILD_EXAMPLES=OFF -DCUDNN_LIBRARY=C:\cudnn\lib\x64\cudnn.lib -DCUDNN_INCLUDE_DIR=C:\cudnn\include -DWITH_VTK=OFF -DBUILD_JAVA=OFF -DBUILD_opencv_java_bindings_generator=OFF -DBUILD_opencv_world=ON -DBUILD_opencv_gapi=OFF -DCUDA_ARCH_PTX=6.1 -DCUDA_ARCH_BIN=6.1

4. Build with:

msbuild INSTALL.vcxproj -m -p:Configuration=Release

5. It fails. 😲

Here's the full console output:
opencv_build_fail_log.txt

Issue submission checklist

• [v] I report the issue, it's not a question
• [v] I checked the problem with documentation, FAQ, open issues,
  answers.opencv.org, Stack Overflow, etc and have not found solution
• [v] I updated to latest OpenCV version and the issue is still there
• [v] There is reproducer code and related data files: videos, images, onnx, etc

[alalek]

Could you please check code from 3.4 branch?

[moooV252]

I've checked out opencv 3.4 branch plus opencv-contrib 3.4 branch.

Now it fails to configure with an error:

CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
CUDA_nppicom_LIBRARY (ADVANCED)

I've taken a look at the cuda folder - the required library is missing indeed. Maybe they've refactored it out in cuda11, who knows. At least, it was a clean cuda installation on my machine.

The full console output is here:
opencv_3.4_config_fail_log.txt

[iwilliamson]

I'm seeing this now too.

OpenCV => master latest
Operating System / Platform => Windows 64 Bit
Compiler => Visual Studio 2019

set CMAKE_CONFIG_GENERATOR="Visual Studio 14 2015 Win64"
set CMAKE_OPTIONS=-DBUILD_PERF_TESTS:BOOL=OFF -DBUILD_TESTS:BOOL=OFF -DBUILD_DOCS:BOOL=OFF
set CMAKE_OPTIONS=%CMAKE_OPTIONS% -DWITH_CUDA:BOOL=ON
set CMAKE_OPTIONS=%CMAKE_OPTIONS% -DBUILD_EXAMPLES:BOOL=ON -DINSTALL_CREATE_DISTRIB=ON
cmake -G%CMAKE_CONFIG_GENERATOR% %CMAKE_OPTIONS% ..
cmake --build . --config release

Worked fine about a month ago. Using CUDA 10.2

[moooV252]

I've downgraded cuda to 10.2 and tested it.
4.4.0-pre (master latest) fails to build the same way it did with 11RC
4.3.0-release builds fine but does not load to python saying that DLLs are missing. Dependency walker does not show any anomalies.

It's really strage - if it's builds ok then it's able to find everything it needs. But if you load this module into python it can't.
All environment variables are exactly the same.

Python 3.8.3 (tags/v3.8.3:6f8c832, May 13 2020, 22:37:02) [MSC v.1924 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import cv2
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Python38\lib\site-packages\cv2\__init__.py", line 96, in <module>
    bootstrap()
  File "C:\Python38\lib\site-packages\cv2\__init__.py", line 86, in bootstrap
    import cv2
ImportError: DLL load failed while importing cv2: The specified module could not be found.

[thangict210]

I just comment out the following block of code and it works perfectly.

[iwilliamson]

@thangict210 - I can confirn that commenting out the CMAKE code you pointed out worked for me as well.

[onism26]

cuda accelerate image/video decode/encode ,gpuMat need nvidia Video_Codec_SDK and nv-codec-headers(download here: git clone https://git.videolan.org/git/ffmpeg/nv-codec-headers.git)

[OOF-dura]

I just comment out the following block of code and it works perfectly.

What should we set on Ubuntu?