This fix does indeed resolve the cmake error, but the generated solution won't properly compile any .cu file

>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

I've noticed that cmake trace shows
-- CUDA detected: 10.2
-- CUDA NVCC target flags: -gencode;arch=compute_30,code=sm_30;-gencode;arch=compute_35,code=sm_35;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_52,code=sm_52;-gencode;arch=compute_53,code=sm_53;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-D_FORCE_INLINES
when it used to show
-- CUDA NVCC target flags: -ccbin;$(VCInstallDir)Tools/MSVC/$(VCToolsVersion)/bin/Host$(Platform)/$(PlatformTarget);-gencode;arch=compute_30,code=sm_30;-gencode;arch=compute_35,code=sm_35;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_52,code=sm_52;-gencode;arch=compute_53,code=sm_53;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-D_FORCE_INLINES

@JulienMaille BTW, CMake output for CUDA NVCC target flags differs between separate launches (it is a bug - run of cmake . should be idempotent).
Clear CMake cache files to get stable result until bug got investigated/fixed.

Thank you @alalek
I tried from a clean build folder, but issue is the same. Actually I spent quite some time reverting changes until I found the section that breaks compilation on my setup:

@JulienMaille could you please share your environment variable situation, especially the CUDA_HOST_COMPILER please ?
i.e. try following command on command prompt

>echo %CUDA_HOST_COMPILER%

I expect that you see

$(VCInstallDir)Tools/MSVC/$(VCToolsVersion)/bin/Host$(Platform)/$(PlatformTarget)

when I see

%CUDA_HOST_COMPILER%

as a result. The output on my side indicates that this variable is not set.
This environment variable needs to be set manually or otherwise it's supposed to be not set.
Did you set it manually ? If so, could you let me know what's the purpose ?
If you didn't set by purpose, is it possible to remove this environment variable, please ?

I did not set CUDA_HOST_COMPILER and in my terminal echo shows it is not set. I suppose this is set earlier by cmake?

@JulienMaille
What I can only say is, please add message(STATUS "CUDA_HOST_COMPILER ${CUDA_HOST_COMPILER}") some where before here and share the result here.

Please please make sure that you start from clean directory to avoid cache of CMake

There are several patches about ccbin which breaks logic in CMake's CUDA scripts (from CMake).
My suggestion is to revert these patches (most of them are on the master branch only). PR will be created soon.

Until that do not use cmake-gui or run cmake multiple times for the same build directory (or perform cleaning of CMake cache from build directory)