Cannot compile on Windows with CUDA

[QubaB]

I am not able compile NGsolve on windows with CUDA. I am obtaining error:

C:\ngsolve-netgen\build-CUDA\ngsolve\ngscuda>“C:\Programs\NVIDIA_CUDA\13_3/bin\nvcc.exe” --use-local-env -ccbin “C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.42.34433\bin\HostX64\x64” -x cu -rdc=true  -I"C:\ngsolve-netgen\src\ngscuda" -I"C:\ngsolve-netgen\src\solve" -I"C:\ngsolve-netgen\src\parallel" -I"C:\ngsolve-netgen\src\multigrid" -I"C:\ngsolve-netgen\src\bem" -I"C:\ngsolve-netgen\src\comp" -I"C:\ngsolve-netgen\src\linalg" -I"C:\ngsolve-netgen\src\fem" -I"C:\ngsolve-netgen\src\basiclinalg" -I"C:\ngsolve-netgen\src\ngstd" -I"C:\ngsolve-netgen\src\include" -I"C:\ngsolve-netgen\build-CUDA\ngsolve" -IC:\python313\Include -I"C:\install-CUDA\include\include" -I"C:\install-CUDA\include" -IC:\Programs\NVIDIA_CUDA\13_3\include -IC:\Programs\NVIDIA_CUDA\13_3\include\cccl -I"C:\Program Files (x86)\Intel\oneAPI\mkl\2024.0\include" -IC:\Programs\NVIDIA_CUDA\13_3\include     --keep-dir ngscudalib\x64\RelWithDebInfo  -maxrregcount=0     --machine 64 --compile -cudart static -std=c++17 --expt-relaxed-constexpr --extended-lambda --diag-suppress=611 --diag-suppress=20011 --diag-suppress=20012 --diag-suppress=20013 --diag-suppress=20014 --diag-suppress=20015 -rdc=true /bigobj /arch:AVX512 /std:c++17 /wd4068 -Xcompiler=“/EHsc -Zi -Ob1”   -D_WINDOWS -DNDEBUG -DMAX_SYS_DIM=3 -DNGS_EXPORTS -DCUDA -DNETGEN_PYTHON -DNG_PYTHON -DPYBIND11_SIMPLE_GIL_MANAGEMENT -D_WIN32_WINNT=0x1000 -DWNT -DWNT_WINDOW -DNOMINMAX -DMSVC_EXPRESS -D_CRT_SECURE_NO_WARNINGS -DHAVE_STRUCT_TIMESPEC -DWIN32 -DHAVE_NETGEN_SOURCES -DUSE_TIMEOFDAY -DTCL -DLAPACK -DUSE_PARDISO -DNGS_PYTHON -DUSE_UMFPACK -D"CMAKE_INTDIR="RelWithDebInfo"" -Dngscudalib_EXPORTS -D_WINDLL -D_MBCS -DWIN32 -D_WINDOWS -DNDEBUG -DMAX_SYS_DIM=3 -DNGS_EXPORTS -DCUDA -DNETGEN_PYTHON -DNG_PYTHON -DPYBIND11_SIMPLE_GIL_MANAGEMENT -D_WIN32_WINNT=0x1000-DWNT -DWNT_WINDOW -DNOMINMAX -DMSVC_EXPRESS -D_CRT_SECURE_NO_WARNINGS -DHAVE_STRUCT_TIMESPEC -DHAVE_NETGEN_SOURCES -DUSE_TIMEOFDAY -DTCL -DLAPACK -DUSE_PARDISO -DNGS_PYTHON -DUSE_UMFPACK -D"CMAKE_INTDIR="RelWithDebInfo"" -Dngscudalib_EXPORTS -Xcompiler “/EHsc /W1 /nologo /O2 /FS /Zi  /MD /GR” -Xcompiler “/Fdngscudalib.dir\RelWithDebInfo\vc143.pdb” -o ngscudalib.dir\RelWithDebInfo\cuda_profiler.obj “C:\ngsolve-netgen\src\ngscuda\cuda_profiler.cu”

nvcc fatal : A single input file is required for a non-link phase when an outputfile is specified

I think problem is with:

/bigobj /arch:AVX512 /std:c++17 /wd4068

which are C++ parameter not nvcc.

If I manually change this part of command line to

-Xcompiler"/bigobj /arch:AVX512 /std:c++17 /wd4068"

then compilation finishes.

I am not enough skilled to change ngscuda\CMakeLists.txt to properly generate command line.