Error when compiling : tiny-cuda-nn and error MSB3721

[gregbeast]

Hi community,

the last step "cmake --build build --config RelWithDebInfo -j 16" generate some errors.
All steps before seems good.

`tiny-cuda-nn\src\fully_fused_mlp.cu(416): error : name followed by "::" must be a class or namespace nam
e [C:\AI\TinyCuda\tiny-cuda-nn\build\src\tiny-cuda-nn.vcxproj]

C:\AI\TinyCuda\tiny-cuda-nn\src\fully_fused_mlp.cu(496): error : name followed by "::" must be a class or namespace nam
e [C:\AI\TinyCuda\tiny-cuda-nn\build\src\tiny-cuda-nn.vcxproj]

C:\AI\TinyCuda\tiny-cuda-nn\src\fully_fused_mlp.cu(496): error : name followed by "::" must be a class or namespace nam
e [C:\AI\TinyCuda\tiny-cuda-nn\build\src\tiny-cuda-nn.vcxproj]

C:\AI\TinyCuda\tiny-cuda-nn\include\tiny-cuda-nn/encodings/grid.h(249): error : no operator "+=" matches these operands
[C:\AI\TinyCuda\tiny-cuda-nn\build\src\tiny-cuda-nn.vcxproj]

:\Program Files (x86)\Microsoft Visual Studio\2019\Community\MSBuild\Microsoft\VC\v160\BuildCustomizations\CUDA 11.6.t
argets(790,9): error MSB3721: The command ""C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin\nvcc.exe" -gen
code=arch=compute_52,code="sm_52,compute_52" --use-local-env -ccbin "C:\Program Files (x86)\Microsoft Visual Studio\2
019\Community\VC\Tools\MSVC\14.29.30133\bin\HostX64\x64" -x cu -I"C:\AI\TinyCuda\tiny-cuda-nn\include" -I"C:\AI\TinyC
uda\tiny-cuda-nn\dependencies" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\include" --keep-dir x64
RelWithDebInfo -maxrregcount=0 --machine 64 --compile -cudart static --extended-lambda --expt-relaxed-constexpr -std=
c++14 -Xcompiler="/EHsc -Zi -Ob1" -D_WINDOWS -DNDEBUG -DTCNN_MIN_GPU_ARCH=86 -DTCNN_SHAMPOO -D"CMAKE_INTDIR="RelWith
DebInfo"" -D_MBCS -D"CMAKE_INTDIR="RelWithDebInfo"" -Xcompiler "/EHsc /W1 /nologo /O2 /FdC:\AI\TinyCuda\tiny-cuda-nn
\build\src\RelWithDebInfo\tiny-cuda-nn.pdb /FS /Zi /MD /GR" -o tiny-cuda-nn.dir\RelWithDebInfo\cutlass_mlp.obj "C:\AI
TinyCuda\tiny-cuda-nn\src\cutlass_mlp.cu"" exited with code 1. [C:\AI\TinyCuda\tiny-cuda-nn\build\src\tiny-cuda-nn.vcxp
roj]`

My environment:
ThreadRipper 3970 X
GeForce RTX 3090
Win10
Cuda 11.6
Cmake 3.23
Visual Studio 2019 updated
OptiX 7.4
Pyhton 3.9

Let me know if you know this issue and feel free to share tips,

[vBitreel]

I am having this issue as well and my environment is exactly identical to yours.

[Tom94]

It looks like CMake generated sm_52,compute_52 nvcc arguments despite having detected your GPU architecture correctly (-DTCNN_MIN_GPU_ARCH=86). This is puzzling to me and shouldn't happen -- the correct arch arguments would be sm_86,compute_86.

Unfortunately, everything seems to be working fine my machine on the latest CMake, CUDA, and VS 2019, so I can't easily look for a fix.

Shot in the dark: perhaps hardcoding the appropriate flags

list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_86,code=\"sm_86,compute_86\"")

near the top of instant-ngp/CMakeLists.txt and instant-ngp/dependencies/tiny-cuda-nn/CMakeLists.txt will fix it on your systems.

[Ghanimalg]

what do you mean by that? im new to this sorry

[vBitreel]

Thank you, hardcoding the flags seems to work!

If anyone like me finds themselves wondering how to escape the quotes correctly, I ended up using:
list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_86,code=\"sm_86,compute_86\"")

[gregbeast]

I confirm! Works like a charm.
Thank you!

[braveness23]

I'm using a GeForce RTX 4000 and setting it to

list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_75,code=\"sm_75,compute_75\"")

worked for me!