Compilation issues

[adw62]

Hello,

I'm having trouble compiling this project on my system. I'm using Ubuntu 20.04, RTX 2070 super, CUDA 11.6/ Driver Version: 510.47, cmake 3.23.0-rc1 and gcc 9.3.0.

The errors I get are when running:
tiny-cuda-nn$ cmake --build build --config RelWithDebInfo -j 16
and I see:

[  5%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/common_device.cu.o
[ 15%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/common.cu.o
[ 15%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/cpp_api.cu.o
[ 21%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/cutlass_mlp.cu.o
[ 26%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/encoding.cu.o
[ 31%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/cutlass_resnet.cu.o
[ 36%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/loss.cu.o
[ 42%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/network.cu.o
[ 47%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/object.cu.o
[ 52%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/optimizer.cu.o
[ 57%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/reduce_sum.cu.o
[ 63%] Building CUDA object src/CMakeFiles/tiny-cuda-nn.dir/fully_fused_mlp.cu.o
/home/a/Downloads/tiny-cuda-nn-1.4/src/fully_fused_mlp.cu(416): error: explicit type is missing ("int" assumed)

/home/a/Downloads/tiny-cuda-nn-1.4/src/fully_fused_mlp.cu(416): error: expected a ")"

/home/a/Downloads/tiny-cuda-nn-1.4/src/fully_fused_mlp.cu(496): error: explicit type is missing ("int" assumed)

/home/a/Downloads/tiny-cuda-nn-1.4/src/fully_fused_mlp.cu(496): error: expected a ")"

etc ... I have attached the full output for reference err.txt

Ultimately I would like to use tiny-cuda-nn with https://github.com/NVlabs/instant-ngp but I see the same problems with tiny-cuda-nn when compiling that also. Any help with this would be appreciated.

Cheers,
Alex

[OctoberKat]

I have met these errors too. It is caused by the namespace nvcuda not imported successfully. Refer to this website: https://stackoverflow.com/questions/53634287/cuda-fails-to-recognize-nvcuda-namespace-during-compilation
Specifying the correct architecture makes the error disappear:
nvcc -arch=sm_70 blablabla.cu
Just append the following code in 130 line of CMakeLists.txt solved my issue:
list(APPEND CUDA_NVCC_FLAGS "-arch=sm_70")

[adw62]

Great find, this fix also works for me.

[Holmes-Alan]

I have met these errors too. It is caused by the namespace nvcuda not imported successfully. Refer to this website: https://stackoverflow.com/questions/53634287/cuda-fails-to-recognize-nvcuda-namespace-during-compilation Specifying the correct architecture makes the error disappear: nvcc -arch=sm_70 blablabla.cu Just append the following code in 130 line of CMakeLists.txt solved my issue: list(APPEND CUDA_NVCC_FLAGS "-arch=sm_70")

which CMakeLists.txt should I modify? The one under the the major folder "tiny-cuda-nn" or somewhere under "tiny-cuda-nn/src"? I tried to add this: list(APPEND CUDA_NVCC_FLAGS "-arch=sm_70") but it does not work.

[adw62]

The main CMake if I remember. It has since been updated I put the APPEND line here

tiny-cuda-nn/CMakeLists.txt

Line 130 in e7826c3

list(APPEND CUDA_NVCC_FLAGS "--extended-lambda")

which is now here

tiny-cuda-nn/CMakeLists.txt

Line 137 in bd29d77

list(APPEND CUDA_NVCC_FLAGS "--extended-lambda")

Maybe moving it to line 137 would work?