Ubuntu 24.04 & CUDA 12 Support #923
Comments
|
Works like a charm, thanks |
|
Thank you very much!!! It saves me for hours of work & debugging. |
|
Did you encounter any issues with installing simple-knn except for including the float.h header as you mentioned @pens ? I seem unable to get this going on my system currently and im seeing some other discussions around the issues with simple-knn and different cuda versions but cant seem to fix the installation on my side. Ive looked into these but my system still does not want to finish the install Did you have any additional changes or install specific versions of gcc g++ or anything like that for your setup with wsl? |
To be honest, I don't recall. You mention WSL though, so I should point out that this was done on a bare metal Ubuntu installation. The system itself was more or less a clean install. |
|
Hi! I am using the patch file you provided. It still has some errors when I compile the SIBR. This is the way I use the patch file: # In the directory gaussian-splatting/SIBR_viewers
patch -p1 < SIBR_viewers.patch.txtThen I build by using the README way cmake -Bbuild . -DCMAKE_BUILD_TYPE=Release -G Ninja # add -G Ninja to build faster
cmake --build build --target install --config RelWithDebInfoHere is the error info, I picked some text. [36/136] Building CUDA object /<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/build/CMakeFiles/CudaRasterizer.dir/cuda_rasterizer/rasterizer_impl.cu.o
FAILED: /<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/build/CMakeFiles/CudaRasterizer.dir/cuda_rasterizer/rasterizer_impl.cu.o
/usr/local/cuda-12.6/bin/nvcc -forward-unknown-to-host-compiler -DBOOST_ALL_DYN_LINK -DBOOST_ALL_NO_LIB -DEIGEN_INITIALIZE_MATRICES_BY_ZERO -DGLEW_EGL -DOPENCV_TRAITS_ENABLE_DEPRECATED -I/usr/include/eigen3 -I/home/aruix/miniconda3/envs/3d/include -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/imgui -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/imgui/imgui -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/nativefiledialog -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/nativefiledialog/nativefiledialog -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/mrf -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/mrf/mrf -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/nanoflann -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/nanoflann/nanoflann -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/picojson -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/picojson/picojson -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/rapidxml -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/rapidxml/rapidxml -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/xatlas -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/xatlas/xatlas -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/xatlas/xatlas/source/xatlas -I/<my-path>/gaussian-splatting/SIBR_viewers/src/. -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer -I/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/third_party/glm -I/usr/local/cuda-12.6/targets/x86_64-linux/include -O3 -DNDEBUG -std=c++14 "--generate-code=arch=compute_70,code=[compute_70,sm_70]" "--generate-code=arch=compute_75,code=[compute_75,sm_75]" "--generate-code=arch=compute_86,code=[compute_86,sm_86]" -fpermissive -fPIC -Wall -Wno-unknown-pragmas -Wno-sign-compare -g -std=c++17 -D__forceinline="inline __attribute__((always_inline))" -MD -MT /<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/build/CMakeFiles/CudaRasterizer.dir/cuda_rasterizer/rasterizer_impl.cu.o -MF /<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/build/CMakeFiles/CudaRasterizer.dir/cuda_rasterizer/rasterizer_impl.cu.o.d -x cu -c /<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.cu -o /<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/build/CMakeFiles/CudaRasterizer.dir/cuda_rasterizer/rasterizer_impl.cu.o
nvcc warning : incompatible redefinition for option 'std', the last value of this option was used
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(24): error: namespace "std" has no member "uintptr_t"
std::size_t offset = (reinterpret_cast<std::uintptr_t>(chunk) + alignment - 1) & ~(alignment - 1);
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(40): error: identifier "uint32_t" is undefined
uint32_t* point_offsets;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(41): error: identifier "uint32_t" is undefined
uint32_t* tiles_touched;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(49): error: identifier "uint32_t" is undefined
uint32_t* n_contrib;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(58): error: identifier "uint64_t" is undefined
uint64_t* point_list_keys_unsorted;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(59): error: identifier "uint64_t" is undefined
uint64_t* point_list_keys;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(60): error: identifier "uint32_t" is undefined
uint32_t* point_list_unsorted;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.h(61): error: identifier "uint32_t" is undefined
uint32_t* point_list;
^
/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.cu(188): warning #549-D: variable "binning" is used before its value is set
obtain(chunk, binning.point_list, P, 128);
^
Remark: The warnings can be suppressed with "-diag-suppress <warning-number>"
8 errors detected in the compilation of "/<my-path>/gaussian-splatting/SIBR_viewers/extlibs/CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.cu". |
|
lifesaver!! 😍 |
|
@aruiplex, that file is missing the necessary include (https://en.cppreference.com/w/cpp/types/integer). Try adding |
|
Thank you, you are my God！！！ |
|
Hey @pens |
|
I used the following configuration： And I got this error: Pip subprocess error: × python setup.py bdist_wheel did not run successfully. note: This error originates from a subprocess, and is likely not a problem with pip. × python setup.py bdist_wheel did not run successfully. note: This error originates from a subprocess, and is likely not a problem with pip. × Running setup.py install for simple_knn did not run successfully. note: This error originates from a subprocess, and is likely not a problem with pip. × Encountered error while trying to install package. note: This is an issue with the package mentioned above, not pip. failed CondaEnvException: Pip failed |
|
thank you very much!!! |
Hello, did you solve it? I had the same problem |
|
I solve it by changing g++ version |
What did you change it do? |
|
Using a newer python version can solve the name: gaussian_splatting
channels:
- pytorch
- conda-forge
- defaults
- nvidia
dependencies:
- cuda-toolkit=12.4
- plyfile
- python=3.12.7
- pip=24.3.1
- pytorch
- torchaudio
- torchvision
- tqdm
- pytorch-cuda=12.4
- pip:
- submodules/diff-gaussian-rasterization
- submodules/simple-knnYou might also need to add a line of code #include <cinttypes>in |
|
For anyone still struggling, I added g++ as @kurryd suggested. |
6666666666666 Oh my GOd，NB |
I have the same issue.I solved it by change g++13\gcc13 version to g++10\gcc10 |
|
here is another pathset that should (tm) work: RongLiu-Leo/Gaussian-Splatting-Monitor#20 implemented with a lot of chatgpt help. |
Apologies as this isn't an issue per se.
I just finished getting training and the local viewer working on Ubuntu 24.04 with CUDA 12. As the process was a little involved I hope that sharing the changes needed might be useful.
environment.yml
submodules/simple-knn/simple-knn.cu
The newer NVIDIA headers (seem to) no longer export FLT_MIN & FLT_MAX, so
float.hneeds to be explicitly included.simple-knn.patch.txt
SIBR_viewers
Both Embree and FFmpeg have removed parts of their APIs that SIBR was using. I didn't test this with the remote viewer at all, as I was really just trying to get the local viewer building and running.
I updated the FFmpeg part with something that seemed in the right direction, but it's probably broken. I really have no idea if I hit the Embree path or not.
SIBR_viewers.patch.txt
CudaRasterizer
Also probably due to NVIDIA's headers,
CudaRasterizer/CudaRasterizer/cuda_rasterizer/rasterizer_impl.hrequirescstdintto be included.The text was updated successfully, but these errors were encountered: