NVlabs / tiny-cuda-nn

Lightning fast C++/CUDA neural network framework
Other
3.67k stars 449 forks source link

help !install torch binding failed ! #304

Open epicgzs1112 opened 1 year ago

epicgzs1112 commented 1 year ago

ubuntu22.04 cuda11.3 pytorch 1.10.0
i have build tinycudann successful , but when i do excute tiny-cuda-nn/bindings/torch$ python setup.py install i got error:

g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/tinycudann/bindings.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../dependencies/fmt/src/format.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../dependencies/fmt/src/os.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/cpp_api.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/common.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/common_device.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/encoding.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/network.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/cutlass_mlp.o: No such file or directory g++: error: /home/lch/Downloads/nvdiffrec-main/tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/../../src/fully_fused_mlp.o: No such file or directory error: command 'g++' failed with exit status 1

MenxLi commented 1 year ago

Did you install torch in a conda enviroment? I had the same issue and re-install torch and others via pip, then no errors again, you can have a try

YG256Li commented 1 year ago

@MenxLi I meet the same issue,can you share your torch, cuda, gcc, pip version?

YG256Li commented 1 year ago

@epicgzs1112 the same error. And can build tinycudann successful, but cannot python setup.py install to torch. have you solve the problem?

MenxLi commented 1 year ago

@MenxLi I meet the same issue,can you share your torch, cuda, gcc, pip version?

I have ubuntu18.04, cuda11.7, gcc10.3, (succeed in pip23.1.2, torch2.0.1) / (failed in pip23.0.1, torch2.0)

QwertyLqLq commented 1 year ago

@MenxLi I meet the same issue,can you share your torch, cuda, gcc, pip version?

I have ubuntu18.04, cuda11.7, gcc10.3, (succeed in pip23.1.2, torch2.0.1) / (failed in pip23.0.1, torch2.0)

For me, succeed in cuda11.7+torch2.0.0, but failed in cuda11.7+torch2.0.1😂

ArtoriasAbyssslayer commented 1 year ago

Is there any way to resolve the build failures?

dlehgo14 commented 11 months ago

With cuda11.5 + torch2.0.0, I succeed to install (Ubuntu 22.04, 4090). With 2.1.0 torch, I failed.

dahaigui commented 5 months ago

I found the solution to this problem! There is a file generated during the building process, the path is tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-3.8/build.ninja. we need to use ninja to build it, the command is ninja -f build .ninja, I also have a building process in the error Python.h:44:10: fatal error: crypt.h: No such file or directory check and found that there is no header file, so I want to find ways to add header files, after searching a variety of network resources, I found that it is because there is no install the libxcrypt library, use conda install libxcrypt to install it, then use ninja -f build.ninja to build, then generate tiny-cuda-nn/bindings/torch/build/temp.linux-x86_64-cpython -38/build.ninja, then ninja -f build.ninja, followed by python installation, python setup.py install can be installed successfully.

chais-xp commented 5 months ago

@MenxLi I meet the same issue,can you share your torch, cuda, gcc, pip version?

I have ubuntu18.04, cuda11.7, gcc10.3, (succeed in pip23.1.2, torch2.0.1) / (failed in pip23.0.1, torch2.0)

thank you very much , this way solve my question,