Error building extension '_pvcnn_backend' about pvcnn HOT 5 CLOSED

I think it's related to your CUDA environment since there is the following error in your message:

/usr/local/cuda/bin/nvcc: not found

from pvcnn.

I think it's related to your CUDA environment since there is the following error in your message:

/usr/local/cuda/bin/nvcc: not found

here is my ~/.bashrc:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/x86_64-linux-gnu
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib/x86_64-linux-gnu
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export PATH=$PATH:/usr/local/cuda/bin
export CUDA_HOME=$CUDA_HOME:/usr/local/cuda

And I run other model with cuda is ok like this:

(pc) gz@gz:~$ python
Python 3.8.5 (default, Sep  4 2020, 07:30:14) 
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.cuda.is_available())
True
>>> a = torch.Tensor([1]).cuda()
>>> print(a)
tensor([1.], device='cuda:0')

from pvcnn.

(pc) gz@gz:~$ python
Python 3.8.5 (default, Sep  4 2020, 07:30:14) 
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.cuda.is_available())
True
>>> a = torch.Tensor([1]).cuda()
>>> print(a)
tensor([1.], device='cuda:0')

@LeopoldACC Before this, did you check nvcc --version and which nvcc? @zhijian-liu is correct to identify the problem. Your installed version of torch contains precompiled binaries therefore does not explicitly need nvcc to run, it just needs working cudnn for its operations. On the contrary, here you are trying to compile source code written in cuda, so you need a functioning nvcc.

from pvcnn.

(pc) gz@gz:~$ python
Python 3.8.5 (default, Sep  4 2020, 07:30:14) 
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.cuda.is_available())
True
>>> a = torch.Tensor([1]).cuda()
>>> print(a)
tensor([1.], device='cuda:0')

@LeopoldACC Before this, did you check nvcc --version and which nvcc? @zhijian-liu is correct to identify the problem. Your installed version of torch contains precompiled binaries therefore does not explicitly need nvcc to run, it just needs working cudnn for its operations. On the contrary, here you are trying to compile source code written in cuda, so you need a functioning nvcc.

I run the command without problem.

(pc) gz@gz:~$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Tue_Sep_15_19:10:02_PDT_2020
Cuda compilation tools, release 11.1, V11.1.74
Build cuda_11.1.TC455_06.29069683_0
(pc) gz@gz:~$ which nvcc
/usr/local/cuda/bin/nvcc

Here is log of cuda file:

(pc) gz@gz:~$ cd /usr/local/cuda/bin && ls
bin2c                         cuobjdump                    nvcc.profile
computeprof                   fatbinary                    nvdisasm
compute-sanitizer             ncu                          nvlink
crt                           ncu-ui                       nv-nsight-cu
cudafe++                      nsight_ee_plugins_manage.sh  nv-nsight-cu-cli
cuda-gdb                      nsight-sys                   nvprof
cuda-gdbserver                nsys                         nvprune
cuda-install-samples-11.1.sh  nsys-exporter                nvvp
cuda-memcheck                 nsys-ui                      ptxas
cuda-uninstaller              nvcc

from pvcnn.

Hi,@zhijian-liu @digital-idiot
I solved the problem refer to the link
change the export CUDA_HOME=$CUDA_HOME:/usr/local/cuda to export CUDA_HOME=/usr/local/cuda in ~/.bashrc

now my ~/.bashrc is shown as below:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/x86_64-linux-gnu
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib/x86_64-linux-gnu
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export PATH=$PATH:/usr/local/cuda/bin
export CUDA_HOME=/usr/local/cuda

it seems that the reason is the :/usr/local/cuda been regard as ':/usr/local/cuda' but not report in error log...

from pvcnn.