spconv简介、环境配置与安装以及遇到的各种报错处理

此篇博客将介绍spconv、如何安装spconv 1.x和spconv2.x，以及遇到的各种报错处理，更多详细内容可以参考官方的GitHub链接https://github.com/traveller59/spconv。

spconv介绍

spconv为空间稀疏卷积库，一个提供高度优化的稀疏卷积实现和张量核心支持的项目。spconv的设计旨在有效地处理包含大量零元素的稀疏数据，主要用于3D点云的卷积操作。

spconv安装

博主自己的硬件配置如下：

• Linux（Ubuntu 20.04）
• NVIDIA GeForce RTX 3090
• NVIDIA显卡驱动版本：12.0
• CUDA version：11.3

首先安装pytorch、cuda等相关软件包：

conda create --name env_name python=3.7 cmake=3.22.1
conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge
conda install cudnn -c conda-forge
conda install boost

打开bashrc文件：vim ~/.bashrc，设置环境变量：

export PATH=/usr/local/cuda/bin:$PATH
export CUDA_HOME=/usr/local/cuda/bin:$CUDA_HOME
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export C_INCLUDE_PATH=$C_INCLUDE_PATH:/[$YourUserName]/anaconda3/envs/[$YourEnvName]/include
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/[$YourUserName]/anaconda3/envs/[$YourEnvName]/include
export LIBRARY_PATH=$LIBRARY_PATH:/[$YourUserName]/anaconda3/envs/[$YourEnvName]/lib
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/[$YourUserName]/anaconda3/envs/[$YourEnvName]/lib

源码安装spconv 1.x

spconv 1.x官方已经舍弃掉不维护了，推荐使用spconv 2.x。而且spconv 2.x相较于spconv 1.x做了优化，速度、效率等有所提升。若由于版本需要，不得不需要安装spconv 1.x，则需要源码编译安装。
下面是官方给出的安装spconv 1.2.1的步骤：

# STEP 1: get source code
git clone https://github.com/traveller59/spconv.git 
cd spconv
git checkout v1.2.1
git submodule update --init --recursive 

# STEP 2: compile
python setup.py bdist_wheel

# STEP 3: install
cd ./dist
pip install spconv-1.2.1-cp37-cp37m-linux_x86_64.whl

# check if is successfully installed
python 
import spconv

pip安装spconv 2.x

spconv 2.x目前已经支持通过pip install的方式进行安装了，可以根据自己的环境进行选择运行以下命令：

# 预先安装cumm和timm
pip install cumm timm

# cpu
pip install spconv
# cuda 10.2
pip install spconv-cu102
# cuda 11.3
pip install spconv-cu113
# cuda 11.4
pip install spconv-cu114
# cuda 11.6
pip install spconv-cu116
# cuda 11.7
pip install spconv-cu117
# cuda 11.8
pip install spconv-cu118
# cuda 12.0
pip install spconv-cu120

报错处理

在此主要列举了在安装spconv 1.x的报错和解决方法。

报错一

error This file requires compiler and library support for the ISO C++ 2011 standard. This support is currently experimental, and must be enabled with the -std=c++11 or -std=gnu++11 compiler options.

输入gcc --version获取gcc版本是4.8.5，解决办法是安装5.4.0版本的gcc。在此，我使用的是在conda虚拟环境中安装：

conda install https://anaconda.org/brown-data-science/gcc/5.4.0/download/linux-64/gcc-5.4.0-0.tar.bz2

这个package有一个坑在于，在miniconda3/envs/env_name/lib文件下的libstdc++.so和libstdc++.so.6软链接指向的是libstdc++.so.6.0.21动态链接库，并且这个软链接会覆盖环境内原有的软链接，也就是说，之后使用的都会是这个package所提供的动态链接库。

这就会导致如果你的pytorch版本较高时，由于libstdc++.so.6.0.21版本较旧，所以会报错：

ImportError: /xxx/miniconda3/envs/env_name/lib/libstdc++.so.6: version `GLIBCXX_3.4.22' not found (required by /xxx/miniconda3/envs/env_name/lib/python3.7/site-packages/open3d/open3d.cpython-37m-x86_64-linux-gnu.so)

解决方法是，进入miniconda3/envs/env_name/lib，然后将软链接libstdc++.so和libstdc++.so.6指向高版本的动态链接库，例如同目录下的libstdc++.so.6.0.32，命令如下：

# 先删除原来的软连接
rm libstdc++.so
rm libstdc++.so.6
# 建立新的软连接
ln -s libstdc++.so.6.0.32 libstdc++.so
ln -s libstdc++.so.6.0.32 libstdc++.so.6

参考链接：https://blog.csdn.net/j___t/article/details/107308883

报错二

/xxx/spconv/include/tensorview/tensorview.h:741:23: error: template declaration of ‘constexpr const char* const tv::type_s’
 constexpr const char *type_s = detail::TypeToString<T>::value;
                       ^
/xxx/spconv/include/tensorview/tensorview.h: In member function ‘std::string tv::TensorView<T, Rank, PtrTraits, Tindex>::repr(Os&) const’:
/xxx/spconv/include/tensorview/tensorview.h:1124:26: error: ‘type_s’ was not declared in this scope
       ss << "Tensor[" << type_s<T> << "]" << std::endl;
                          ^
/xxx/spconv/include/tensorview/tensorview.h:1124:34: error: expected primary-expression before ‘>’ token
       ss << "Tensor[" << type_s<T> << "]" << std::endl;
                                  ^
/xxx/spconv/include/tensorview/tensorview.h:1124:36: error: expected primary-expression before ‘<<’ token
       ss << "Tensor[" << type_s<T> << "]" << std::endl;
                                    ^
/xxx/spconv/include/tensorview/tensorview.h:1135:24: error: ‘type_s’ was not declared in this scope
     ss << "Tensor[" << type_s<T> << "]: shape=" << shape()
                        ^
/xxx/spconv/include/tensorview/tensorview.h:1135:32: error: expected primary-expression before ‘>’ token
     ss << "Tensor[" << type_s<T> << "]: shape=" << shape()
                                ^
/xxx/spconv/include/tensorview/tensorview.h:1135:34: error: expected primary-expression before ‘<<’ token
     ss << "Tensor[" << type_s<T> << "]: shape=" << shape()
                                  ^
/xxx/spconv/include/tensorview/tensorview.h: At global scope:
/xxx/spconv/include/tensorview/tensorview.h:1270:23: error: template declaration of ‘constexpr const char* const tv::detail::type_printf_format_v’
 constexpr const char *type_printf_format_v = TypePrintfFormat<T>::value;
                       ^
/xxx/spconv/include/tensorview/tensorview.h: In function ‘void tv::printTensorView(tv::TensorView<T, Rank, PtrTraits, Tindex>)’:
/xxx/spconv/include/tensorview/tensorview.h:1343:34: error: ‘type_printf_format_v’ is not a member of ‘tv::detail’
   return printTensorView(tensor, detail::type_printf_format_v<Traw>);
                                  ^
/xxx/spconv/include/tensorview/tensorview.h:1343:67: error: expected primary-expression before ‘>’ token
   return printTensorView(tensor, detail::type_printf_format_v<Traw>);
                                                                   ^
/xxx/spconv/include/tensorview/tensorview.h:1343:68: error: expected primary-expression before ‘)’ token
   return printTensorView(tensor, detail::type_printf_format_v<Traw>);
                                                                    ^
/xxx/spconv/include/tensorview/tensorview.h: In function ‘void tv::printTensorView(const T*, tv::Shape)’:
/xxx/spconv/include/tensorview/tensorview.h:1349:26: error: ‘type_printf_format_v’ is not a member of ‘tv::detail’
                          detail::type_printf_format_v<Traw>);
                          ^
/xxx/spconv/include/tensorview/tensorview.h:1349:59: error: expected primary-expression before ‘>’ token
                          detail::type_printf_format_v<Traw>);
                                                           ^
/xxx/spconv/include/tensorview/tensorview.h:1349:60: error: expected primary-expression before ‘)’ token
                          detail::type_printf_format_v<Traw>);

解决方法是直接将build文件夹删掉，重新运行python setup.py bdist_wheel就解决了，如果没有解决，可以尝试一下将boost/. 复制到spconv/include/，但此方法我没有尝试，不确定是否可以解决。

报错三

以下两个报错均是由CUDA环境变量未找到或不正确导致：

CMake Error at /disk1/zhuhe/miniconda3/envs/env_name/lib/python3.7/site-packages/cmake/data/share/cmake-3.28/Modules/CMakeDetermineCUDACompiler.cmake:270 (message):
  Failed to detect a default CUDA architecture.



  Compiler output:

Call Stack (most recent call first):
  CMakeLists.txt:9 (project)

CMake Error at /xxx/miniconda3/envs/env_name/lib/python3.7/site-packages/torch/share/cmake/Caffe2/public/cuda.cmake:87 (message):
  FindCUDA says CUDA version is (usually determined by nvcc), but the CUDA
  headers say the version is ERROR: ld.so: object

  10.2.  This often occurs when you set both CUDA_HOME and
  CUDA_NVCC_EXECUTABLE to non-standard locations, without also setting PATH
  to point to the correct nvcc.  Perhaps, try re-running this command again
  with PATH=/usr/local/cuda-10.2/bin:$PATH.  See above log messages for more
  diagnostics, and see https://github.com/pytorch/pytorch/issues/8092 for
  more details.
Call Stack (most recent call first):
  /xxx/miniconda3/envs/env_name/lib/python3.7/site-packages/torch/share/cmake/Caffe2/Caffe2Config.cmake:88 (include)
  /xxx/miniconda3/envs/env_name/lib/python3.7/site-packages/torch/share/cmake/Torch/TorchConfig.cmake:68 (find_package)
  CMakeLists.txt:22 (find_package)

打开bashrc文件：vim ~/.bashrc，设置好CUDA的相关环境变量。

export PATH=/usr/local/cuda/bin:$PATH
export CUDA_HOME=/usr/local/cuda
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH