Win10下CUDA/C++的混合编译——DepthAwareCNN中depthconv模块

chwei0130

已于 2022-06-17 13:27:40 修改

阅读量371

点赞数

文章标签： c++ pytorch 深度学习

于 2022-06-14 16:06:55 首次发布

本文链接：https://blog.csdn.net/chwei0130/article/details/125165281

版权

本文详细记录了在Windows 10系统中，使用CUDA和C++编译深度学习模块DepthAwareCNN的过程，包括源码修改、nvcc和cl编译命令、链接步骤以及解决函数参数类型不匹配的问题。通过设置setup.py实现了一次性编译，并提供了VS+CUDA的一次性编译方案。

摘要由CSDN通过智能技术生成

源码：GitHub - laughtervv/DepthAwareCNN: Depth-aware CNN for RGB-D Segmentation, ECCV 2018

环境：Win10，python 3.7，torch 1.7.1+cu110

depthconv_cuda文件的debug

//nDimension全部替换为dim()
//size[x]全部替换为size(x)
//THCState_getCurrentStream(state)都替换为c10::cuda::getCurrentCUDAStream()
//THCudaBlas_Sgemv替换为at::cuda::blas::gemv<double>，加入头文件#include <ATen/cuda/CUDABlas.h> 该方法能够编译，但是无法链接
//THCudaBlas_xxxxx方法定义于D:\ProgramTensor\Anaconda3\envs\pytorch\Lib\site-packages\torch\include\THC\THCblas.h
//最终使用cublasSgemv，加入头文件#include <ATen/cuda/CUDABlas.h>
//参考DCNv2  https://github.com/lbin/DCNv2/tree/pytorch_1.6    https://github.com/lbin/DCNv2/

//链接错误：将 extern THCState *state; 改成THCState *state = at::globalContext().lazyInitCUDA();

使用nvcc命令编译depthconv_cuda_kernel.cu
具体指令参考make.sh

使用cl命令编译depthconv.cpp和depthconv_cuda.cpp

depthconv.c和depthconv_cuda.c修改后缀为.cpp

cl -DMS_WIN64 -I "C:\Program Files (x86)\Windows Kits\10\Include\10.0.17763.0\shared" -I"C:\Program Files (x86)\Windows Kits\10\Include\10.0.17763.0\ucrt" -I"D:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\include" -ID:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include -I"D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include\torch\csrc\api\include" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include\TH" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include\THC" -I"D:\ProgramTensor\NVIDIA\CUDA\v10.0\include" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\include" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\include"  -IE:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\src -c E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\src\depthconv.c /FoE:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\temp.win-amd64-3.7\Release\src\depthconv.o -Zi -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=depthconv -D_GLIBCXX_USE_CXX11_ABI=0

cl -DMS_WIN64 -I "C:\Program Files (x86)\Windows Kits\10\Include\10.0.17763.0\shared" -I"C:\Program Files (x86)\Windows Kits\10\Include\10.0.17763.0\ucrt" -I"D:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\include" -ID:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include -I"D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include\torch\csrc\api\include" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include\TH" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\include\THC" -I"D:\ProgramTensor\NVIDIA\CUDA\v10.0\include" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\include" -I"D:\ProgramTensor\Anaconda3\envs\pytorch\include"  -IE:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\src -c E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\src\depthconv_cuda.cpp /FoE:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\temp.win-amd64-3.7\Release\src\depthconv_cuda.o -Zi -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=depthconv -D_GLIBCXX_USE_CXX11_ABI=0

添加bindings.cpp，并使用cl编译

该cpp用于打包cuda/c++中的函数并导出

#pragma once
#include <pybind11/pybind11.h>

#include <THC/THC.h>
#include "depthconv.h"
#include "depthconv_cuda.h"
#include "depthconv_cuda_kernel.h"

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
  m.def("depthconv_forward", &depthconv_forward);
  m.def("depthconv_backward", &depthconv_backward);

  m.def("depthconv_forward_cuda", &depthconv_forward_cuda);
  m.def("depthconv_backward_input_cuda", &depthconv_backward_input_cuda);
  m.def("depthconv_backward_parameters_cuda", &depthconv_backward_parameters_cuda);

  //m.def("depthconv_im2col", &depthconv_im2col);
  //m.def("depthconv_col2im", &depthconv_col2im);
}

使用link命令链接

link -dll E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\temp.win-amd64-3.7\Release\src\depthconv.o E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\temp.win-amd64-3.7\Release\src\depthconv_cuda.o E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\temp.win-amd64-3.7\Release\src \depthconv_cuda_kernel.o E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\temp.win-amd64-3.7\Release\src    \bindings.o /libpath:"D:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Tools\MSVC\14.16.27023\lib\onecore\x64" /libpath:"D:\ProgramTensor\Anaconda3\libs" /libpath:"C:\Program Files (x86)\Windows Kits\10\Lib\10.0.17763.0\um\x64" /libpath:"C:\Program Files (x86)\Windows Kits\10\Lib\10.0.17763.0\ucrt\x64" -LIBPATH:D:\ProgramTensor\Anaconda3\envs\pytorch\lib\site-packages\torch\lib -LIBPATH:D:\ProgramTensor\NVIDIA\CUDA\v10.0\lib/x64 -LIBPATH:D:\ProgramTensor\Anaconda3\envs\pytorch\libs -LIBPATH:D:\ProgramTensor\Anaconda3\envs\pytorch\PCbuild\amd64 c10.lib c10_cuda.lib torch.lib torch_cpu.lib torch_cuda.lib torch_python.lib cudart.lib cublas.lib python37.lib  -OUT:E:\ZTensor\programs\DepthAwareCNN-master\models\ops\depthconv\build\lib.win-amd64-3.7\_ext\depthconv.cp37-win_amd64.pyd

以上步骤为单步编译、链接。

按上述修改程序后，其实可以直接构建setup.py，一次性编译。

具体过程请参见《Win10下CUDA/C++的混合编译——PCT中PointNet2模块》

调用

拷贝depthconv.cp37-win_amd64.pyd至depthconv\_ext目录下，即可调用。

实际调用中存在问题：函数定义中输入张量均为THCudaTensor *，但是传入的都是torch.Tensor，运行提示参数类型不支持。对此有两种解决方案：

1）改变传入参数的类型（需要高手加以指点）；

2）修改CUDA/C++源代码，替换THCudaTensor *指针为at::Tensor对象（可参考https://github.com/lbin/DCNv2/blob/pytorch_1.6/src/cuda/dcn_v2_cuda.cu）。

成功配置VS+CUDA后，可以一次性编译，参见

《Win10下CUDA/C++的混合编译——更新》