win10 gtx1660ti 配置vs opencv cuda加速

环境：

操作系统：Windows 10
显卡：GeForce GTX1660 Ti

第一步：安装显卡驱动：461.40-desktop-win10-64bit-international-nsd-dch-whql.exe

下载地址： https://cn.download.nvidia.com/Windows/461.40/461.40-notebook-win10-64bit-international-nsd-dch-whql.exe

第二步：安装Visual Studio：2015

下载地址：https://jingyan.baidu.com/article/c45ad29c223421051753e23a.html

第三步：安装CUDA： cuda_10.0.130_411.31_win10

下载地址：https://developer.nvidia.com/cuda-10.0-download-archive
添加环境变量：

 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin
 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\libnvvp
 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\lib\x64
 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\extras\CUPTI\libx64

在cmd命令窗口中输入：nvcc -V，查看cuda是否安装成功：

下载Cudnn: cudnn-10.0-windows10-x64-v7.6.1.34.zip

下载地址：https://developer.nvidia.com/rdp/cudnn-download

解压cudnn压缩包，将Cudnn包中所带的文件放到Cuda目录(上面的CUDA_PATH的路径)下对应的文件夹即可
（cuda默认安装路径为C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0）

cuda测试：
cmd到C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\extras\demo_suite目录下

执行bandwidthTest.exe

执行deviceQuery.exe

两次结果均为pass即为安装成功，记住显卡算力为7.5。

第四步：安装TBB：tbb2018_20171205oss

下载地址：https://github.com/oneapi-src/oneTBB/releases

解压好以后，添加环境变量：

D:\Program Files\tbb2018_20171205oss\bin\intel64\vc14

第五步：安装cmake: cmake-3.4.3-win32-x86.exe

下载地址：https://cmake.org/download/

添加环境变量：D:\Program Files (x86)\CMake\bin

改完之后重启一次

第六步：下载OpenCV：opencv-3.0.0.exe

下载地址：https://opencv.org/releases.html
下载后解压到D:\opencv300cuda

OpenCV-contrib：opencv_contrib-3.0.0.tar.gz

下载地址：https://github.com/opencv/opencv_contrib/releases
下载后解压到D:\opencv300cuda

第七步：cmake编译设置：

Configure后选择Visual Studio 14 2015 Win64

Cmake中红色的部分说明还没有更新，需要点击Configure更新，每次更改完一些配置或者参数之后都要点击Configure更新一下。最后Configure后没有输出红色信息后，确定配置无误再点Generate，生成项目文件。

第八步： VS编译生成库文件（进行编译前先阅读相关问题，少走弯路，减少编译时间）

打开D:\opencv300cuda\opencv\cudabuild100下的OpenCV.sln
编译，找到“ALL_BUILD”，右键->“生成”，然后开始漫长的等待……
编译输出没有错误，失败0个后，找到“CmakeTargets”下的“INSTALL”，右键->"仅用于项目“->“仅生成INSTALL”。

相关问题：

1. 取消勾选BUILD_EXAMPLES、BUILD_PERF_TESTS、BUILD_TBB、WITH_MATLAB、BUILD_opencv_world、BUILD_opencv_face，减少编译时间
   勾选WITH_CUDA、WITH_TBB

2. 特别注意，Cmake的版本一定要用3.4.3，不然会出现很多问题，比如报错：

无法打开输入文件“…\lib\Release\opencv_bioinspired300.lib”

3. “ALL_BUILD”前一定要先单独编译opencv_core、opencv_bioinspired、opencv_cudaarithm、opencv_cudabgsegm以及opencv_cudalegacy几个容易出错的项目试试，编译方法，在项目上 右键->生成。单独编译这几个项目没有报错后再进行“ALL_BUILD”。

4. 编译过程中有个别几个项目生成失败，不必担心，可以到输出目录下的modules文件夹下找到生成失败的模块，打开对应的.sln解决方案，进行Build、Install。

5. 用VS打开OpenCV.sln工程，编译生成Debug库时，VS报错：

无法打开文件\lib\Debug\opencv_bioinspired300d.lib

参考链接：https://blog.csdn.net/akadiao/article/details/78975786

解决方法：在VS中打开位于路径“……\ opencv_contrib-3.1.0\ modules\bioinspired\ src\ opencl ” 下的文件retina_kernel.cl：
把所有的注释的==//*****************************==删掉。然后重新使用cmake 配置生成。

6. 编译opencv_cudalegacy项目时报错：

error C2061: 语法错误: 标识符“NppiGraphcutState”

参考链接：https://blog.csdn.net/hollisjoe/article/details/80063938

解决方案：找到…\sources\modules\cudalegacy\src目录下的graphcuts.cpp文件，将

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)

改为

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION>= 8000)

7. 编译VS项目时报错：

error C2382: “std::tuplecv::Size,perf:`anonymous-namespace’::MatDepth,perf::`anonymous-namespace’::MatCn::operator
=”: 重定义

解决办法：在cmake编译时取消勾选BUILD_PERF_TESTS。

8. cmake编译时输出红色提示信息：

CMake Deprecation Warning at CMakeLists.txt:69 (cmake_policy): The
OLD behavior for policy CMP0022 will be removed from a future version
of CMake.

The cmake-policies(7) manual explains that the OLD behaviors of all
policies are deprecated and that a policy should be set to OLD only
under specific short-term circumstances. Projects should be ported
to the NEW behavior and not rely on setting a policy to OLD.

CMake Deprecation Warning at CMakeLists.txt:74 (cmake_policy): The
OLD behavior for policy CMP0026 will be removed from a future version
of CMake.

The cmake-policies(7) manual explains that the OLD behaviors of all
policies are deprecated and that a policy should be set to OLD only
under specific short-term circumstances. Projects should be ported
to the NEW behavior and not rely on setting a policy to OLD.

cmake版本太高，将cmake版本换为3.4.3。

9. cmake编译报错：

CMake Error: The following variables are used in this project, but
they are set to NOTFOUND. Please set them or make sure they are set
and tested correctly in the CMake files: CUDA_nppi_LIBRARY (ADVANCED)

参考链接：https://blog.csdn.net/u014613745/article/details/78310916#reply

解决方案如下：

1）.找到FindCUDA.cmake文件，找到行

find_cuda_helper_libs(nppi)

改为

 find_cuda_helper_libs(nppial)
  find_cuda_helper_libs(nppicc)
  find_cuda_helper_libs(nppicom)
  find_cuda_helper_libs(nppidei)
  find_cuda_helper_libs(nppif)
  find_cuda_helper_libs(nppig)
  find_cuda_helper_libs(nppim)
  find_cuda_helper_libs(nppist)
  find_cuda_helper_libs(nppisu)
  find_cuda_helper_libs(nppitc)

2）.找到行

set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}")

改为

set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppial_LIBRARY};${CUDA_nppicc_LIBRARY};${CUDA_nppicom_LIBRARY};${CUDA_nppidei_LIBRARY};${CUDA_nppif_LIBRARY};${CUDA_nppig_LIBRARY};${CUDA_nppim_LIBRARY};${CUDA_nppist_LIBRARY};${CUDA_nppisu_LIBRARY};${CUDA_nppitc_LIBRARY};${CUDA_npps_LIBRARY}")

3）.找到行

unset(CUDA_nppi_LIBRARY CACHE)

改为

unset(CUDA_nppial_LIBRARY CACHE)
unset(CUDA_nppicc_LIBRARY CACHE)
unset(CUDA_nppicom_LIBRARY CACHE)
unset(CUDA_nppidei_LIBRARY CACHE)
unset(CUDA_nppif_LIBRARY CACHE)
unset(CUDA_nppig_LIBRARY CACHE)
unset(CUDA_nppim_LIBRARY CACHE)
unset(CUDA_nppist_LIBRARY CACHE)
unset(CUDA_nppisu_LIBRARY CACHE)
unset(CUDA_nppitc_LIBRARY CACHE)

4）.找到文件 OpenCVDetectCUDA.cmake，修改以下几行

  set(__cuda_arch_ptx "")
  if(CUDA_GENERATION STREQUAL "Fermi")
    set(__cuda_arch_bin "2.0")
  elseif(CUDA_GENERATION STREQUAL "Kepler")
    set(__cuda_arch_bin "3.0 3.5 3.7")

改为

  set(__cuda_arch_ptx "")
  if(CUDA_GENERATION STREQUAL "Kepler")
    set(__cuda_arch_bin "3.0 3.5 3.7")
  elseif(CUDA_GENERATION STREQUAL "Maxwell")
    set(__cuda_arch_bin "5.0 5.2")

5）.cuda9中有一个单独的halffloat(cuda_fp16.h)头文件,也应该被包括在opencv的目录里，将头文件cuda_fp16.h添加至 …\opencv\modules\cudev\include\opencv2\cudev\common.hpp
即在common.hpp中添加

#include 

重新生成即可

10. cmake编译报错：

error MSB6006: “cmd.exe”已退出,代码为 1。

参考链接：https://blog.csdn.net/foso1994/article/details/96307491

解决办法：将cmake中CUDA_HOST_COMPILER设置为

D:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin\cl.exe

11. VS报错：

unsupported gpu architecture‘compute_20’

参考链接：https://blog.csdn.net/jialuo0238/article/details/88574113

解决办法：cmake中将CUDA_ARCH_BIN 和CUDA_ARCH_PTX对应的小于3的值全部删掉。

第九步：配置OpenCV环境

1.编译完成之后，在目标文件夹中多了一个install文件夹，里边就有配置OpenCV的Debug和Release版本需要的各种文件，可以将其拷贝至自定义目录下，我的拷贝路径为D:\opencv300cuda。
添加环境变量：D:\opencv300cuda\x64\vc14\bin，之后重启电脑。

2.VS新建项目，配置属性：

1）VC++目录->包含目录 添加：

D:\opencv300cuda\include
D:\opencv300cuda\include\opencv
D:\opencv300cuda\include\opencv2

2)VC++目录->库目录 添加：

D:\opencv300cuda\x64\vc14\lib

链接器->输入->附加依赖项 添加:

debug版本：

opencv_bgsegm300d.lib
opencv_bioinspired300d.lib
opencv_calib3d300d.lib
opencv_ccalib300d.lib
opencv_core300d.lib
opencv_cudaarithm300d.lib
opencv_cudabgsegm300d.lib
opencv_cudacodec300d.lib
opencv_cudafeatures2d300d.lib
opencv_cudafilters300d.lib
opencv_cudaimgproc300d.lib
opencv_cudalegacy300d.lib
opencv_cudaobjdetect300d.lib
opencv_cudaoptflow300d.lib
opencv_cudastereo300d.lib
opencv_cudawarping300d.lib
opencv_cudev300d.lib
opencv_features2d300d.lib
opencv_flann300d.lib
opencv_hal300d.lib
opencv_highgui300d.lib
opencv_imgcodecs300d.lib
opencv_imgproc300d.lib
opencv_latentsvm300d.lib
opencv_line_descriptor300d.lib
opencv_ml300d.lib
opencv_objdetect300d.lib
opencv_optflow300d.lib
opencv_photo300d.lib
opencv_reg300d.lib
opencv_rgbd300d.lib
opencv_saliency300d.lib
opencv_shape300d.lib
opencv_superres300d.lib
opencv_surface_matching300d.lib
opencv_text300d.lib
opencv_tracking300d.lib
opencv_ts300d.lib
opencv_video300d.lib
opencv_videoio300d.lib
opencv_videostab300d.lib
opencv_xfeatures2d300d.lib
opencv_ximgproc300d.lib
opencv_xobjdetect300d.lib
opencv_xphoto300d.lib

release版本：

opencv_bgsegm300.lib
opencv_bioinspired300.lib
opencv_calib3d300.lib
opencv_ccalib300.lib
opencv_core300.lib
opencv_cudaarithm300.lib
opencv_cudabgsegm300.lib
opencv_cudacodec300.lib
opencv_cudafeatures2d300.lib
opencv_cudafilters300.lib
opencv_cudaimgproc300.lib
opencv_cudalegacy300.lib
opencv_cudaobjdetect300.lib
opencv_cudaoptflow300.lib
opencv_cudastereo300.lib
opencv_cudawarping300.lib
opencv_cudev300.lib
opencv_features2d300.lib
opencv_flann300.lib
opencv_hal300.lib
opencv_highgui300.lib
opencv_imgcodecs300.lib
opencv_imgproc300.lib
opencv_latentsvm300.lib
opencv_line_descriptor300.lib
opencv_ml300.lib
opencv_objdetect300.lib
opencv_optflow300.lib
opencv_photo300.lib
opencv_reg300.lib
opencv_rgbd300.lib
opencv_saliency300.lib
opencv_shape300.lib
opencv_stitching300.lib
opencv_superres300.lib
opencv_surface_matching300.lib
opencv_text300.lib
opencv_tracking300.lib
opencv_ts300.lib
opencv_video300.lib
opencv_videoio300.lib
opencv_videostab300.lib
opencv_xfeatures2d300.lib
opencv_ximgproc300.lib
opencv_xobjdetect300.lib
opencv_xphoto300.lib

3.将D:\opencv300cuda\x64\vc14\bin下所有的.dll文件复制到C:\Windows\System32和C:\Windows\SysWOW64下。

第十步：环境测试

程序1（m1.cpp）：

#include "opencv2/opencv.hpp"
//#include "opencv2/core.hpp"
//#include "opencv2/highgui.hpp"
//#include "opencv2/videoio.hpp"
//#include "opencv2/core/cuda.hpp"

#include
#include
#include
#include
#include

#include
#include "opencv2/cudabgsegm.hpp"

#include

using namespace std;
using namespace cv;
using namespace cv::cuda;


int main()
{
	/*int num_devices = cv::cuda::getCudaEnabledDeviceCount();
	if (num_devices <= 0) {
	cerr << "There is no device." << endl;
	return -1;
	}

	int enable_device_id = -1;
	for (int i = 0; i < num_devices; i++) {
	cv::cuda::DeviceInfo dev_info(i);
	if (dev_info.isCompatible()) {
	enable_device_id = i;
	}
	}
	if (enable_device_id < 0) {
	cerr << "GPU module isn't built for GPU" << endl;
	}
	cv::cuda::setDevice(enable_device_id);*/

	cuda::printCudaDeviceInfo(cuda::getDevice());
	int count = cuda::getCudaEnabledDeviceCount();
	printf("GPU Device Number:%d\n", count);
	printf("**************************************************\n");


	/*-------------------------以下四种验证方式任意选取一种即可-------------------------*/
	//获取显卡简单信息
	cuda::printShortCudaDeviceInfo(cuda::getDevice());  //有显卡信息表示GPU模块配置成功
	cuda::printCudaDeviceInfo(cuda::getDevice());  //有显卡信息表示GPU模块配置成功
												   
	int Device_Num = cuda::getCudaEnabledDeviceCount();//获取显卡设备数量
	cout << Device_Num << endl;  //返回值大于0表示GPU模块配置成功
	 
	cuda::DeviceInfo Device_State;//获取显卡设备状态
	bool Device_OK = Device_State.isCompatible();
	cout << "Device_State: " << Device_OK << endl;  //返回值大于0表示GPU模块配置成功


	waitKey();
	return 0;
}

***** VIDEOINPUT LIBRARY - 0.1995 - TFW07 *****

*** CUDA Device Query (Runtime API) version (CUDART static linking) ***

Device count: 1

Device 0: "GeForce GTX 1660 Ti"
  CUDA Driver Version / Runtime Version          11.20 / 10.0
  CUDA Capability Major/Minor version number:    7.5
  Total amount of global memory:                 6144 MBytes (6442450944 bytes)
  GPU Clock Speed:                               1.59 GHz
  Max Texture Dimension Size (x,y,z)             1D=(131072), 2D=(131072,65536), 3D=(16384,16384,16384)
  Max Layered Texture Size (dim) x layers        1D=(32768) x 2048, 2D=(32768,32768) x 2048
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per block:           1024
  Maximum sizes of each dimension of a block:    1024 x 1024 x 64
  Maximum sizes of each dimension of a grid:     2147483647 x 65535 x 65535
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and execution:                 Yes with 6 copy engine(s)
  Run time limit on kernels:                     Yes
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Concurrent kernel execution:                   Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support enabled:                No
  Device is using TCC driver mode:               No
  Device supports Unified Addressing (UVA):      Yes
  Device PCI Bus ID / PCI location ID:           1 / 0
  Compute Mode:
      Default (multiple host threads can use ::cudaSetDevice() with device simultaneously)

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version  = 11.20, CUDA Runtime Version = 10.0, NumDevs = 1

GPU Device Number:1
**************************************************
Device 0:  "GeForce GTX 1660 Ti"  6144Mb, sm_75, Driver/Runtime ver.11.20/10.0
*** CUDA Device Query (Runtime API) version (CUDART static linking) ***

Device count: 1

Device 0: "GeForce GTX 1660 Ti"
  CUDA Driver Version / Runtime Version          11.20 / 10.0
  CUDA Capability Major/Minor version number:    7.5
  Total amount of global memory:                 6144 MBytes (6442450944 bytes)
  GPU Clock Speed:                               1.59 GHz
  Max Texture Dimension Size (x,y,z)             1D=(131072), 2D=(131072,65536), 3D=(16384,16384,16384)
  Max Layered Texture Size (dim) x layers        1D=(32768) x 2048, 2D=(32768,32768) x 2048
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per block:           1024
  Maximum sizes of each dimension of a block:    1024 x 1024 x 64
  Maximum sizes of each dimension of a grid:     2147483647 x 65535 x 65535
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and execution:                 Yes with 6 copy engine(s)
  Run time limit on kernels:                     Yes
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Concurrent kernel execution:                   Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support enabled:                No
  Device is using TCC driver mode:               No
  Device supports Unified Addressing (UVA):      Yes
  Device PCI Bus ID / PCI location ID:           1 / 0
  Compute Mode:
      Default (multiple host threads can use ::cudaSetDevice() with device simultaneously)

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version  = 11.20, CUDA Runtime Version = 10.0, NumDevs = 1

1
Device_State: 1

程序2（m2.cpp）:

#include "opencv2/opencv.hpp"
//#include "opencv2/core.hpp"
//#include "opencv2/highgui.hpp"
//#include "opencv2/videoio.hpp"
//#include "opencv2/core/cuda.hpp"

#include
#include
#include
#include
#include

#include
#include "opencv2/cudabgsegm.hpp"

#include

using namespace std;
using namespace cv;
using namespace cv::cuda;


int main()
{
	Mat src_image = imread("E:/finished202102/picture/test.png");
	imshow("image", src_image);
	Mat dst_image, dst_imge1;

	vector<Mat> channels;//定义存储的容器
	split(src_image, channels);
	Mat src_image0 = channels[0];//b通道的图像

	cuda::GpuMat d_src_img(src_image);//upload src image to gpu
	cuda::GpuMat d_dst_img, d_src_img00;
	cuda::cvtColor(d_src_img, d_dst_img, CV_BGR2GRAY);
	d_dst_img.download(dst_image);

	cuda::GpuMat d_src_img0(src_image0);
	cuda::equalizeHist(d_src_img0, d_src_img00);
	Mat hist_image;
	d_src_img00.download(hist_image);
	imshow("gray", dst_image);
	imshow("hist", hist_image);
	
	waitKey();
	return 0;
}

程序3（m3.cpp）:

#include 

#include "opencv2/opencv_modules.hpp"
#include "opencv2/core.hpp"
#include "opencv2/features2d.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/cudafeatures2d.hpp"
#include "opencv2/xfeatures2d/cuda.hpp"

#include "opencv2/opencv.hpp"
#include "opencv2/core.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/videoio.hpp"
#include "opencv2/core/cuda.hpp"
#include
#include
#include
#include
#include


#include
#include "opencv2/cudabgsegm.hpp"


using namespace std;
using namespace cv;
using namespace cv::cuda;

int main()
{
	cuda::printShortCudaDeviceInfo(cuda::getDevice());

	GpuMat img1, img2;
	img1.upload(imread("1.jpg", IMREAD_GRAYSCALE));
	img2.upload(imread("2.jpg", IMREAD_GRAYSCALE));

	SURF_CUDA surf;

	// detecting keypoints & computing descriptors
	GpuMat keypoints1GPU, keypoints2GPU;
	GpuMat descriptors1GPU, descriptors2GPU;
	surf(img1, GpuMat(), keypoints1GPU, descriptors1GPU);
	surf(img2, GpuMat(), keypoints2GPU, descriptors2GPU);

	cout << "FOUND " << keypoints1GPU.cols << " keypoints on first image" << endl;
	cout << "FOUND " << keypoints2GPU.cols << " keypoints on second image" << endl;

	// matching descriptors
	Ptr<cuda::DescriptorMatcher> matcher = cuda::DescriptorMatcher::createBFMatcher(surf.defaultNorm());
	vector<DMatch> matches;
	matcher->match(descriptors1GPU, descriptors2GPU, matches);

	// downloading results
	vector<KeyPoint> keypoints1, keypoints2;
	vector<float> descriptors1, descriptors2;
	surf.downloadKeypoints(keypoints1GPU, keypoints1);
	surf.downloadKeypoints(keypoints2GPU, keypoints2);
	surf.downloadDescriptors(descriptors1GPU, descriptors1);
	surf.downloadDescriptors(descriptors2GPU, descriptors2);

	// drawing the results
	Mat img_matches;
	drawMatches(Mat(img1), keypoints1, Mat(img2), keypoints2, matches, img_matches);

	namedWindow("matches", 0);
	imshow("matches", img_matches);
	waitKey(0);
	return 0;
}

***** VIDEOINPUT LIBRARY - 0.1995 - TFW07 *****

Device 0:  "GeForce GTX 1660 Ti"  6144Mb, sm_75, Driver/Runtime ver.11.20/10.0
FOUND 643 keypoints on first image
FOUND 1161 keypoints on second image

参考链接：

WIN10 + OpenCV3.4 + CUDA8.0 + Cmake3.9.0 + TBB + VS2015配置/重编译：
https://blog.csdn.net/gggttt222/article/details/79482033

Win10 下Cmake编译配置 Opencv3.1 + Cuda7.5 + VS2013：
https://www.cnblogs.com/asmer-stone/p/5530868.html