【记录一个问题】linux + opencv + gpu视频解码，好不容易编译通过，运行又coredump了

1.首先编译了opencv + cuda
   编译选项中使用了以下关于cuvid库的内容：

//"nvcuvid" library
CUDA_nvcuvid_LIBRARY:FILEPATH=/usr/local/lib/libnvcuvid.so

//Path to a library.
CUDA_rt_LIBRARY:FILEPATH=/usr/local/cuda-9.0/lib64/libcudart_static.a

//Include NVidia Video Decoding library support
WITH_NVCUVID:BOOL=ON

2.下载了这个opencv + cuvid视频解码例子：
https://github.com/opencv/opencv/blob/master/samples/gpu/video_reader.cpp

编译和链接命令行：

g++ -c video_reader.cpp -o video_reader.o -g -I"/home/admin/opencv/2019-05-09_cuda/include/opencv4/" -std=c++11

g++ -o video_reader video_reader.o -L"/home/admin/opencv/2019-05-09_cuda/lib64/" -lopencv_core -lopencv_videoio -lopencv_cudacodec -lopencv_imgproc -lopencv_imgcodecs -lopencv_flann  -lopencv_img_hash  -lopencv_photo -L"/usr/local/cuda/lib64/stubs" -lcuda -lcublas -L"/usr/local/cuda/lib64" -lcudart -L"/usr/local/lib/" -lnvcuvid -lz -ldl -lpthread -lopencv_imgcodecs -lopencv_core -lopencv_imgcodecs -L"/home/admin/opencv/2019-05-09_cuda/lib64/opencv4/3rdparty" -lIlmImf -lopencv_core -lopencv_imgproc

3. 运行程序，出现coredump，GPU驱动的版本是 384.81

./video_reader stand_input.mp4

Program received signal SIGSEGV, Segmentation fault.

堆栈信息如下：

(gdb) bt
#0  0x00000000007eb4aa in cv::cudacodec::detail::CuvidVideoSource::CuvidVideoSource (
    this=0x1731810, fname=...)
    at /home/admin/opencv/opencv_contrib-master/modules/cudacodec/src/cuvid_video_source.cpp:63
#1  0x00000000007e6943 in cv::cudacodec::createVideoReader (filename=...)
    at /home/admin/opencv/opencv_contrib-master/modules/cudacodec/src/video_reader.cpp:206
#2  0x000000000040e0cb in main (argc=2, argv=0x7fffffffe5a8) at video_reader.cpp:33
(gdb) f 0
#0  0x00000000007eb4aa in cv::cudacodec::detail::CuvidVideoSource::CuvidVideoSource (
    this=0x1731810, fname=...)
    at /home/admin/opencv/opencv_contrib-master/modules/cudacodec/src/cuvid_video_source.cpp:63
63          CUresult cuRes = cuvidCreateVideoSource(&videoSource_, fname.c_str(), &params);

4. 搜索了一下，这篇帖子貌似提供了解决办法：
https://github.com/opencv/opencv/issues/10201

'Segmentation fault' with gpu video decoding #10201

下一步尝试自己下载一个 Video_Codec_SDK 库来替换系统自带的版本。

====================

2019-05-12补充：

按照以上帖子的提示，下载 Video_Codec_SDK_8.0.14.zip

unzip Video_Codec_SDK_8.0.14.zip

/home/admin/cuvid/Video_Codec_SDK_8.0.14/Samples

make

# make 出现错误，没关系， dynlink_nvcuvid.o生成就好了

代码的头文件中include Video_Codec_SDK_8.0.14/Samples/common/inc/dynlink_nvcuvid.h

然后代码中加上：cuda_res = cuvidInit(0);

最后，链接的路径中加上： "Video_Codec_SDK_8.0.14/Samples/NvDecodeGL/dynlink_nvcuvid.o" -L"/usr/local/nvidia/lib64/" -lnvcuvid

运行成功

然后又遇到了GpuMat的格式问题，请看下一篇！