nvcuvid.h: No such file or directory

[Luyang1125]

Hi,

I got the error when compiling this project on Ubuntu 16.04: nvcuvid.h: No such file or directory. Could you let me know how to solve this issue? @tfogal

The detail logs shown on the terminal is here:

-- The C compiler identification is GNU 5.4.0
-- The CXX compiler identification is GNU 5.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found CUDA: /usr/local/cuda (found version "9.0")
-- Found PkgConfig: /usr/bin/pkg-config (found version "0.29.1")
-- Could NOT find EGL (missing: EGL_LIBRARY EGL_opengl_LIBRARY EGL_INCLUDE_DIR)
-- Could NOT find GLEW (missing: GLEW_INCLUDE_DIR GLEW_LIBRARY)
-- Configuring done
-- Generating done
-- Build files have been written to: /home/luyang/Documents/NvPipe-master/build
luyang@luyang-ubuntu:~/Documents/NvPipe-master/build$ make
[ 12%] Building NVCC (Device) object CMakeFiles/nvpipe.dir/convert.o
Scanning dependencies of target nvpipe
[ 25%] Building C object CMakeFiles/nvpipe.dir/debug.c.o
[ 37%] Building C object CMakeFiles/nvpipe.dir/decode.c.o
/home/luyang/Documents/NvPipe-master/decode.c:44:21: fatal error: nvcuvid.h: No such file or directory
compilation terminated.
CMakeFiles/nvpipe.dir/build.make:93: recipe for target 'CMakeFiles/nvpipe.dir/decode.c.o' failed
make[2]: *** [CMakeFiles/nvpipe.dir/decode.c.o] Error 1
CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/nvpipe.dir/all' failed
make[1]: *** [CMakeFiles/nvpipe.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2

[tfogal]

Hi Luyang,

I have heard that CUDA 9 changed the setup of the SDK, as mentioned in issue #11. If you could confirm that the header is not present on your system, and furthermore that it is installed as part of a CUDA 8 SDK, that would clarify.

Of course, patches would be welcome too!

[renewagner]

I can confirm that nvcuvid.h used to be shipped in CUDA < 9.0 but is gone now. Starting with CUDA 9.0 there's a dynlink_nvcuvid.h instead which appears to be API compatible. I've created a pull request to use this depending on the CUDA version. I cannot test decoding, unfortunately, but encoding still works fine with either CUDA 8.0 or 9.1 on a Quadro K2000M and GTX 1080.

[sasmaster]

Hi! I have a related question. I have downloaded the latest video SDK. And for NVDEC there is no dynlink_nvcuvid anymore but just nvcuvid. And there is a also a static lib for Windows only. What does that mean? How do I link on Linux? Do the latest drivers for linux come with static lib for nvcuvid?

[tbiedert]

I've just pushed a redesigned NvPipe which works with CUDA 9 and is based on the reusable NvCodec classes in the Video Codec SDK. It uses the nvcuvid.h header and compile-time linking against libnvcuvid.
Please refer to the README for its features and usage.

[jtk1919]

I cloned this ffmpeg_libyami repo and copied the 2 cuda headers directly under /usr/include/

https://github.com/intel/ffmpeg_libyami/tree/master/compat/cuda

And then I installed all the ffmpeg -dev libraries with apt-get.

The hack worked. The compilation found /usr/include/nvcuid.h and proceeded the compination through to the end.

[tbiedert]

Thanks @jtk19 ! Which version of NvPipe are you building?

The latest commit including Video Codec SDK 9 should contain everything you need, including the headers you mentioned. That include directory is also considered in CMake.