computers:ffmpeg:using_nvenc_cuda [Jon's Repository of Info]

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== ffmpeg with NVENC/CUDA install on Linux ======

Need to compile, point libraries and build dirs to local user folder for build so that it is easier to remove and clean up if you mess up instead of putting directly into /usr/local, which is default.

Nvidia's guide, is well.. semi accurate, but not verbatim.

===== Software requirements =====

==== Package Requirements ====

The Nvidia guide ((https://docs.nvidia.com/video-technologies/video-codec-sdk/ffmpeg-with-nvidia-gpu/#compiling-for-linux)) does not work with the limited amount of packages required. I referred to the official ffmpeg compile guide ((https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu#FFmpeg)) which gives you the following:

<code>
sudo apt-get update -qq && sudo apt-get -y install autoconf automake build-essential cmake git-core libass-dev libfreetype6-dev libgnutls28-dev libmp3lame-dev libsdl2-dev libtool libva-dev libvdpau-dev libvorbis-dev libxcb1-dev libxcb-shm0-dev libxcb-xfixes0-dev meson ninja-build pkg-config texinfo wget yasm zlib1g-dev
</code>

Honestly, I think just installing pkg-config from the repo will suffice, but this just helps overall to ensure ffmpeg compiles without issues.

==== Nvidia Drivers ====

https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Debian&target_version=11&target_type=deb_network

<code>
wget https://developer.download.nvidia.com/compute/cuda/repos/debian11/x86_64/cuda-keyring_1.0-1_all.deb
sudo dpkg -i cuda-keyring_1.0-1_all.deb
sudo add-apt-repository contrib
sudo apt-get update
sudo apt-get -y install cuda
</code>

=== Install linux headers ===

Per NVIDIA((https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#verify-the-system-has-the-correct-kernel-headers-and-development-packages-installed)), they want you to install headers based on uname -r. Instead, install the package from the base arch level, not specific kernel version. This way when you update, the headers update too. In my case, running Debian, it would be linux-headers-amd64

==== Cuda Headers ====

Per NVIDIA,((https://docs.nvidia.com/video-technologies/video-codec-sdk/ffmpeg-with-nvidia-gpu/)) they want you to download the Cuda Headers from VideoLan's Git page. You can't always trust it though, it may be a version off. In my last go-round, they were still on version 11, while my setup was looking for version 12. Either way, download the files from an alternative source (like FFMpeg's Git Repo https://github.com/FFmpeg/nv-codec-headers), and compile as normal.

==== Showing the driver connections ====

First things first, do an ldconfig -v to see if there are any conflicts with the shared libs

Then run an ldd of the binary you want to check to show all of the dependent shared libs. If any are missing, you will need to check on what was missed in the pre-steps. The NPP libs are part of the cuda toolkit
<code>ldd /usr/local/bin/ffmpeg ##Or wherever you compiled ffmpeg
	linux-vdso.so.1 (0x00007ffd5ed27000)
	libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f45a5902000)
	libnppig.so.11 => /lib/x86_64-linux-gnu/libnppig.so.11 (0x00007f45a3a35000)
	libnppicc.so.11 => /lib/x86_64-linux-gnu/libnppicc.so.11 (0x00007f45a3310000)
	libnppidei.so.11 => /lib/x86_64-linux-gnu/libnppidei.so.11 (0x00007f45a28b7000)
	libnppif.so.11 => /lib/x86_64-linux-gnu/libnppif.so.11 (0x00007f459ee2e000)
	libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007f459ee28000)
	libvpx.so.6 => /lib/x86_64-linux-gnu/libvpx.so.6 (0x00007f459ec0b000)
	libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007f459ebe9000)
	libdav1d.so.4 => /lib/x86_64-linux-gnu/libdav1d.so.4 (0x00007f459eab5000)
	libfdk-aac.so.2 => /lib/x86_64-linux-gnu/libfdk-aac.so.2 (0x00007f459e972000)
	libopus.so.0 => /lib/x86_64-linux-gnu/libopus.so.0 (0x00007f459e917000)
	libx264.so.160 => /lib/x86_64-linux-gnu/libx264.so.160 (0x00007f459e655000)
	libx265.so.192 => /lib/x86_64-linux-gnu/libx265.so.192 (0x00007f459d6d6000)
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f459d511000)
	/lib64/ld-linux-x86-64.so.2 (0x00007f45a7823000)
	libnppc.so.11 => /lib/x86_64-linux-gnu/libnppc.so.11 (0x00007f459d280000)
	librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007f459d275000)
	libstdc++.so.6 => /lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007f459d0a8000)
	libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f459d08e000)
	libnuma.so.1 => /lib/x86_64-linux-gnu/libnuma.so.1 (0x00007f459d07e000)
</code>
===== Install steps ===== 

I do prefer using the full guide, as there are additional steps you'll need to get everything set for ffmpeg

https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#pre-installation-actions

==== Download ffmpeg ====

Grab ffmpeg from its source

<code>git clone https://git.ffmpeg.org/ffmpeg.git ffmpeg/</code>

==== Set Paths ====

You'll want to check out the full docs ((https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#environment-setup)) for the env setups.

<code>export LD_LIBRARY_PATH=/usr/local/cuda-12.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}</code>

Don't think this is needed..
<code>PATH="$HOME/bin:$PATH" PKG_CONFIG_PATH="$HOME/ffmpeg_build/lib/pkgconfig"</code>

=== Running the ffmpeg configure script ===

This is the official way of doing things:
<code>./configure --enable-nonfree --enable-cuda-nvcc --enable-libnpp --extra-cflags=-I/usr/local/cuda/include --extra-ldflags=-L/usr/local/cuda/lib64 --disable-static --enable-shared</code>

This is a WIP, hoping to cut this down to the minimal needed for doing any files thru nvenc only. 
<code>
./configure --pkg-config-flags="--static" --enable-nonfree --enable-cuda-nvcc --enable-libnpp --extra-cflags="-I$HOME/ffmpeg_build/include" --extra-cflags=-I/usr/local/cuda/include --extra-ldflags="-L$HOME/ffmpeg_build/lib" --extra-ldflags=-L/usr/local/cuda/lib64 --extra-libs="-lpthread -lm" --ld="g++" --enable-gpl --enable-libfdk-aac --enable-libopus --enable-libdav1d --enable-libvpx --enable-libx264 --enable-libx265
</code>

==== Optional - Move files into /usr/ for global use ====

<code>cp ffmppegdir/ffmpeg /usr/local/bin/</code>

===== Resources =====

  * https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#pre-installation-actions
  * https://trac.ffmpeg.org/wiki/CompilationGuide/Generic
  * https://trac.ffmpeg.org/wiki/CompilationGuide/Ubuntu
  * https://trac.ffmpeg.org/wiki/HWAccelIntro
  * https://trac.ffmpeg.org/ticket/7782
      * This addresses where to put the codec-headers if you want to compile it yourself per the guide, and want it to get picked up by the config script.
  * https://git.ffmpeg.org/ffmpeg.git
  * https://github.com/NVIDIA/cuda-samples/issues/46
  * https://blog.roberthallam.org/2021/06/ffmpeg-error-while-loading-shared-libraries-libavdevice-so-59-cannot-open-shared-object-file-no-such-file-or-directory/