FFmpeg显卡版编译安装

本文内容包括：

在Linux环境下安装FFmpeg
通过命令行实现视频格式识别和转码
有Nvidia显卡的情况下，在Linux下使用GPU进行视频转码加速的方法

FFmpeg编译安装

在FFmpeg官网Download FFmpeg可以下载到ubunto/debian的发行包，其他Linux发行版需自行编译。同时，如果要使用GPU进行硬件加速的话，也是必须自己编译FFmpeg的，所以本节将介绍从源码编译安装FFmpeg的方法（基于RHEL/Centos）

安装依赖工具

yum install autoconf automake bzip2 cmake freetype-devel gcc gcc-c++ git libtool make mercurial pkgconfig zlib-devel

准备工作

在$HOME下创建ffmpeg_sources目录

编译并安装依赖库

本节中的依赖库基本都是必须的，建议全部安装

nasm

汇编编译器，编译某些依赖库的时候需要

cd ~/ffmpeg_sources
curl -O -L http://www.nasm.us/pub/nasm/releasebuilds/2.13.02/nasm-2.13.02.tar.bz2
tar xjvf nasm-2.13.02.tar.bz2
cd nasm-2.13.02
./autogen.sh
./configure --prefix="$HOME/ffmpeg_build" --bindir="$HOME/bin"
make
make install

yasm

汇编编译器，编译某些依赖库的时候需要

cd ~/ffmpeg_sources
curl -O -L http://www.tortall.net/projects/yasm/releases/yasm-1.3.0.tar.gz
tar xzvf yasm-1.3.0.tar.gz
cd yasm-1.3.0
./configure --prefix="$HOME/ffmpeg_build" --bindir="$HOME/bin"
make
make install

libx264

H.264视频编码器，如果需要输出H.264编码的视频就需要此库，所以可以说是必备

cd ~/ffmpeg_sources
git clone --depth 1 http://git.videolan.org/git/x264
cd x264
PKG_CONFIG_PATH="$HOME/ffmpeg_build/lib/pkgconfig" ./configure --prefix="$HOME/ffmpeg_build" --bindir="$HOME/bin" --enable-static
make
make install

libx265

H.265/HEVC视频编码器。
如果不需要此编码器，可以跳过，并在ffmpeg的configure命令中移除--enable-libx265

cd ~/ffmpeg_sources
hg clone https://bitbucket.org/multicoreware/x265
cd ~/ffmpeg_sources/x265/build/linux
cmake -G "Unix Makefiles" -DCMAKE_INSTALL_PREFIX="$HOME/ffmpeg_build" -DENABLE_SHARED:bool=off ../../source
make
make install

libfdk_acc

AAC音频编码器，必备

cd ~/ffmpeg_sources
git clone --depth 1 --branch v0.1.6 https://github.com/mstorsjo/fdk-aac.git
cd fdk-aac
autoreconf -fiv
./configure --prefix="$HOME/ffmpeg_build" --disable-shared
make
make install

libmp3lame

MP3音频编码器，必备

cd ~/ffmpeg_sources
curl -O -L http://downloads.sourceforge.net/project/lame/lame/3.100/lame-3.100.tar.gz
tar xzvf lame-3.100.tar.gz
cd lame-3.100
./configure --prefix="$HOME/ffmpeg_build" --bindir="$HOME/bin" --disable-shared --enable-nasm
make
make install

libops

OPUS音频编码器
如果不需要此编码器，可以跳过，并在ffmpeg的configure命令中移除--enable-libopus

cd ~/ffmpeg_sources
curl -O -L https://archive.mozilla.org/pub/opus/opus-1.2.1.tar.gz
tar xzvf opus-1.2.1.tar.gz
cd opus-1.2.1
./configure --prefix="$HOME/ffmpeg_build" --disable-shared
make
make install

libogg

被libvorbis依赖

cd ~/ffmpeg_sources
curl -O -L http://downloads.xiph.org/releases/ogg/libogg-1.3.3.tar.gz
tar xzvf libogg-1.3.3.tar.gz
cd libogg-1.3.3
./configure --prefix="$HOME/ffmpeg_build" --disable-shared
make
make install

libvorbis

Vorbis音频编码器
如果不需要此编码器，可以跳过，并在ffmpeg的configure命令中移除--enable-libvorbis

cd ~/ffmpeg_sources
curl -O -L http://downloads.xiph.org/releases/vorbis/libvorbis-1.3.5.tar.gz
tar xzvf libvorbis-1.3.5.tar.gz
cd libvorbis-1.3.5
./configure --prefix="$HOME/ffmpeg_build" --with-ogg="$HOME/ffmpeg_build" --disable-shared
make
make install

libvpx

VP8/VP9视频编/解码器
如果不需要此编/解码器，可以跳过，并在ffmpeg的configure命令中移除--enable-libvpx

cd ~/ffmpeg_sources
git clone --depth 1 https://github.com/webmproject/libvpx.git
cd libvpx
./configure --prefix="$HOME/ffmpeg_build" --disable-examples --disable-unit-tests --enable-vp9-highbitdepth --as=yasm
make
make install

编译安装ffmpeg 3.3.8

cd ~/ffmpeg_sources
curl -O -L https://ffmpeg.org/releases/ffmpeg-3.3.8.tar.bz2
tar xjvf ffmpeg-3.3.8.tar.bz2
cd ffmpeg-3.3.8
PATH="$HOME/bin:$PATH" PKG_CONFIG_PATH="$HOME/ffmpeg_build/lib/pkgconfig" ./configure \--prefix="$HOME/ffmpeg_build" \--pkg-config-flags="--static" \--extra-cflags="-I$HOME/ffmpeg_build/include" \--extra-ldflags="-L$HOME/ffmpeg_build/lib" \--extra-libs=-lpthread \--extra-libs=-lm \--bindir="$HOME/bin" \--enable-gpl \--enable-libfdk_aac \--enable-libfreetype \--enable-libmp3lame \--enable-libopus \--enable-libvorbis \--enable-libvpx \--enable-libx264 \--enable-libx265 \--enable-nonfree
make
make install
hash -r

验证安装

ffmpeg -h

使用FFmpeg

识别视频信息

通过ffprobe命令识别并输出视频信息

ffprobe -v error -show_streams -print_format json <input>

为方便程序解析，将视频信息输出为json格式，样例如下：

{"streams": [{"index": 0,"codec_name": "h264","codec_long_name": "H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10","profile": "High","codec_type": "video","codec_time_base": "61127/3668400","codec_tag_string": "avc1","codec_tag": "0x31637661","width": 1920,"height": 1080,"coded_width": 1920,"coded_height": 1080,"has_b_frames": 0,"sample_aspect_ratio": "0:1","display_aspect_ratio": "0:1","pix_fmt": "yuv420p","level": 40,"color_range": "tv","color_space": "bt709","color_transfer": "bt709","color_primaries": "bt709","chroma_location": "left","refs": 1,"is_avc": "true","nal_length_size": "4","r_frame_rate": "30/1","avg_frame_rate": "1834200/61127","time_base": "1/600","start_pts": 0,"start_time": "0.000000","duration_ts": 61127,"duration": "101.878333","bit_rate": "16279946","bits_per_raw_sample": "8","nb_frames": "3057","disposition": {"default": 1,"dub": 0,"original": 0,"comment": 0,"lyrics": 0,"karaoke": 0,"forced": 0,"hearing_impaired": 0,"visual_impaired": 0,"clean_effects": 0,"attached_pic": 0,"timed_thumbnails": 0},"tags": {"rotate": "90","creation_time": "2018-08-09T09:13:33.000000Z","language": "und","handler_name": "Core Media Data Handler","encoder": "H.264"},"side_data_list": [{"side_data_type": "Display Matrix","displaymatrix": "\n00000000:            0       65536           0\n00000001:       -65536           0           0\n00000002:     70778880           0  1073741824\n","rotation": -90}]},{"index": 1,"codec_name": "aac","codec_long_name": "AAC (Advanced Audio Coding)","profile": "LC","codec_type": "audio","codec_time_base": "1/44100","codec_tag_string": "mp4a","codec_tag": "0x6134706d","sample_fmt": "fltp","sample_rate": "44100","channels": 1,"channel_layout": "mono","bits_per_sample": 0,"r_frame_rate": "0/0","avg_frame_rate": "0/0","time_base": "1/44100","start_pts": 0,"start_time": "0.000000","duration_ts": 4492835,"duration": "101.878345","bit_rate": "91595","max_bit_rate": "96000","nb_frames": "4390","disposition": {"default": 1,"dub": 0,"original": 0,"comment": 0,"lyrics": 0,"karaoke": 0,"forced": 0,"hearing_impaired": 0,"visual_impaired": 0,"clean_effects": 0,"attached_pic": 0,"timed_thumbnails": 0},"tags": {"creation_time": "2018-08-09T09:13:33.000000Z","language": "und","handler_name": "Core Media Data Handler"}},{"index": 2,"codec_type": "data","codec_tag_string": "mebx","codec_tag": "0x7862656d","r_frame_rate": "0/0","avg_frame_rate": "0/0","time_base": "1/600","start_pts": 0,"start_time": "0.000000","duration_ts": 61127,"duration": "101.878333","bit_rate": "119","nb_frames": "17","disposition": {"default": 1,"dub": 0,"original": 0,"comment": 0,"lyrics": 0,"karaoke": 0,"forced": 0,"hearing_impaired": 0,"visual_impaired": 0,"clean_effects": 0,"attached_pic": 0,"timed_thumbnails": 0},"tags": {"creation_time": "2018-08-09T09:13:33.000000Z","language": "und","handler_name": "Core Media Data Handler"}},{"index": 3,"codec_type": "data","codec_tag_string": "mebx","codec_tag": "0x7862656d","r_frame_rate": "0/0","avg_frame_rate": "0/0","time_base": "1/600","start_pts": 0,"start_time": "0.000000","duration_ts": 61127,"duration": "101.878333","nb_frames": "1","disposition": {"default": 1,"dub": 0,"original": 0,"comment": 0,"lyrics": 0,"karaoke": 0,"forced": 0,"hearing_impaired": 0,"visual_impaired": 0,"clean_effects": 0,"attached_pic": 0,"timed_thumbnails": 0},"tags": {"creation_time": "2018-08-09T09:13:33.000000Z","language": "und","handler_name": "Core Media Data Handler"}}]
}

折叠

可以看到一共返回了4个流，其中第0个是视频流，1是音频流，2和3是附加数据，没什么用
如果想指定分析视频流或音频流的话，可以加上参数-show_streams -v或-show_streams -a，这样就会只输出视频/音频流的分析结果

视频转码

ffmpeg -i <input> -c:v libx264 -b:v 2048k -vf scale=1280:-1 -y <output>

上述命令将输入视频转码为h264编码的视频

-c:v：指定编码器，编码器列表可以使用ffmpeg -codecs查看
-vf scale：指定输出视频的宽高，高-1代表按照比例自动适应
-b:v：指定输出视频的码率，即输出视频每秒的bit数
libx264支持的其他参数请使用ffmpeg -h encoder=libx264命令查询，如转码为其他编码，也可使用类似命令查询可用参数

使用Nvidia显卡GPU进行转码

重头戏来了，这块的资料相当少，我也是费了一番力气才搞定

CUDA

CUDA是Nvidia出的一个GPU计算库，让程序员可以驱动Nvidia显卡的GPU进行各种工作，其中就包含了视频的编解码

安装CUDA

首先验证一下显卡驱动是否装好

nvidia-smi

如果驱动正常的话，此命令会输出显卡的型号、驱动版本、现存/GPU占用等信息。如何安装显卡驱动本文不描述，请参考其他资料。

到CUDA官网CUDA Toolkit 11.7 Downloads | NVIDIA Developer下载对应平台的发行包，这里我选择Centos7对应的rpm包cuda-repo-rhel7-9-2-local-9.2.148-1.x86_64.rpm

执行如下命令安装：

rpm -i cuda-repo-rhel7-9-2-local-9.2.148-1.x86_64.rpm
yum clean all
yum install cuda

一共大概要安装90多个依赖库，注意一下安装完成后的报告，我首次安装时有一个库不知道为什么安装失败了，又单独yum install了该库一次才成功

验证安装

/usr/local/cuda-9.2/bin/nvcc -V

安装成功的话，会输出类似文本：

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Tue_Jun_12_23:07:04_CDT_2018
Cuda compilation tools, release 9.2, V9.2.148

重新编译ffmpeg

要让ffmpeg能够使用CUDA提供的GPU编解码器，必须重新编译ffmpeg，让其能够通过动态链接调用CUDA的能力

首先要编译安装nv-codec-headers库

git clone https://git.videolan.org/git/ffmpeg/nv-codec-headers.git
make PREFIX="$HOME/ffmpeg_build" BINDDIR="$HOME/bin"
make install PREFIX="$HOME/ffmpeg_build" BINDDIR="$HOME/bin"

进入~/ffmepg_sources/ffmpeg-3.3.8/目录重新执行ffmpeg的编译和安装
注意configure命令参数和之前configure命令参数的区别

PATH="$HOME/bin:$PATH" PKG_CONFIG_PATH="$HOME/ffmpeg_build/lib/pkgconfig" ./configure \--prefix="$HOME/ffmpeg_build" \--pkg-config-flags="--static" \--extra-cflags="-I$HOME/ffmpeg_build/include -I/usr/local/cuda/include" \--extra-ldflags="-L$HOME/ffmpeg_build/lib -L/usr/local/cuda/lib64" \--extra-libs=-lpthread \--extra-libs=-lm \--bindir="$HOME/bin" \--enable-gpl \--enable-libfdk_aac \--enable-libfreetype \--enable-libmp3lame \--enable-libopus \--enable-libvorbis \--enable-libvpx \--enable-libx264 \--enable-libx265 \--enable-nonfree \--enable-cuda \--enable-cuvid \--enable-nvenc \--enable-libnpp
make
make install
hash -r ffmpegln -s /opt/ffmpeg/ffmpeg/ffmpeg /usr/bin/ffmpeg

验证安装

重新安装完ffmpeg，使用ffmpeg -hwaccels命令查看支持的硬件加速选项

Hardware acceleration methods:
cuvid

可以看到多出来一种叫做cuvid的硬件加速选项，这就是CUDA提供的GPU视频编解码加速选项

然后查看cuvid提供的GPU编解码器ffmpeg -codecs | grep cuvid

 DEV.LS h264                 H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (decoders: h264 h264_cuvid ) (encoders: libx264 libx264rgb h264_nvenc nvenc nvenc_h264 )DEV.L. hevc                 H.265 / HEVC (High Efficiency Video Coding) (decoders: hevc hevc_cuvid ) (encoders: libx265 nvenc_hevc hevc_nvenc )DEVIL. mjpeg                Motion JPEG (decoders: mjpeg mjpeg_cuvid )DEV.L. mpeg1video           MPEG-1 video (decoders: mpeg1video mpeg1_cuvid )DEV.L. mpeg2video           MPEG-2 video (decoders: mpeg2video mpegvideo mpeg2_cuvid )DEV.L. mpeg4                MPEG-4 part 2 (decoders: mpeg4 mpeg4_cuvid )D.V.L. vc1                  SMPTE VC-1 (decoders: vc1 vc1_cuvid )DEV.L. vp8                  On2 VP8 (decoders: vp8 libvpx vp8_cuvid ) (encoders: libvpx )DEV.L. vp9                  Google VP9 (decoders: vp9 libvpx-vp9 vp9_cuvid ) (encoders: libvpx-vp9 )

所有带有"cuvid"或"nvenc"的，都是CUDA提供的GPU编解码器
可以看到，我们现在可以进行h264/hevc/mjpeg/mpeg1/mpeg2/mpeg4/vc1/vp8/vp9格式的GPU解码，以及h264/hevc格式的GPU编码

使用GPU进行视频转码

用GPU进行转码的命令和软转码命令不太一样，CPU转码的时候，我们可以依赖ffmpeg识别输入视频的编码格式并选择对应的解码器，但ffmpeg只会自动选择CPU解码器，要让ffmpeg使用GPU解码器，必须先用ffprobe识别出输入视频的编码格式，然后在命令行中指定对应的GPU解码器。

例如，将h264编码的源视频转码为指定尺寸和码率的h264编码视频：

ffmpeg -hwaccel cuvid -c:v h264_cuvid -i <input> -c:v h264_nvenc -b:v 2048k -vf scale_npp=1280:-1 -y <output>

-hwaccel cuvid：指定使用cuvid硬件加速
-c:v h264_cuvid：使用h264_cuvid进行视频解码
-c:v h264_nvenc：使用h264_nvenc进行视频编码
-vf scale_npp=1280:-1：指定输出视频的宽高，注意，这里和软解码时使用的-vf scale=x:x不一样

转码期间使用nvidia-smi查看显卡状态，能够看到ffmpeg确实是在使用GPU进行转码：

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0     62543      C   ffmpeg                                       193MiB |
+-----------------------------------------------------------------------------+

GPU转码效率测试

在配有两颗Intel-E5-2630v3 CPU和两块Nvidia Tesla M4显卡的服务器上，进行h264视频转码测试，成绩如下：

GPU转码平均耗时：8s
CPU转码平均耗时：25s

并行转码时，CPU软转的效率有所提高，3个转码任务并行时32颗核心全被占满，此时的成绩

GPU转码平均耗时：8s
CPU转码平均耗时：18s

不难看出，并行时GPU的转码速度并没有提高，可见一颗GPU同时只能执行一个转码任务。那么，如果服务器上插有多块显卡，ffmpeg是否会使用多颗GPU进行并行转码呢？

很遗憾，答案是否。

ffmpeg并不具备自动向不同GPU分配转码任务的能力，但经过一番调查后，发现可以通过-hwaccel_device参数指定转码任务使用的GPU！

向不同GPU提交转码任务

ffmpeg -hwaccel cuvid -hwaccel_device 0 -c:v h264_cuvid -i <input> -c:v h264_nvenc -b:v 2048k -vf scale_npp=1280:-1 -y <output>
ffmpeg -hwaccel cuvid -hwaccel_device 1 -c:v h264_cuvid -i <input> -c:v h264_nvenc -b:v 2048k -vf scale_npp=1280:-1 -y <output>

-hwaccel_device N：指定某颗GPU执行转码任务，N为数字

此时nvidia-smi显示：

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0     96931      C   ffmpeg                                       193MiB |
|    1     96930      C   ffmpeg                                       193MiB |
+-----------------------------------------------------------------------------+

可以进行并行GPU转码了！

那么在占满服务器资源时，GPU转码和CPU转码的效率如下：

GPU转码平均耗时：4s
CPU转码平均耗时：18s

GPU效率是CPU的4.5倍