(FFmpeg) VP9 Vaapi encoding to a .mp4 or .webm container from given official ffmpeg example

1
votes

I'm trying to implement vp9 hardware acceleration encoding process. I followed ffmpeg offical github's example (Here -> vaapi_encode.c).

But given example only save a .yuv file to .h264 file, I would like to save the frames to either .mp4 or .webm container. And having the ability to control the quality, and etc.

I'm not reading frames from a file, I'm collecting frames from a live feed. When having full 5 secs of frames from the live feed, encode those frames using vp9_vaapi to a 5 secs .mp4 file.

I'm able to save all the 5 secs frames from my live feed to a .mp4 or .webm file, but they couldn't be played correctly (more precisely: keep loading, and I open).

The result from the official site's example:

enter image description here

The cpu encoded vp9 .mp4 file result:

enter image description here

Edit: Result enter image description here

1
encodingffmpegmp4hardware-accelerationvaapi

1 Answers

2
votes

You will need to use FFmpeg directly, where you may optionally add the vp9_superframe and the vp9_raw_reorder bitstream filters in the same command line if you enable B-frames in the vp9_vaapi encoder.

Example:

ffmpeg -threads 4 -vaapi_device /dev/dri/renderD128 \
-hwaccel vaapi -hwaccel_output_format vaapi \
-i http://server:port \
-c:v vp9_vaapi -global_quality 50 -bf 1 \
-bsf:v vp9_raw_reorder,vp9_superframe \
-f segment -segment_time 5 -segment_format_options movflags=+faststart output%03d.mp4

Adjust your input and output paths/urls as needed.

What this command does:

It will create 5 second long mp4 segments, via the segment muxer. See the usage of the movflags=+faststart , and how it has been passed as a format option to the underlying mp4 muxer via -segment_format_options flag above.

The segment lengths may not be exactly 5 seconds long, as each segment begins (gets cut on) (with) a keyframe.

However, I'd not recommend enabling B-frames in that encoder, as these bitstream filters have other undesired effects, such as mucking around with the encoder's rate control and triggering bugs like this one. This is not desirable in a production environment. This is why the scripts below do not have that option enabled, and instead, we define a set rate control mode directly in the encoder options.

If you need to take advantage of 1:N encoding with VAAPI, use these snippets:

  1. If you need to deinterlace, call up the deinterlace_vaapi filter:
    ffmpeg -loglevel debug -threads 4 \
    -init_hw_device vaapi=va:/dev/dri/renderD128 -hwaccel vaapi \
    -hwaccel_device va -filter_hw_device va \
    -hwaccel_output_format vaapi \
    -i 'http://server:port' \
    -filter_complex "[0:v]deinterlace_vaapi,split=3[n0][n1][n2]; \
    [n0]scale_vaapi=1152:648[v0]; \
    [n1]scale_vaapi=848:480[v1];
    [n2]scale_vaapi=640:360[v2]" \
    -b:v:0 2250k -maxrate:v:0 2250k -bufsize:v:0 360k -c:v:0 vp9_vaapi -g:v:0 50 -r:v:0 25 -rc_mode:v:0 2 \
    -b:v:1 1750k -maxrate:v:1 1750k -bufsize:v:1 280k -c:v:1 vp9_vaapi -g:v:1 50 -r:v:1 25 -rc_mode:v:1 2 \
    -b:v:2 1000k -maxrate:v:2 1000k -bufsize:v:2 160k -c:v:2 vp9_vaapi -g:v:2 50 -r:v:2 25 -rc_mode:v:2 2 \
    -c:a aac -b:a 128k -ar 48000 -ac 2 \
    -flags -global_header -f tee -use_fifo 1 \
    -map "[v0]" -map "[v1]" -map "[v2]" -map 0:a \
    "[select=\'v:0,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path0/output%03d.mp4| \
     [select=\'v:1,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path1/output%03d.mp4| \
     [select=\'v:2,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path2/output%03d.mp4"
  1. Without deinterlacing:
    ffmpeg -loglevel debug -threads 4 \
    -init_hw_device vaapi=va:/dev/dri/renderD128 -hwaccel vaapi \
    -hwaccel_device va -filter_hw_device va -hwaccel_output_format vaapi \
    -i 'http://server:port' \
    -filter_complex "[0:v]split=3[n0][n1][n2]; \
    [n0]scale_vaapi=1152:648[v0]; \
    [n1]scale_vaapi=848:480[v1];
    [n2]scale_vaapi=640:360[v2]" \
    -b:v:0 2250k -maxrate:v:0 2250k -bufsize:v:0 2250k -c:v:0 vp9_vaapi -g:v:0 50 -r:v:0 25 -rc_mode:v:0 2  \
    -b:v:1 1750k -maxrate:v:1 1750k -bufsize:v:1 1750k -c:v:1 vp9_vaapi -g:v:1 50 -r:v:1 25 -rc_mode:v:1 2  \
    -b:v:2 1000k -maxrate:v:2 1000k -bufsize:v:2 1000k -c:v:2 vp9_vaapi -g:v:2 50 -r:v:2 25 -rc_mode:v:2 2  \
    -c:a aac -b:a 128k -ar 48000 -ac 2 \
    -flags -global_header -f tee -use_fifo 1 \
    -map "[v0]" -map "[v1]" -map "[v2]" -map 0:a \
    "[select=\'v:0,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path0/output%03d.mp4| \
     [select=\'v:1,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path1/output%03d.mp4| \
     [select=\'v:2,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path2/output%03d.mp4"
  1. Using Intel's QuickSync (on supported platforms):

On Intel Icelake and above, you can use the vp9_qsv encoder wrapper with the following known limitations (for now):

(a). You must enable low_power mode because only the VDENC decode path is exposed by the iHD driver for now.

(b). Coding option1 and extra_data are not supported by MSDK.

(c). The IVF header will be inserted in MSDK by default, but it is not needed for FFmpeg, and remains disabled by default.

See the examples below:

  1. If you need to deinterlace, call up the vpp_qsv filter:
    ffmpeg -nostdin -y -fflags +genpts \
    -init_hw_device vaapi=va:/dev/dri/renderD128,driver=iHD \
    -filter_hw_device va -hwaccel vaapi -hwaccel_output_format vaapi
    -threads 4 -vsync 1 -async 1 \
    -i 'http://server:port' \
    -filter_complex "[0:v]hwmap=derive_device=qsv,format=qsv,vpp_qsv=deinterlace=2:async_depth=4,split[n0][n1][n2]; \
    [n0]vpp_qsv=w=1152:h=648:async_depth=4[v0]; \
    [n1]vpp_qsv=w=848:h=480:async_depth=4[v1];
    [n2]vpp_qsv=w=640:h=360:async_depth=4[v2]" \
    -b:v:0 2250k -maxrate:v:0 2250k -bufsize:v:0 360k -c:v:0 vp9_qsv -g:v:0 50 -r:v:0 25 -low_power:v:0 2 \
    -b:v:1 1750k -maxrate:v:1 1750k -bufsize:v:1 280k -c:v:1 vp9_qsv -g:v:1 50 -r:v:1 25 -low_power:v:1 2 \
    -b:v:2 1000k -maxrate:v:2 1000k -bufsize:v:2 160k -c:v:2 vp9_qsv -g:v:2 50 -r:v:2 25 -low_power:v:2 2 \
    -c:a aac -b:a 128k -ar 48000 -ac 2 \
    -flags -global_header -f tee -use_fifo 1 \
    -map "[v0]" -map "[v1]" -map "[v2]" -map 0:a \
    "[select=\'v:0,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path0/output%03d.mp4| \
     [select=\'v:1,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path1/output%03d.mp4| \
     [select=\'v:2,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path2/output%03d.mp4"
  1. Without deinterlacing:
    ffmpeg -nostdin -y -fflags +genpts \
    -init_hw_device vaapi=va:/dev/dri/renderD128,driver=iHD \
    -filter_hw_device va -hwaccel vaapi -hwaccel_output_format vaapi
    -threads 4 -vsync 1 -async 1 \
    -i 'http://server:port' \
    -filter_complex "[0:v]hwmap=derive_device=qsv,format=qsv,split=3[n0][n1][n2]; \
    [n0]vpp_qsv=w=1152:h=648:async_depth=4[v0]; \
    [n1]vpp_qsv=w=848:h=480:async_depth=4[v1];
    [n2]vpp_qsv=w=640:h=360:async_depth=4[v2]" \
    -b:v:0 2250k -maxrate:v:0 2250k -bufsize:v:0 2250k -c:v:0 vp9_qsv -g:v:0 50 -r:v:0 25 -low_power:v:0 2  \
    -b:v:1 1750k -maxrate:v:1 1750k -bufsize:v:1 1750k -c:v:1 vp9_qsv -g:v:1 50 -r:v:1 25 -low_power:v:1 2  \
    -b:v:2 1000k -maxrate:v:2 1000k -bufsize:v:2 1000k -c:v:2 vp9_qsv -g:v:2 50 -r:v:2 25 -low_power:v:2 2  \
    -c:a aac -b:a 128k -ar 48000 -ac 2 \
    -flags -global_header -f tee -use_fifo 1 \
    -map "[v0]" -map "[v1]" -map "[v2]" -map 0:a \
    "[select=\'v:0,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path0/output%03d.mp4| \
     [select=\'v:1,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path1/output%03d.mp4| \
     [select=\'v:2,a\':f=segment:segment_time=5:segment_format_options=movflags=+faststart]$output_path2/output%03d.mp4"

Note that we use the vpp_qsv filter with the async_depth option set to 4. This massively improves transcode performance over using scale_qsv and deinterlace_qsv. See this commit on FFmpeg's git.

Note: If you use the QuickSync path, note that MFE (Multi-Frame encoding mode) will be enabled by default if the Media SDK library on your system supports it.

Formulae used to derive the snippet above:

Optimal bufsize:v = target bitrate (-b:v value)

Set GOP size as: 2 * fps (GOP interval set to 2 seconds).

We limit the thread counts for the video encoder(s) via -threads:v to prevent VBV overflows.

Resolution ladder used: 640p, 480p and 360p in 16:9, see this link. Adjust this as needed.

Substitute the variables above ($output_path{0-2}, the input, etc) as needed.

Test and report back.

Current observations:

On my system, I'm able to encode up to 5 streams at real-time with VP9 using Apple's recommended resolutions and bit-rates for HEVC encoding for HLS as a benchmark. See the picture below on system load, etc.

system load and stats with 5 simultaneous VAAPI encodes in progress

Platform details:

I'm on a Coffee-lake system, using the i965 driver for this workflow:

libva info: VA-API version 1.5.0
libva info: va_getDriverName() returns 0
libva info: User requested driver 'i965'
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/i965_drv_video.so
libva info: Found init function __vaDriverInit_1_5
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.5 (libva 2.4.0.pre1)
vainfo: Driver version: Intel i965 driver for Intel(R) Coffee Lake - 2.4.0.pre1 (2.3.0-11-g881e67a)
vainfo: Supported profile and entrypoints
      VAProfileMPEG2Simple            : VAEntrypointVLD
      VAProfileMPEG2Simple            : VAEntrypointEncSlice
      VAProfileMPEG2Main              : VAEntrypointVLD
      VAProfileMPEG2Main              : VAEntrypointEncSlice
      VAProfileH264ConstrainedBaseline: VAEntrypointVLD
      VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice
      VAProfileH264ConstrainedBaseline: VAEntrypointEncSliceLP
      VAProfileH264Main               : VAEntrypointVLD
      VAProfileH264Main               : VAEntrypointEncSlice
      VAProfileH264Main               : VAEntrypointEncSliceLP
      VAProfileH264High               : VAEntrypointVLD
      VAProfileH264High               : VAEntrypointEncSlice
      VAProfileH264High               : VAEntrypointEncSliceLP
      VAProfileH264MultiviewHigh      : VAEntrypointVLD
      VAProfileH264MultiviewHigh      : VAEntrypointEncSlice
      VAProfileH264StereoHigh         : VAEntrypointVLD
      VAProfileH264StereoHigh         : VAEntrypointEncSlice
      VAProfileVC1Simple              : VAEntrypointVLD
      VAProfileVC1Main                : VAEntrypointVLD
      VAProfileVC1Advanced            : VAEntrypointVLD
      VAProfileNone                   : VAEntrypointVideoProc
      VAProfileJPEGBaseline           : VAEntrypointVLD
      VAProfileJPEGBaseline           : VAEntrypointEncPicture
      VAProfileVP8Version0_3          : VAEntrypointVLD
      VAProfileVP8Version0_3          : VAEntrypointEncSlice
      VAProfileHEVCMain               : VAEntrypointVLD
      VAProfileHEVCMain               : VAEntrypointEncSlice
      VAProfileHEVCMain10             : VAEntrypointVLD
      VAProfileHEVCMain10             : VAEntrypointEncSlice
      VAProfileVP9Profile0            : VAEntrypointVLD
      VAProfileVP9Profile0            : VAEntrypointEncSlice
      VAProfileVP9Profile2            : VAEntrypointVLD

My ffmpeg build info:

ffmpeg -buildconf
ffmpeg version N-93308-g1144d5c96d Copyright (c) 2000-2019 the FFmpeg developers
  built with gcc 7 (Ubuntu 7.3.0-27ubuntu1~18.04)
  configuration: --pkg-config-flags=--static --prefix=/home/brainiarc7/bin --bindir=/home/brainiarc7/bin --extra-cflags=-I/home/brainiarc7/bin/include --extra-ldflags=-L/home/brainiarc7/bin/lib --enable-cuda-nvcc --enable-cuvid --enable-libnpp --extra-cflags=-I/usr/local/cuda/include/ --extra-ldflags=-L/usr/local/cuda/lib64/ --enable-nvenc --extra-cflags=-I/opt/intel/mediasdk/include --extra-ldflags=-L/opt/intel/mediasdk/lib --extra-ldflags=-L/opt/intel/mediasdk/plugins --enable-libmfx --enable-libass --enable-vaapi --disable-debug --enable-libvorbis --enable-libvpx --enable-libdrm --enable-opencl --enable-gpl --cpu=native --enable-opengl --enable-libfdk-aac --enable-libx265 --enable-openssl --extra-libs='-lpthread -lm' --enable-nonfree
  libavutil      56. 26.100 / 56. 26.100
  libavcodec     58. 47.103 / 58. 47.103
  libavformat    58. 26.101 / 58. 26.101
  libavdevice    58.  6.101 / 58.  6.101
  libavfilter     7. 48.100 /  7. 48.100
  libswscale      5.  4.100 /  5.  4.100
  libswresample   3.  4.100 /  3.  4.100
  libpostproc    55.  4.100 / 55.  4.100

  configuration:
    --pkg-config-flags=--static
    --prefix=/home/brainiarc7/bin
    --bindir=/home/brainiarc7/bin
    --extra-cflags=-I/home/brainiarc7/bin/include
    --extra-ldflags=-L/home/brainiarc7/bin/lib
    --enable-cuda-nvcc
    --enable-cuvid
    --enable-libnpp
    --extra-cflags=-I/usr/local/cuda/include/
    --extra-ldflags=-L/usr/local/cuda/lib64/
    --enable-nvenc
    --extra-cflags=-I/opt/intel/mediasdk/include
    --extra-ldflags=-L/opt/intel/mediasdk/lib
    --extra-ldflags=-L/opt/intel/mediasdk/plugins
    --enable-libmfx
    --enable-libass
    --enable-vaapi
    --disable-debug
    --enable-libvorbis
    --enable-libvpx
    --enable-libdrm
    --enable-opencl
    --enable-gpl
    --cpu=native
    --enable-opengl
    --enable-libfdk-aac
    --enable-libx265
    --enable-openssl
    --extra-libs='-lpthread -lm'
    --enable-nonfree

And output from inxi:

inxi -F
System:    Host: cavaliere Kernel: 5.0.0 x86_64 bits: 64 Desktop: Gnome 3.28.3 Distro: Ubuntu 18.04.2 LTS
Machine:   Device: laptop System: ASUSTeK product: Zephyrus M GM501GS v: 1.0 serial: N/A
           Mobo: ASUSTeK model: GM501GS v: 1.0 serial: N/A
           UEFI: American Megatrends v: GM501GS.308 date: 10/01/2018
Battery    BAT0: charge: 49.3 Wh 100.0% condition: 49.3/55.0 Wh (90%)
CPU:       6 core Intel Core i7-8750H (-MT-MCP-) cache: 9216 KB
           clock speeds: max: 4100 MHz 1: 2594 MHz 2: 3197 MHz 3: 3633 MHz 4: 3514 MHz 5: 3582 MHz 6: 3338 MHz
           7: 3655 MHz 8: 3684 MHz 9: 1793 MHz 10: 3651 MHz 11: 3710 MHz 12: 3662 MHz
Graphics:  Card-1: Intel Device 3e9b
           Card-2: NVIDIA GP104M [GeForce GTX 1070 Mobile]
           Display Server: x11 (X.Org 1.19.6 ) drivers: modesetting,nvidia (unloaded: fbdev,vesa,nouveau)
           Resolution: [email protected]
           OpenGL: renderer: GeForce GTX 1070/PCIe/SSE2 version: 4.6.0 NVIDIA 418.43
Audio:     Card-1 Intel Cannon Lake PCH cAVS driver: snd_hda_intel Sound: ALSA v: k5.0.0
           Card-2 NVIDIA GP104 High Definition Audio Controller driver: snd_hda_intel
           Card-3 Kingston driver: USB Audio
Network:   Card: Intel Wireless-AC 9560 [Jefferson Peak] driver: iwlwifi
           IF: wlo1 state: up mac: (redacted)
Drives:    HDD Total Size: 3050.6GB (94.5% used)
           ID-1: /dev/nvme0n1 model: Samsung_SSD_960_EVO_1TB size: 1000.2GB
           ID-2: /dev/sda model: Crucial_CT2050MX size: 2050.4GB
Partition: ID-1: / size: 246G used: 217G (94%) fs: ext4 dev: /dev/nvme0n1p5
           ID-2: swap-1 size: 8.59GB used: 0.00GB (0%) fs: swap dev: /dev/nvme0n1p6
RAID:      No RAID devices: /proc/mdstat, md_mod kernel module present
Sensors:   System Temperatures: cpu: 64.0C mobo: N/A gpu: 61C
           Fan Speeds (in rpm): cpu: N/A
Info:      Processes: 412 Uptime: 3:32 Memory: 4411.3/32015.5MB Client: Shell (bash) inxi: 2.3.56

Why that last bit is included:

I'm running the latest Linux kernel to date, version 5.0. The same also applies to the graphics driver stack, on Ubuntu 18.04LTS. FFmpeg was built as shown here as this laptop has both NVIDIA+ Intel GPU enabled via Optimus. That way, I can tap into VAAPI, QuickSync and NVENC hwaccels as needed. Your mileage may vary even if our hardware is identical.

References:

  1. See the encoder options, including rate control methods supported:
ffmpeg -h encoder=vp9_vaapi
  1. See the deinterlace_vaapi filter usage options:
ffmpeg -h filter=deinterlace_vaapi
  1. On the vpp_qsv filter usage, see:
ffmpeg -h filter=vpp_qsv

For instance, if you want field-rate output rather than frame-rate output from the deinterlacer, you could pass the rate=field option to it instead:

-vf=vaapi_deinterlace=rate=field

This feature, for instance, is tied to encoders that support MBAFF. Others, such as the NVENC-based ones in FFmpeg, do not have this implemented (as of the time of writing).

Tips on getting ahead with FFmpeg:

Where possible, infer to the built-in docs, as with the examples shown above. They can uncover potential pitfalls that you may be able to avoid by understanding how filter chaining and encoder initialization works, unsupported features, etc, and the impact on performance.

For example, you'll see that in the snippets above, we call up the deinterlacer only once, then split its' output via the split filter to separate scalers. This is done to lower the overhead that would be incurred had we called up the deinterlacer more than once, and it would've been wasteful.

Warning:

Note that the SDK requires at least 2 threads to prevent deadlock, see this code block. This is why we set -threads 4 in ffmpeg.