Add support for Intel QSV codecs

[ladaapp]

Adds support for Intel hardware-accelerated QSV encoders and decoders available for discrete (Arc) and integrated (Xe) Intel GPUs.

How to test

Check out this PR branch and build pyav-ffmpeg

cd pyav-ffmpeg
python -m venv .venv
source .venv/bin/activate
python scripts/build-ffmpeg.py pyav_ffmpeg_qsv_build
PYAV_FFMPEG_BUILD=$(pwd)/pyav_ffmpeg_qsv_build

Checkout PyAV and build it using the build artifacts from above

git clone https://github.com/PyAV-Org/PyAV.git
cd PyAV
python -m venv .venv
source .venv/bin/activate
pip install setuptools cython
python setup.py build_ext --inplace --ffmpeg-dir="$PYAV_FFMPEG_BUILD"

Check if qsv codecs show up

python -m av --codecs | grep qsv
 DEV.L. av1_qsv            AV1 (Intel Quick Sync Video acceleration)
 DEV.LS h264_qsv           H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (Intel Quick Sync Video acceleration)
 DEV.L. hevc_qsv           HEVC (Intel Quick Sync Video acceleration)
 DEVIL. mjpeg_qsv          MJPEG (Intel Quick Sync Video acceleration)
 DEV.L. mpeg2_qsv          MPEG-2 video (Intel Quick Sync Video acceleration)
 D.V.L. vc1_qsv            VC1 video (Intel Quick Sync Video acceleration)
 D.V.L. vp8_qsv            VP8 video (Intel Quick Sync Video acceleration)
 DEV.L. vp9_qsv            VP9 video (Intel Quick Sync Video acceleration)
 D.V.L. vvc_qsv            VVC video (Intel Quick Sync Video acceleration)

Similar to AMD or Nvidia codecs we need a runtime component to be able to use it. I installed these packages on ArchLinux. On other distros the package(s) may be named differently. I did not test on Windows.

sudo pacman -S libvpl vpl-gpu-rt

Test transcoding using h264_qsv encoder:

import av
import av.datasets

input_container = av.open(av.datasets.curated("pexels/time-lapse-video-of-night-sky-857195.mp4"))
output_container = av.open("h264_qsv_encoded.mp4", "w", options={})

input_stream = input_container.streams.video[0]
output_stream = output_container.add_stream('h264_qsv', input_stream.average_rate)
output_stream.width = input_stream.width
output_stream.height = input_stream.height
output_stream.pix_fmt = "nv12"

for frame in input_container.decode(video=0):
    packet = output_stream.encode(frame)
    if packet: output_container.mux(packet)

input_container.close()
output_container.close()

Documentation:

• https://github.com/intel/libvpl
• https://trac.ffmpeg.org/wiki/Hardware/QuickSync

[ladaapp]

The CI currently doesn't work as nasm package cannot be downloaded as their website is down:

Failed to connect to www.nasm.us port 443: Connection timed out

I found this mirror and manually downloaded it to fix this locally on my system
https://distfiles.macports.org/nasm/nasm-2.16.03.tar.bz2

Let's hope the official nasm.us site comes back 🤞

[ladaapp]

@WyattBlue https://nasm.us/ now seems to be back online. Unfortunately it looks like that I don't have permissions to re-run the pipeline. Can you trigger that for me?