Jump to content

Draft:Dav1d: Difference between revisions

From Wikipedia, the free encyclopedia
Browse history interactively
← Previous editNext edit →
Content deleted Content added
VisualWikitext
Trudluc (talk | contribs)
59 edits
m Dav1d use at Meta
Trudluc (talk | contribs)
59 edits
m Firefox uses dav1d
Line 26: Line 26:
The first usable version of dav1d, called 0.1.0 Gazelle, was released in December 2018.<ref>{{Cite web |title=First release of dav1d, the AV1 decoder - Jean-Baptiste Kempf's Website |url=https://jbkempf.com/blog/First-release-of-dav1d/ |access-date=2024-04-29 |website=jbkempf.com}}</ref> Releases 0.1.0 to 0.7.1 were primarily dedicated to [[Color depth|low-bitdepth]] [[Single instruction, multiple data|SIMD]] optimizations, aimed at enhancing decoding efficiency on x86 and ARM.<ref>{{Cite web |title=dav1d 0.7.1 - Jean-Baptiste Kempf's Website |url=https://jbkempf.com/blog/dav1d-0.7.1/ |access-date=2024-04-29 |website=jbkempf.com}}</ref> In 2019, [[Benchmark (computing)|benchmarks]] showed that dav1d 0.5 was the fastest AV1 decoder in comparison to libgav1 and libaom.<ref>{{Cite web |last=Hoeven |first=Ewout ter |date=2019-10-10 |title=AV1 is ready for prime time Part 2: Decoding performance |url=https://medium.com/@ewoutterhoeven/av1-is-ready-for-prime-time-part-2-decoding-performance-d3428221313 |access-date=2024-04-29 |website=Medium |language=en}}</ref> Subsequent releases, from 0.7.1 to 1.0, shifted focus to high-bitdepth SIMD optimizations.<ref>{{Cite web |title=dav1d 0.9.1: a ton of asm - Jean-Baptiste Kempf's Website |url=https://jbkempf.com/blog/2021/dav1d-0.9.1-a-ton-of-asm/ |access-date=2024-04-29 |website=jbkempf.com}}</ref>
The first usable version of dav1d, called 0.1.0 Gazelle, was released in December 2018.<ref>{{Cite web |title=First release of dav1d, the AV1 decoder - Jean-Baptiste Kempf's Website |url=https://jbkempf.com/blog/First-release-of-dav1d/ |access-date=2024-04-29 |website=jbkempf.com}}</ref> Releases 0.1.0 to 0.7.1 were primarily dedicated to [[Color depth|low-bitdepth]] [[Single instruction, multiple data|SIMD]] optimizations, aimed at enhancing decoding efficiency on x86 and ARM.<ref>{{Cite web |title=dav1d 0.7.1 - Jean-Baptiste Kempf's Website |url=https://jbkempf.com/blog/dav1d-0.7.1/ |access-date=2024-04-29 |website=jbkempf.com}}</ref> In 2019, [[Benchmark (computing)|benchmarks]] showed that dav1d 0.5 was the fastest AV1 decoder in comparison to libgav1 and libaom.<ref>{{Cite web |last=Hoeven |first=Ewout ter |date=2019-10-10 |title=AV1 is ready for prime time Part 2: Decoding performance |url=https://medium.com/@ewoutterhoeven/av1-is-ready-for-prime-time-part-2-decoding-performance-d3428221313 |access-date=2024-04-29 |website=Medium |language=en}}</ref> Subsequent releases, from 0.7.1 to 1.0, shifted focus to high-bitdepth SIMD optimizations.<ref>{{Cite web |title=dav1d 0.9.1: a ton of asm - Jean-Baptiste Kempf's Website |url=https://jbkempf.com/blog/2021/dav1d-0.9.1-a-ton-of-asm/ |access-date=2024-04-29 |website=jbkempf.com}}</ref>


From the beginning, dav1d was designed to be multithreaded, allowing for a combination of both frame-level [[Parallelism (computing)|parallelism]] and tile-level parallelism. The 1.0.0 release introduced a [[thread pool]] design, where each component in the decoding loop runs as a generic task with a simple dependency management mechanism<ref>{{Cite AV media |url=https://www.youtube.com/watch?v=a4uretCJh_4 |title=Ronald S. Bultje - Low-level wizardry in dav1d |language=en |access-date=2024-05-03 |via=www.youtube.com}}</ref>. In this design, worker threads iterate over a list of available tasks, allowing for a more resource-efficient parallel decoding scheme that is independent of bitstream features like tiles.
From the beginning, dav1d was designed to be multithreaded, allowing for a combination of both frame-level [[Parallelism (computing)|parallelism]] and tile-level parallelism. The 1.0.0 release introduced a [[thread pool]] design, where each component in the decoding loop runs as a generic task with a simple dependency management mechanism.<ref>{{Cite AV media |url=https://www.youtube.com/watch?v=a4uretCJh_4 |title=Ronald S. Bultje - Low-level wizardry in dav1d |language=en |access-date=2024-05-03 |via=www.youtube.com}}</ref> In this design, worker threads iterate over a list of available tasks, allowing for a more resource-efficient parallel decoding scheme that is independent of bitstream features like tiles.


== Software that uses dav1d ==
== Software that uses dav1d ==
In February 2020, [[Netflix]] stated that their [[Android (operating system)|Android]] app relied on dav1d to decode 10-bit AV1 content<ref>{{Cite web |last=Blog |first=Netflix Technology |date=2020-02-05 |title=Netflix Now Streaming AV1 on Android |url=https://netflixtechblog.com/netflix-now-streaming-av1-on-android-d5264a515202 |access-date=2024-05-07 |website=Medium |language=en}}</ref>.
In May 2019, [[Mozilla]] announced that [[Firefox]] switched from libaom to dav1d as the default AV1 decoder on all desktop platforms (Windows, OSX and Linux) for both 32-bit and 64-bit systems.<ref>{{Cite web |title=Firefox brings you smooth video playback with the world’s fastest AV1 decoder – Mozilla Hacks - the Web developer blog |url=https://hacks.mozilla.org/2019/05/firefox-brings-you-smooth-video-playback-with-the-worlds-fastest-av1-decoder |access-date=2024-05-07 |website=Mozilla Hacks – the Web developer blog |language=en-US}}</ref>


In February 2023, [[Meta (Company)|Meta]] reported using dav1d to decode AV1 content for [[Facebook]] Reels and [[Instagram]] Reels in their [[iOS]] app<ref>{{Cite web |date=2023-02-21 |title=How Meta brought AV1 to Reels |url=https://engineering.fb.com/2023/02/21/video-engineering/av1-codec-facebook-instagram-reels/ |access-date=2024-05-07 |website=Engineering at Meta |language=en-US}}</ref>.
In February 2020, [[Netflix]] stated that their [[Android (operating system)|Android]] app relied on dav1d to decode 10-bit AV1 content.<ref>{{Cite web |last=Blog |first=Netflix Technology |date=2020-02-05 |title=Netflix Now Streaming AV1 on Android |url=https://netflixtechblog.com/netflix-now-streaming-av1-on-android-d5264a515202 |access-date=2024-05-07 |website=Medium |language=en}}</ref>

In February 2023, [[Meta (Company)|Meta]] reported using dav1d to decode AV1 content for [[Facebook]] Reels and [[Instagram]] Reels in their [[iOS]] app.<ref>{{Cite web |date=2023-02-21 |title=How Meta brought AV1 to Reels |url=https://engineering.fb.com/2023/02/21/video-engineering/av1-codec-facebook-instagram-reels/ |access-date=2024-05-07 |website=Engineering at Meta |language=en-US}}</ref>


== References ==
== References ==

Revision as of 14:23, 7 May 2024

dav1d
Developer(s)VideoLAN and FFmpeg communities
Repositorycode.videolan.org/videolan/dav1d.git
Written inAssembly language
C99
TypeVideo decoder
LicenseBSD 2-clause "Simplified" License
Websitewww.videolan.org/projects/dav1d.html

dav1d is a free and open-source software library and a command-line utility to decode video streams in the AV1 video coding format. It is developed by the VideoLAN and FFmpeg communities and sponsored by the Alliance for Open Media.[1] It is designed to be fast, small, cross-platform and multithreaded.[2]

The name dav1d is derived from a combination of factors. Pronounced "David", it merges the recursive acronym "dav1d is an AV1 decoder" with the leetspeak transformation of the name David, where the "i" is replaced by a "1" due to the visual similarity of their glyphs. This playful adaptation captures both the essence of the project as an AV1 decoder and pays homage to the hacker culture, where recursive acronyms and leetspeak are common occurrences.

dav1d is released under the BSD 2-clause "Simplified" License, a distinction from the Copyleft licensing favored by projects like FFmpeg and VideoLAN. This strategic choice of a more permissive license was made to encourage wider adoption of AV1. The decision to adopt a permissive license model for dav1d finds its roots in the Vorbis project, where similar licensing strategies were employed to promote adoption and prevent the entrenchment of proprietary formats.[3]

History

The Alliance for Open Media released AV1 in June 2018 with a production grade and SIMD optimized software decoder called aomdec as part of libaom. A widespread opinion was that the libaom software implementation was slow[4] both in terms of encoding and decoding speeds. The VideoLAN and FFmpeg communities, who had previously written faster software decoders for both VP8[5] and VP9[6] video formats, saw this as a barrier to entry for AV1 and decided to resolve this by writting a faster software decoder.

In October 2018, Ronald Bultje and Jean-Baptiste Kempf announced dav1d at the VideoLAN Developer Days 2018 conference.[7] The goal of the project being to develop a fast AV1 software decoder to jumpstart the AV1 ecosystem and act as a stopgap solution until AV1 hardware decoders are common place.

The first usable version of dav1d, called 0.1.0 Gazelle, was released in December 2018.[8] Releases 0.1.0 to 0.7.1 were primarily dedicated to low-bitdepth SIMD optimizations, aimed at enhancing decoding efficiency on x86 and ARM.[9] In 2019, benchmarks showed that dav1d 0.5 was the fastest AV1 decoder in comparison to libgav1 and libaom.[10] Subsequent releases, from 0.7.1 to 1.0, shifted focus to high-bitdepth SIMD optimizations.[11]

From the beginning, dav1d was designed to be multithreaded, allowing for a combination of both frame-level parallelism and tile-level parallelism. The 1.0.0 release introduced a thread pool design, where each component in the decoding loop runs as a generic task with a simple dependency management mechanism.[12] In this design, worker threads iterate over a list of available tasks, allowing for a more resource-efficient parallel decoding scheme that is independent of bitstream features like tiles.

Software that uses dav1d

In May 2019, Mozilla announced that Firefox switched from libaom to dav1d as the default AV1 decoder on all desktop platforms (Windows, OSX and Linux) for both 32-bit and 64-bit systems.[13]

In February 2020, Netflix stated that their Android app relied on dav1d to decode 10-bit AV1 content.[14]

In February 2023, Meta reported using dav1d to decode AV1 content for Facebook Reels and Instagram Reels in their iOS app.[15]

References

  1. ^ "dav1d - dav1d is an AV1 decoder - VideoLAN". www.videolan.org. Retrieved 2024-04-26.
  2. ^ "Introducing dav1d: a new AV1 decoder - Jean-Baptiste Kempf's Website". jbkempf.com. Retrieved 2024-04-26.
  3. ^ "LWN.net: RMS on the Ogg Vorbis license". lwn.net. Retrieved 2024-04-29.
  4. ^ Ozer, Jan (2018-08-31). "AV1: A First Look".{{cite web}}: CS1 maint: url-status (link)
  5. ^ "Diary Of An x264 Developer » Announcing the world's fastest VP8 decoder: ffvp8". web.archive.org. 2010-08-11. Retrieved 2024-05-03.
  6. ^ "The world's fastest VP9 decoder: ffvp9 | Ronald S. Bultje". Retrieved 2024-05-03.
  7. ^ Video Dev Days 2018: Dav1d: a fast new AV1 decoder. Retrieved 2024-04-29 – via www.youtube.com.
  8. ^ "First release of dav1d, the AV1 decoder - Jean-Baptiste Kempf's Website". jbkempf.com. Retrieved 2024-04-29.
  9. ^ "dav1d 0.7.1 - Jean-Baptiste Kempf's Website". jbkempf.com. Retrieved 2024-04-29.
  10. ^ Hoeven, Ewout ter (2019-10-10). "AV1 is ready for prime time Part 2: Decoding performance". Medium. Retrieved 2024-04-29.
  11. ^ "dav1d 0.9.1: a ton of asm - Jean-Baptiste Kempf's Website". jbkempf.com. Retrieved 2024-04-29.
  12. ^ Ronald S. Bultje - Low-level wizardry in dav1d. Retrieved 2024-05-03 – via www.youtube.com.
  13. ^ "Firefox brings you smooth video playback with the world's fastest AV1 decoder – Mozilla Hacks - the Web developer blog". Mozilla Hacks – the Web developer blog. Retrieved 2024-05-07.
  14. ^ Blog, Netflix Technology (2020-02-05). "Netflix Now Streaming AV1 on Android". Medium. Retrieved 2024-05-07.
  15. ^ "How Meta brought AV1 to Reels". Engineering at Meta. 2023-02-21. Retrieved 2024-05-07.
Retrieved from "https://en.wikipedia.org/w/index.php?title=Draft:Dav1d&oldid=1222713388"