VideoLAN's dav2d: Ushering in the AV2 Era with High-Performance Decoding

The advent of new video codecs consistently pushes the boundaries of compression efficiency, enabling higher quality video at lower bitrates. The Alliance for Open Media (AOMedia) is at the forefront of this innovation with AV2, the successor to the widely adopted AV1 codec. VideoLAN, known for its VLC media player and its high-performance dav1d AV1 decoder, is now introducing dav2d, a new software decoder designed to be the fastest and most efficient for AV2 across all platforms. This development is crucial as it lays the groundwork for widespread AV2 adoption, much like dav1d accelerated the rollout of AV1.

The initial announcement of dav2d on Hacker News, while briefly hampered by a server error on the project's GitLab instance, quickly garnered significant attention. The community's discussion highlighted both the technical anticipation for AV2 and the strategic importance of a robust software decoder from the outset.

What is dav2d?

dav2d is VideoLAN's new video decoder implementation specifically for the AV2 video coding specification. Its core mission, as described in the project's brief, is to be "the fastest AV2 decoder on all platforms," emphasizing its design for being "small, portable and very fast." This ambition is not new for VideoLAN; it mirrors the success of dav1d, their AV1 decoder, which played a pivotal role in making AV1 practical for widespread use before dedicated hardware decoding became common.

One commenter provided essential context:

Project description: dav2d is the fastest AV2 decoder on all platforms :) Targeted to be small, portable and very fast. If you're out of the loop like me: AV2 is the next-generation video coding specification from the Alliance for Open Media (AOMedia). Building on the foundation of AV1, AV2 is engineered to provide superior compression efficiency, enabling high-quality video delivery at significantly lower bitrates. It is optimized for the evolving demands of streaming, broadcasting, and real-time video conferencing. - from https://av2.aomedia.org/

The Promise of AV2

AV2 is positioned as a significant leap forward from AV1. Building on its predecessor's foundation, AV2 is engineered to deliver superior compression efficiency, which translates to high-quality video at significantly lower bitrates. This is particularly vital for modern demands in streaming, broadcasting, and real-time video conferencing, where bandwidth is often a constraint.

Early insights suggest that AV2 could offer substantial improvements:

AV2 video codec delivers 30% lower bitrate than AV1, final spec due in late 2025 (videocardz.com)

This 30% reduction, if realized, would be a game-changer for content delivery and consumption, making high-definition and even ultra-high-definition content more accessible globally. However, as one comment noted, the full benefits will only be seen once robust encoders become available, a process that historically has taken time, as evidenced by the development of SVT-AV1 for AV1.

Technical Implementation and Design Philosophy

Following the successful blueprint of dav1d, dav2d is primarily written in C, with critical performance paths leveraging hand-coded assembly. This approach allows for meticulous optimization, squeezing out every last bit of performance from various hardware architectures.

Mostly ASM for performance critical paths is a pattern that never gets old. The VideoLAN team did the same with dav1d and it paid off. Curious how much of dav2d ends up staying C as AV2 matures.

This choice, while performance-driven, sparked a discussion regarding memory safety in codecs:

We must not continue to develop media codecs in memory unsafe languages. Small, auditable sections can opt-out perhaps, but choosing default-unsafe for this type of software is close to professional negligence.

While memory safety is a valid concern, especially in security-critical software, the dav1d precedent demonstrates that highly optimized C and assembly can achieve unparalleled performance, which is often a paramount requirement for video decoding. The VideoLAN team's expertise in this domain suggests a careful balance between performance and code quality.

Impact on Adoption and Hardware Development

The existence of a high-quality, performant software decoder like dav2d from the early stages of AV2's lifecycle is expected to have a profound impact on its adoption. As one commenter eloquently put it, it creates a "reverse chicken and egg scenario":

Dav1d was the surprisingly fast assembly implementation of AV1 decoding. Even for something in hand-coded platform-specific-assembly I think the general impression was that they'd done amazing work to really chase down every last bit of potential performance. It didn't initially exist when AV1 was first rolled out and its arrival was a step change in powering adoption on devices without hardware decoding. Dav2d is likely to play a similar role, but it exists from the start of AV2 and can build on the work of dav1d, so should have an even bigger effect. In a weird reverse chicken and egg scenario, having really good software decode that can be deployed will spur on hardware development and adoption due to network effects.

This means that the availability of dav2d can accelerate the deployment of AV2 content, which in turn creates demand for hardware manufacturers to include AV2 decoding capabilities in their chips, ultimately driving broader ecosystem adoption.

Broader Context and Future Considerations

The discussion also touched on several other aspects of the video codec landscape:

• Naming Convention: The name dav2d follows a recursive pattern, similar to "GNU's Not Unix," where Dav1d is an AV1 Decoder, and Dav2d is an AV2 Decoder. This was clarified amidst some initial confusion regarding popular YouTubers or other public figures.
• Ecosystem Challenges: The open-source codec space is not without its challenges, including patent disputes. One comment referenced ongoing concerns about entities like Dolby potentially attacking the AV1 ecosystem, highlighting the need for organized defense by AOMedia.
• Future ISAs: A forward-looking perspective questioned the continued focus on "legacy ISAs like x86 or ARM" when "RVA23 boards shipping this month" suggest that RISC-V could be dominant by the time AV2 is fully deployed. This points to the need for future-proofing and broader architectural support.
• User Experience: Some comments reflected on the general state of the internet, noting the increasing friction from bot checks, cookie consent, and Cloudflare challenges, which can detract from the user experience when trying to access technical resources.

As dav2d matures alongside the AV2 specification, its role in enabling the next generation of high-efficiency video will be critical. The project's early release and VideoLAN's proven track record suggest a promising future for AV2 adoption across various platforms and applications.