dav2d Takes the Helm for AV2 Decoding – A Community‑First Approach

Why a New Decoder Matters Now

The AV2 specification, the latest royalty‑free codec from the Alliance for Open Media, has just been released. Early benchmarks suggest ~25 % better compression than AV1, but they also reveal that software decoding is roughly five times more demanding. On today’s CPUs, real‑time playback would be out of reach without aggressive, architecture‑specific optimization.

In the AV1 era, the VideoLAN community answered a similar gap with dav1d, a lean, fast software decoder that quickly became the de‑facto reference implementation. The same pattern is repeating: a codec arrives, hardware lags, and developers need a production‑ready software fallback. dav2d is positioned to fill that gap for AV2.

Evidence of Momentum

• Feature‑complete AVM v15 core: The repository already supports 8‑bit and 10‑bit streams, handling bitstream parsing, entropy decoding, intra/inter prediction, transforms, CDEF, Wiener filtering, and film‑grain synthesis.
• Cross‑architecture work in progress:
  • x86: AVX2 paths for inverse transforms and deblocking.
  • ARM (AArch64): NEON implementations for entropy, SAD, intra prediction, palette prediction, and motion compensation.
  • Early RISC‑V support is being re‑enabled.
• Checkasm integration: The same benchmarking framework that helped dav1d mature now validates dav2d’s SIMD kernels from day one, accelerating safe optimization.
• Open development: All code, issues, and merge requests live publicly on VideoLAN’s GitLab instance, encouraging contributions and transparent conformance testing.

These signals show a project that is not merely an announcement but a working decoder already capable of decoding AV2 streams, albeit with performance still being tuned.

Counter‑Perspectives and Open Questions

While the community enthusiasm is palpable, several concerns temper the optimism:

1. Hardware acceleration timeline – Major silicon vendors have hinted at AV2 decode blocks in upcoming GPUs and dedicated video ASICs, but concrete roadmaps remain vague. If hardware arrives sooner than expected, the window for a software‑only solution may shrink, potentially limiting dav2d’s long‑term relevance.
2. Complexity vs. portability – AV2’s five‑fold increase in decode complexity forces heavy SIMD use. Maintaining a clean, portable C fallback while still delivering real‑time performance on low‑end devices could become a maintenance burden.
3. Ecosystem readiness – Browsers and operating systems have already integrated dav1d, but they will need to evaluate dav2d’s API stability, licensing compatibility, and conformance scores before committing. Early adopters may face integration hurdles if the decoder’s public API evolves rapidly.
4. Testing depth – The specification is fresh, and edge‑case bitstreams are still being discovered. Ensuring exhaustive conformance across the myriad of real‑world content types will require a large, community‑driven test suite, something that takes months to mature.

Looking Ahead

The roadmap outlined by the VideoLAN team includes:

• Continued conformance testing against the official AV2 spec.
• Expansion of SIMD coverage on x86 (AVX‑512) and ARM (SVE2).
• Memory‑usage reductions and improved threading models to better exploit multi‑core CPUs.
• Broader high‑bit‑depth (12‑bit) support, which is critical for HDR workflows.

If these milestones are met, dav2d could replicate dav1d’s trajectory: becoming the reference implementation for AV2, easing integration for browsers, media players, and even operating‑system video stacks.

Community Sentiment

The announcement has sparked a mix of excitement and caution on forums such as the VideoLAN mailing list and the AOM Discord channel. Many developers praise the early start—"Having a decoder now lets us experiment with AV2 content before hardware arrives"—while others warn against over‑reliance on software performance: "We need to keep hardware in the loop, otherwise we risk building a solution that becomes obsolete quickly."

Bottom Line

dav2d embodies a pragmatic response to the AV2 challenge: deliver a fast, open‑source decoder now, learn from the dav1d experience, and lay the groundwork for a robust AV2 ecosystem. Its success will hinge on how quickly the project can close the performance gap, prove its conformance, and gain trust from the broader media community.

---

Resources

• AV2 specification: https://av2.aomedia.org/
• dav2d repository: https://code.videolan.org/videolan/dav2d
• dav1d history (for context): https://code.videolan.org/videolan/dav1d