Update libavif (v0.12-main) and AOM (3.6) for improved compression and speed

[andrerom]

Noticed there are several newer AOM and libavif releases where out that have focused a lot on compression quality and speed. There is no libavif tag yet using AOM 3.6 so used latests commit here.

---

Quick comparison locally:

• Default settings (Quality: 30, Effort: 4): dev (3.1) at 22kb vs update_libavif_aom (3.6) 14.8kb, similar perceived compression artifacts on edges of hearth (305% zoom)
• Higher settings (Quality: 40, Effort: 7): Artifacts goes away on 3.6 build (20.4kb) but not yet on 3.1 build (19kb)
• Speed: 3.6 build seem maybe 5-10% faster on all mentioned quality/effort levels.

Credit: https://unsplash.com/photos/PMxT0XtQ--A

• Default settings (Quality: 30, Effort: 4): dev (3.1) at 425kb vs update_libavif_aom (3.6) 438kb, similar perceived smoothing artifacts on skin (when zoomed)
• Higher settings (Quality: 40, Effort: 7): Smoothing effect goes away on 3.6 build (714kb) but not yet on 3.1 build (644kb)
• Speed: 3.6 build seem maybe 10-20% faster on all mentioned quality/effort levels.

---

Result is a bit of a mixed bag considering building this took 3 hours (on M1). Was hoping for both quality, size and speed improvements across the board, but 2 out of 3 in most cases at least.

x86-64 users might see better speed numbers as there are a fair bit of SSE (AOM 3.3) and AVX (AOM 3.5) optimizations recently in AOM (assuming WASM SIMD is used here for AVIF).

[google-cla[bot]]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[andrerom]

Since Google bot does not update it's comment with this: CLA has been signed, either click link above and click "Refresh", or directly the new invocation url.

[jakearchibald]

Thanks for this! I'll get the AVIF team to take a look, as I'd also like to know if there are new features we could be exposing.

[andrerom]

Great! And yes, might be there are incoming changes in libavif worth waiting for, and/or libavif/AOM features Squoosh could easily take advantage of for more speed/quality/features.

[jakearchibald]

I'm not so bothered about waiting, more that there may have been features already in this new version that we could expose. I also want them to check our glue code is still the best way to interface with the library. I've seen some colour shifting occasionally that makes me worry that we're doing something incorrect with colour spaces, but I've also seen others say that colour shifting is a general AVIF issue.

[jakearchibald]

I've updated things for the latest AVIF recommended internally. Let's land this! Thanks for the help @andrerom