rusty_h264-accel

The acceleration crate — and the only crate in the workspace that contains unsafe. It vendors Cisco openh264's BSD-2 x86-64 SIMD assembly kernels, assembles them with nasm, and exposes them behind safe Rust wrappers so the rest of rusty_h264 can stay #![forbid(unsafe_code)] while still running at SIMD speed.

Part of Remade With Rust by Mata Network.

The design rule

The codec has one unsafe boundary, and this is it.

rusty_h264-common / -encoder / -decoder     #![forbid(unsafe_code)]
        │  call only safe wrappers
        ▼
rusty_h264-accel                            the FFI + `unsafe` quarantine
        │  extern "C"
        ▼
vendored openh264 .asm  →  nasm  →  .obj

Every kernel here has a pure-Rust scalar twin in rusty_h264-common that acts as its oracle: satd_asm_compare / satd_avg_compare tests assert the asm and the scalar path agree, and every speed brick is gated byte-identical on full bitstreams. Turn the feature off and you get 100% safe, portable Rust with identical output — just slower.

The same boundary is the intended place to plug in your own kernels: custom intrinsics, hand-written assembly, or a pure-Rust implementation tuned for your micro-architecture. The safe core never sees unsafe and never needs to change.

What's vendored

The .asm files under vendor/ are copied verbatim from Cisco openh264 (codec/{common,encoder,decoder}/…/x86/), which is BSD-2-Clause — © 2013 Cisco Systems, see vendor/LICENSE.openh264. Vendoring makes this crate self-contained: no external openh264 checkout is needed, only nasm.

Kernel families wrapped today:

Area	Kernels
Cost metrics	`WelsSampleSatd{4x4,8x8,16x8,8x16,16x16}_sse2`, `WelsSampleSad{16x16,16x8,8x16}_sse2`
Transforms	`WelsDctFourT4_sse2`, `WelsQuantFour4x4_sse2`, inverse DCT + reconstruct
Motion compensation	`McHorVer20/02/22` (half-pel horizontal, vertical, centre), chroma MC width-8
Deblocking	luma & chroma `Lt4`/`Eq4`, vertical & horizontal, incl. the H↔V transposes
Intra prediction	`WelsI16x16LumaPred{V,H,Dc,Plane}_sse2`, `WelsIChromaPred{V,Plane}_sse2`

Plus satd_avg, satd_16x16_x4 and sad_16x16_x4 — custom fused kernels written for this project (not openh264's), which compute a bi-prediction average and its SATD in one pass, and batch four candidate costs per call for the motion search.

Platform behaviour

Situation	What happens
x86-64 + `nasm` on `PATH`	Kernels are assembled and linked. Full speed.
x86-64, no `nasm`	The build script skips assembly with a warning and the crate still compiles as a lib — so it stays publishable and docs.rs-buildable. Enabling `asm` in a binary then surfaces a clear link error rather than a build-script panic.
Any other architecture (e.g. arm64 macOS)	The whole module is gated on `#![cfg(target_arch = "x86_64")]`: the crate compiles to an empty lib, `nasm` is never invoked, no x86 objects are linked, and the consumer crates fall back to the pure-Rust scalar path via the internal `accel` cfg. A downstream default-features build works on Apple Silicon unchanged.

Environment overrides: NASM (path to the assembler) and OPENH264_DIR (build against a live openh264 tree instead of vendor/, for development).

You probably don't depend on this directly

Enable it through the facade instead:

# SIMD on (the default) — needs nasm:

rusty_h264 = "0.3"



# Pure safe Rust, no nasm, no unsafe anywhere:

rusty_h264 = { version = "0.3", default-features = false }

The asm feature on rusty_h264, rusty_h264-encoder, rusty_h264-decoder and rusty_h264-common is what pulls this crate in.

Where this sits

Crate	Role
`rusty_h264`	the public, safe facade API — depend on this
`rusty_h264-common`	bitstream I/O, transforms, prediction, MC, deblock
`rusty_h264-encoder`	the encode pipeline
`rusty_h264-decoder`	the decode pipeline
`rusty_h264-accel`	← you are here — the SIMD asm + the one `unsafe` boundary

The Remade With Rust ecosystem

Remade With Rust is an initiative by Mata Network to rebuild essential C and C++ tools in Rust — for the memory safety, the predictable performance, and the freedom of a permissive license. Each project is a reimplementation, not a fork: same wire protocols and file formats, new code you can actually depend on. No copyleft. No surprises.

Project	What it is
🎬 remade_ffmpeg_rs	Our FFmpeg alternative. Drop-in `ffmpeg` and `ffprobe` binaries — demux → decode → filter → encode → mux, rebuilt as composable Rust crates with zero GPL/LGPL. Apache-2.0. `rusty_h264` is its H.264 codec.
🧠 FFAI	*Our sister project: media for* AI. "The AI media toolkit, remade with rust." Embedded ASR + TTS (Mercury), OCR (Carmenta) and vision-language captioning (Argus**) behind an ffmpeg-style, swap-by-name architecture — no Python, no CUDA. MIT OR Apache-2.0.
🌐 Mata Network	The home page. "Stop sacrificing your privacy for convenience." Sovereign, self-hostable privacy infrastructure — wallet & identity, password manager, contact manager, and a browser extension that stops information leaking as you browse. Remade With Rust is its open-source arm.

→ All projects: github.com/Remade-With-Rust

rusty_h264-accel 0.5.1