Proxeno.Kiln 0.1.0

<p align="center"> <img src="docs/assets/hero.png" alt="Kiln" width="100%" /> </p>

<h1 align="center">Kiln</h1>

<p align="center"> A pure-managed, SIMD-accelerated <b>H.264 baseline-profile encoder</b> for .NET,<br/> built for real-time game streaming. </p>

<p align="center"> <a href="https://github.com/Proxeno/kiln/actions/workflows/ci.yml"><img src="https://github.com/Proxeno/kiln/actions/workflows/ci.yml/badge.svg" alt="CI" /></a> <a href="https://www.nuget.org/packages/Proxeno.Kiln"><img src="https://img.shields.io/nuget/v/Proxeno.Kiln" alt="NuGet version" /></a> <a href="https://www.nuget.org/packages/Proxeno.Kiln"><img src="https://img.shields.io/nuget/dt/Proxeno.Kiln" alt="NuGet downloads" /></a> <img src="https://img.shields.io/badge/license-Apache--2.0-e8912d" alt="Apache-2.0" /> <img src="https://img.shields.io/badge/.NET-10-e8912d" alt=".NET 10" /> </p>

Kiln is a pure-managed H.264 encoder for .NET / C# with no native dependencies — the entire codec is C# on .NET 10 using hardware intrinsics: NEON/AdvSimd on arm64, AVX2 and SSSE3 on x64, with a scalar fallback everywhere else. It targets real-time, low-latency use — game streaming, screen capture, and WebRTC/RTP pipelines — where a managed, dependency-free encoder beats bundling a native codec. Kernel selection happens at runtime; every SIMD path is covered by parity tests against the scalar reference, and CI runs the full suite on Linux, Windows and macOS so both architectures stay green.

Spec-cited and clean-licensed

Kiln is original work, written against the ITU-T H.264 (ISO/IEC 14496-10) specification. Every numeric table that originates in the spec carries its clause/table citation in the source (Table 9-4, §8.5.9, §9.2.1, …). It has no codec dependencies and links no copyleft code, so it embeds cleanly under Apache-2.0 in commercial products where GPL/LGPL codec linkage is a problem — which is exactly why it exists.

What it is

• Real-time first: constrained baseline profile, IDR + P-frames, CAVLC, Annex B output that streams straight into WebRTC or an RTP packetizer. Predictable per-frame latency over exhaustive search.
• A real encoder: Intra_4x4 / Intra_16x16 with RD mode selection, P-slice inter prediction with sub-pel SATD motion search, P_Skip, multi-slice frames (parallel slice encoding), multiple reference frames, in-loop deblocking, optional greedy trellis quantization, spatial adaptive QP, per-frame rate control.
• Operations-grade: Kiln.RateControl (low-latency rate controller with network feedback) and Kiln.Recovery (IDR budgeting / keyframe recovery policy) are public companion namespaces, production-used in a streaming server.
• Verified: 2,174 tests — spec-roundtrip decoding, SIMD/scalar parity on NEON and AVX2/SSSE3, golden-frame regression, PSNR fidelity floors, adversarial neighbour-availability sweeps, and smoke tests that decode every produced stream with an independent reference decoder as an oracle.

What it is not

Not an x264 competitor. No B-frames, no CABAC, no 8x8 transform, no interlace; 4:2:0 8-bit only; dimensions must be multiples of 16. If you need maximum compression at any CPU cost, use a full-profile encoder. If you need clean-licensed, dependency-free, low-latency encoding inside a .NET process, you are in the right place.

Quick start

using Kiln;

var encoder = new H264BaselineEncoder(1280, 720, new H264BaselineEncoderOptions
{
    QuantizationParameter = 28,
    KeyframeIntervalFrames = 120,
    SliceCount = 4,               // parallel slice encoding
});

var annexB = new byte[1280 * 720 * 2];

// Planar I420 input; u/v are half-resolution planes.
var written = encoder.EncodeFrame(y, u, v, strideY: 1280, strideUv: 640, annexB);
var wasIdr = encoder.LastFrameWasIdr;
// annexB[0..written] is a complete Annex B access unit (SPS/PPS included on IDR).

Options reference

Performance

Measured on Apple M5 Max (arm64, NEON/AdvSimd), .NET 10, BenchmarkDotNet, quiet machine — these are the committed perf-gate baseline numbers (perf/):

A ~2.2 ms steady-state P-frame at 720p on one slice leaves comfortable headroom for 60 fps game streaming; SliceCount > 1 parallelizes further. During extraction the motion-estimation path also got a measured targeted win: two structurally dead caches (0.0% hit rate under production instrumentation) were removed, making the textured-content sub-partition search ~5% faster (paired A/B, median of 6 rounds, faster in 6/6) with bit-identical output.

Perf discipline is part of the repo: bench/Kiln.Benchmarks (BenchmarkDotNet), scripts/h264-simd-capture-baseline.sh and scripts/h264-simd-perf-gate.sh gate changes against the committed baseline in perf/. See docs/perf-gate.md.

Experimental subsystems

Adaptation (resolution/fps ladders) and Queue (latest-frame dropping) ship inside the library, fully tested but not wired into the encoder — experimental, APIs may change or move without notice. See docs/architecture.md.

Installing

Published on nuget.org as Proxeno.Kiln. The package id is Proxeno.Kiln; the assembly and namespace stay Kiln, so code uses using Kiln;.

dotnet add package Proxeno.Kiln

Documentation

• docs/architecture.md — pipeline stages, SIMD kernel structure, subsystems
• docs/perf-gate.md — benchmark baseline + regression gate workflow
• CONTRIBUTING.md — build, test, and contribution rules
• SECURITY.md — reporting vulnerabilities

License

Apache-2.0 — see LICENSE. Copyright the Kiln contributors.