SpawnDev.VoxelEngine 0.2.0-rc.4

SpawnDev.VoxelEngine

High-performance voxel engine built on SpawnDev.ILGPU. GPU-accelerated meshing, culling, LOD, terrain carving, physics, destruction, and VR rendering - from a single C# codebase targeting WebGPU, WebGL, Wasm, CUDA, OpenCL, and CPU.

Status: 0.2.0-rc.3 - release candidate. The block renderer and the GPU meshers (BlockRenderPipeline, ColumnMesher, HeightmapMesher) run AubsCraft's 3D viewer in production; the rest of the API is functional and consumed by Lost Spawns. Expect iteration during the RC cycle.

Write voxel engine code in C# and let SpawnDev.ILGPU handle the GPU backend. In the browser, WebGPU compute shaders power the entire pipeline. On desktop, CUDA and OpenCL take over automatically. The same kernels run everywhere.

Features

Core Pipeline

• Binary greedy meshing - GPU compute kernels that merge adjacent voxel faces into larger quads using bitwise operations. 80-90% vertex reduction vs naive meshing. 50-200 microseconds per chunk.
• GPU-driven culling pipeline - Frustum culling, cylindrical fog culling, and Sodium-style graph-based visibility. (Hi-Z occlusion is reserved for v2; graph + frustum handle 95%+ of voxel-world occlusion.)
• Vertex pooling - Fixed-size bucket allocator backed by a single GPU buffer. O(1) alloc/free. Priority eviction by distance and vertex cost.
• Single-buffer indirect draws - All draw commands in one GPUBuffer for Chrome's 300x D3D12 validation optimization.
• Adaptive LOD - Distance-based detail reduction via DelegateSpecialization kernels. One mesh kernel, multiple LOD behaviors, zero overhead.
• Device-adaptive quality - Queries actual GPU limits at init. Draw distance, vertex budget, LOD bias, and streaming rate all derived from hardware. Quest 3S gets appropriate budgets automatically.
• Reversed-Z depth - Zero Z-fighting at any distance with infinite far plane.
• Face masking - 6 orientation groups per chunk mesh. Skip back-facing groups at draw time for 36% reduction.

Textured Blocks

• BlockRenderPipeline - textured, tinted, lit, fogged blocks from PackedQuad buffers (the Minecraft-style look): opaque, plant (crossed planes, double-sided) and water (alpha blended) passes, alpha cutout, per-face brightness, sun + ambient + fog supplied by the caller, reversed-Z.
• A texture atlas read exactly - BlockTextureAtlas is one 2D texture, a grid of equal tiles; the shader fetches texels itself and wraps inside the tile, so a greedy-merged quad tiles its texture once per block, with no sampler, no seams and no layer limit (an array holds only 256 layers on WebGPU). Animated tiles are a per-tick timeline (SetAnimations).
• ColumnMesher - a whole block column (all its 16-high sections) meshed on the GPU in a handful of dispatches: class-aware (BlockClass: opaque, see-through leaves/glass, water, plants), exact across column borders, greedy-merged, with one tiny counter readback per column. A column can come straight from a JS typed array (CreateColumnBufferFromJS).
• Section placement by instance index - one quad buffer holds many sections; a draw's firstInstance selects the section origin, so drawing a section writes no uniform.
• QuadArena - a few large buffers that hold the quads of hundreds of meshes (ColumnMesher.Arena): a renderer binds a buffer once per pass instead of once per column, which is what a weak CPU spends its frame on. Each list is drawn from its own FirstQuad.

Beyond Meshing

• Terrain carving - TerrainCarveService with sphere/box/cylinder add/remove modes and ITerrainCarve interface. NMS-style sphere terraform shipped in Lost Spawns.
• Voxel physics - VoxelRaycast.CastWorld (DDA traversal returning RaycastHit with face + adjacent cell), VoxelCollision AABB-vs-world, VoxelSphereQuery, and StructuralIntegrity for cave-in / floating-block detection.
• Destruction - ExplosionKernels for radial damage propagation on the GPU.
• SDF mesher - SdfMeshPipeline with Dual Marching Cubes kernels and SDF noise generation for smooth-terrain styles alongside the cubic mesher.
• Terrain generation - HeightmapTerrainProvider and BiomeAssigner for procedural worlds.
• VR / WebXR - StereoRenderer, FoveatedRendering, and WebXRHelper for Quest 3S and other WebXR runtimes.
• Caching - Chunk persistence layer for save/load round-trips.
• Cross-platform - Same kernel code runs in browser (WebGPU/WebGL/Wasm) and desktop (CUDA, OpenCL, CPU) from one NuGet package.

Installation

dotnet add package SpawnDev.VoxelEngine --prerelease

Dependencies

• SpawnDev.ILGPU - GPU compute (WebGPU, WebGL, Wasm, CUDA, OpenCL, CPU)
• SpawnDev.SpawnJS - Browser interop for Blazor WebAssembly

Quick Start

using SpawnDev.SpawnJS;
using SpawnDev.VoxelEngine;
using SpawnDev.VoxelEngine.Carving;

var builder = WebAssemblyHostBuilder.CreateDefault(args);
builder.Services.AddSpawnJSRuntime();

// One-line registration with Lost Spawns defaults (HD realistic, VoxelSize=0.5).
// Use AddVoxelEngineAubsCraft() for blocky Minecraft-style defaults,
// or AddVoxelEngine(cfg => { ... }) to dial in your own values.
builder.Services.AddVoxelEngineLostSpawns();

await builder.Build().SpawnJSRunAsync();

// Inject the services you need:
public class GameLogic
{
    public GameLogic(
        ChunkManager chunks,
        CullingPipeline culling,
        IndirectDrawBuffer drawBuffer,
        BlockRegistry blocks)
    {
        // ... use them in your render loop
    }
}

Architecture

SpawnDev.VoxelEngine provides the GPU-accelerated infrastructure that voxel games need. It handles the hard parts - meshing, culling, buffer management, LOD, carving, physics, destruction, and adaptive quality - so game code can focus on gameplay.

Your Game Code
     |
SpawnDev.VoxelEngine  (meshing, culling, LOD, rendering, carving, physics, SDF, VR)
     |
SpawnDev.ILGPU        (GPU compute - WebGPU, WebGL, Wasm, CUDA, OpenCL, CPU)
     |
SpawnDev.SpawnJS      (browser interop)

Library Layout

Key ILGPU Features Used

Documentation

• Plans - Implementation roadmap
• Research - Optimization techniques and engine comparisons

License

MIT - see LICENSE.txt

Credits

SpawnDev.VoxelEngine is built upon SpawnDev.ILGPU and the excellent ILGPU library. We would like to thank Marcel Koester and the ILGPU contributors for their hard work in providing a high-performance IL-to-GPU compiler for the .NET ecosystem.

• ILGPU Project: https://github.com/m4rs-mt/ILGPU
• ILGPU Authors: Marcel Koester and the ILGPU contributors

The SpawnDev Crew

SpawnDev.VoxelEngine is developed collaboratively by TJ (Todd Tanner / @LostBeard) and a team of AI agents who contribute extensively to research, analysis, architecture design, and code development. This project represents a new model of human-AI collaboration in open source development.

• LostBeard (Todd Tanner) - Captain, library author, keeper of the vision.
• Riker (Claude CLI #1) - First Officer, implementation lead on consuming projects.
• Data (Claude CLI #2) - Operations Officer, deep-library work, test rigor, root-cause analysis. Authored the SpawnDev.VoxelEngine architecture and the optimization masterplan covering binary greedy meshing, GPU-driven culling, adaptive LOD, and Quest 3S adaptive quality.
• Tuvok (Claude CLI #3) - Security/Research Officer, design planning, documentation, code review. Drove the sub-word data type implementation in SpawnDev.ILGPU (v4.9.0) and built the eviction system that proved the architecture.
• Geordi (Claude CLI #4) - Chief Engineer, library internals, GPU kernels, backend work. Designed the 2-worker architecture (JS data worker + Blazor render worker) and built the binary WebSocket + OPFS cache + CopyFromJS data pipeline.
• Gemini (Google AI, in-browser) - Brainstorming partner. TJ's sounding board for cross-checking approaches.

These agents communicate through a shared DevComms system, coordinate tasks autonomously, review each other's work, and produce independent analyses that are compared for convergence - the same methodology used by any high-performing engineering team.

Resources

• SpawnDev.ILGPU - GPU compute for .NET (WebGPU, CUDA, OpenCL, CPU)
• SpawnDev.SpawnJS - Full JS interop for Blazor WebAssembly
• ILGPU Documentation
• WebGPU Specification
• GitHub Repository