nightshade-renderer

A GPU-driven wgpu renderer with a built-in frame graph. It renders one owned input structure per frame and never sees the caller's scene representation, so any engine or tool that can fill out RenderInputs can drive it.

This is the renderer behind the nightshade engine. It runs on DX12, Metal, Vulkan, and WebGPU.

[dependencies]

nightshade-renderer = "0.52.0"

Design

• One input contract. The caller composes a RenderInputs value each frame: the scene snapshot, materials, mesh and texture caches, per-camera snapshots, settings, and drained render commands. The renderer consumes it and hands world-facing writes back through FrameOutputs. Large state moves in and out rather than being cloned.
• GPU-driven drawing. Object transforms, materials, and instance data live in persistent per-world GPU buffers updated by deltas. Frustum and hi-z occlusion culling, LOD selection, batch table construction, and indirect draw argument population run in compute. Draw submission is multi-draw indirect, with the draw count itself sourced from the GPU on hardware with MULTI_DRAW_INDIRECT_COUNT. Bindless material textures are used where the device supports them.
• Composable passes. The built-in pipeline is a set of PassNode<RenderInputs> implementations on a rendergraph: depth prepass, cascaded and atlas shadows, clustered forward PBR meshes and skinned meshes, sky and IBL, water, particles, decals, SSAO, SSGI, SSR, bloom, TAA, depth of field, post-processing, text, retained UI, and debug overlays. Callers add their own passes to the same graph.

Driving it

The default frame driver is wgpu::frame::render_frame(renderer, &mut inputs, &mut outputs). It owns the whole per-frame sequence: render command drain, IBL capture, dirty-state handoff to the passes, surface acquisition with occlusion and loss handling, per-camera dispatch with viewport caching, UI composite, and present.

A host drives it with three steps per frame:

1. Compose a RenderInputs. Persistent state (scene: RendererState, the mesh and texture caches, materials) lives on your side and moves in by value each frame; keep it current with deltas rather than rebuilding it. Frame data (frame: FrameInputs) carries the cameras to dispatch, drained commands, dirty spheres, and debug lines for this frame.
2. Call render_frame. On a skipped frame (occluded or lost surface) the unconsumed dirty state stays in the inputs.
3. Apply FrameOutputs (viewport texture sizes, cleared force-render flags, frame_executed) and take the persistent state back out of the inputs for the next frame.

RenderInputs deliberately has no Default: every field is load-bearing and the move-in/move-out cycle is the contract. Build one compose function against the struct and the compiler walks you through the fields.

What it renders

• Clustered forward PBR with the metallic-roughness workflow and the glTF KHR material extension set
• Cascaded shadow maps for the sun plus a shelf-packed atlas for spotlight and area-light shadows
• HDR environment maps and procedural atmospheres with prefiltered IBL, including day and night snapshot blending
• Skeletal animation and morph targets through a GPU skinning path
• Order-independent transparency, screen-space ambient occlusion, global illumination, and reflections
• Signed distance field text from a dynamic glyph atlas
• Optional grass and terrain passes behind feature flags

Using the frame graph alone

The renderer is built on its own frame graph, exposed as the rendergraph module. To use just the graph without the renderer:

[dependencies]

nightshade-renderer = { version = "0.52.0", default-features = false, features = ["rendergraph"] }

Declare passes and the resources they read and write, compile once, execute every frame. The graph is generic over an inputs type C that you define; passes implement PassNode<C> and receive &C during prepare and execute. The graph handles pass ordering from declared dependencies, transient resource allocation and aliasing, load and store op optimization, dead pass culling, per-pass enable toggles, external per-frame resources like the swapchain, and execute phases so cached-viewport frames can run a cheap compose-only subset.

use nightshade_renderer::rendergraph::{render_graph_new, render_graph_add_pass, render_graph_compile, render_graph_execute};

let mut graph = render_graph_new();

let depth = render_graph_add_depth_texture(&mut graph, "depth")
    .size(1920, 1080)
    .clear_depth(0.0)
    .transient();

let scene_color = render_graph_add_color_texture(&mut graph, "scene_color")
    .format(wgpu::TextureFormat::Rgba16Float)
    .size(1920, 1080)
    .clear_color(wgpu::Color::BLACK)
    .transient();

let swapchain = render_graph_add_color_texture(&mut graph, "swapchain")
    .format(wgpu::TextureFormat::Bgra8Unorm)
    .external();

render_graph_add_pass(&mut graph, Box::new(mesh_pass), &[
    ("color", scene_color),
    ("depth", depth),
])?;

render_graph_add_pass(&mut graph, Box::new(post_process_pass), &[
    ("input", scene_color),
    ("output", swapchain),
])?;

render_graph_compile(&mut graph)?;

// Per frame: register the swapchain texture, then execute.
render_graph_set_external_texture(&mut graph, swapchain, Some(texture), view, width, height);
let command_buffers = render_graph_execute(&mut graph, &device, &queue, &inputs)?;
queue.submit(command_buffers);

Features

default = ["wgpu"]

License

Dual-licensed under MIT or Apache-2.0, at your option.