Skip to main content

Deferred Lighting

deferredLighting is an experimental WebGPU-only fullscreen shader pass that resolves physically based opaque lighting from a GBuffer. Geometry writes surface properties once; the resolve then reconstructs view-space position from depth and evaluates one directional light plus up to 64 point lights from a storage buffer.

The pass stays intentionally composable. It does not traverse a scene, own materials, or allocate the G-buffer. Applications choose their geometry, material packing, light animation, and the effect passes that run before or after the resolve.

GitHub

Material attachment contract​

deferredLighting consumes the standard GBuffer depth and normal-roughness channels plus two named extra color attachments:

G-buffer outputSuggested formatMeaning
@location(0) scene colorrgba16floatBackground/fallback color when depth contains no opaque surface.
@location(1) normal-roughnessrgba8unormView-space normal encoded as normal * 0.5 + 0.5 in RGB; perceptual roughness in A.
@location(2) velocityrg16floatCurrent-minus-previous screen UV velocity for later temporal effects.
@location(3) baseColorMetallicrgba8unormLinear base color in RGB; metallic factor in A.
@location(4) emissiveOcclusionrgba8uintNormalized emissive color in RGB and ambient-occlusion factor in A, packed to 0–255 integers.
depth attachmentdepth24plusSampleable depth used to reconstruct view position.

The extra attachment names are conventions at the application boundary; GBuffer itself permits any names. The resolve bindings use baseColorMetallicTexture and emissiveOcclusionTexture. This layout keeps HDR scene color while totaling 32 color-attachment bytes per sample, so it runs on the portable WebGPU CORE limit without requesting a higher device limit.

Usage​

import {Buffer} from '@luma.gl/core';
import {ShaderPassRenderer} from '@luma.gl/engine';
import {
  createDeferredLightingShaderPassPipeline,
  GBuffer,
  makeDeferredPointLightBufferData,
  MAX_DEFERRED_POINT_LIGHTS
} from '@luma.gl/experimental';

const gBuffer = new GBuffer(device, {
  width,
  height,
  colorFormat: 'rgba16float',
  extraColorAttachments: [
    {name: 'baseColorMetallic', format: 'rgba8unorm'},
    {name: 'emissiveOcclusion', format: 'rgba8uint'}
  ]
});

const pointLights = device.createBuffer({
  data: makeDeferredPointLightBufferData(lights, MAX_DEFERRED_POINT_LIGHTS),
  usage: Buffer.STORAGE | Buffer.COPY_DST
});

const renderer = new ShaderPassRenderer(device, {
  shaderPasses: [createDeferredLightingShaderPassPipeline()]
});

renderer.renderToScreen({
  sourceTexture: gBuffer.colorTexture,
  bindings: {
    depthTexture: gBuffer.depthTexture,
    normalTexture: gBuffer.normalRoughnessTexture,
    baseColorMetallicTexture: gBuffer.getExtraColorTexture('baseColorMetallic'),
    emissiveOcclusionTexture: gBuffer.getExtraColorTexture('emissiveOcclusion'),
    pointLights
  },
  uniforms: {
    deferredLighting: {
      inverseProjectionMatrix,
      ambientColor: [0.03, 0.03, 0.05],
      directionalLightDirectionView,
      directionalLightColor: [1, 0.9, 0.8],
      directionalLightIntensity: 2.5,
      pointLightCount: lights.length
    }
  }
});

Point-light positions and the directional-light direction are view-space values. Update the storage buffer with pointLights.write(makeDeferredPointLightBufferData(...)) when lights move.

API​

deferredLighting​

The exported ShaderPass descriptor. It samples:

  • depthTexture
  • normalTexture
  • baseColorMetallicTexture
  • emissiveOcclusionTexture
  • pointLights

Its uniforms are inverseProjectionMatrix, ambientColor, directionalLightDirectionView, directionalLightColor, directionalLightIntensity, and pointLightCount.

createDeferredLightingShaderPassPipeline(): ShaderPassPipeline​

Returns a one-step pipeline that reads the current previous color and writes the lighting result back into previous. Put it before SSAO, reflections, temporal accumulation, bloom, and tone mapping unless a deliberate stylized ordering needs otherwise.

makeDeferredPointLightBufferData(lights, maxLightCount?): Float32Array​

Packs each light into two vec4f storage records:

position.xyz, range, color.rgb, intensity

The returned array is padded to maxLightCount, allowing one fixed-size GPU buffer to be reused across frames. MAX_DEFERRED_POINT_LIGHTS is 64 and matches the shader loop bound.

  • GBuffer owns the MRT attachments and standard depth, normal, and velocity bindings.
  • Clustered Lighting reuses the same material attachments with compute-built light lists for hundreds of local lights.
  • Shader Passes explains the ordered composable render stack.
  • Advanced Effects shows the broader scene-aware and temporal effect chain after opaque rendering.