Skip to main content

WebGpuPointsRenderer

GPU instanced point sprites (triangle-list quad per instance) with a small style storage buffer and a frame uniform block for map transform and viewport size.

Import​

import {
  WebGpuPointsRenderer,
  createWebGpuPointsRenderer,
  type WebGpuPointsRendererOptions,
} from "stratum-map";

Construction​

createWebGpuPointsRenderer(options)​

Async factory. Requests adapter/device (unless options.device provided), builds pipeline, allocates quad + uniform + style buffers, configures canvas context.

Options (WebGpuPointsRendererOptions)

FieldTypeDescription
canvasHTMLCanvasElementTarget WebGPU canvas
alphaModeGPUCanvasAlphaMode?Default premultiplied
deviceGPUDevice?Share device across layers
stylesreadonly PointStyle[]?Initial style table
gpuBufferPoolGpuBufferPool?Optional pooled vertex buffers

Key methods​

MethodSummary
setMapToClipMatrix(m: Float32Array)Column-major 4×4 map → clip
resize()Sync canvas context after DPR/size changes
setStyleTable(styles)Upload style table to GPU
setVertexBuffers(...) / setGeometry(...)Single-buffer layout
useChunkedGeometryLayout()Switch to per-chunk buffers
ingestTransferredWorkerChunk(...)Register chunk from worker
encodeRenderPass(pass, frame)Encode draws into caller’s pass
renderToCurrentTexture(frame)Convenience full-frame render
pickFeatureIdAtCanvasPixel(x, y)Async GPU pick (backing-store px)
destroy()Release GPU resources

Picking​

Enable and size the pick target on the renderer, then call pickFeatureIdAtCanvasPixel(x, y) with backing-store pixel coordinates. See GPU picking for the full flow.

Performance notes​

  • Prefer partial writeBuffer when only a subrange of vertices changes.
  • In chunked mode, avoid unbounded chunk keys — eviction reclaims GPU memory.
  • Pick target has a size cost; resize only when canvas changes.

Shader exports​

POINTS_WGSL and FRAME_UNIFORM_BYTE_LENGTH are exported for advanced embedding (custom pipelines).

See also​