SplatCapture

class SplatCapture

import { SplatCapture } from '@driftengine/splats';

In depth

Allocated once at the final count and filled progressively, which is what makes a streaming load cost nothing extra: the GPU textures are sized for the whole capture at registration, each block is a sub-upload into storage that already exists, and the drawn count rises. Growing an array as blocks land would mean reallocating and re-uploading a capture's worth of texels several times during a load, which is more work than the load itself.

Every block is a sparse version of the whole capture, because the container writes them that way — see coarseFirstOrder. So a prefix of the arrived splats is a recognisable place at lower density rather than one finished corner, and the load opens on something worth looking at. That property lives in the writer; what lives here is that nothing is re-ordered or re-allocated on the way in.

The unarrived tail is zeroed, which puts it at the origin with no size. Nothing draws it: the caller sorts and draws ready splats and the sorter is told the same number.

Constructor

constructor(options: SplatCaptureOptions)

Properties

Accessors

Methods

append(records: Uint32Array, count: number): SplatAppend