Internal — obtain instances through WebgpuRenderer.createComputeBuffer.
ReadonlybufferThe device allocation — a backend seam, not an application surface: the dispatch and readback paths read it, and nothing else should.
ReadonlybyteAllocation size in bytes.
ReadonlyisThe ComputeBuffer brand — one property load discriminates.
Whether WgpuComputeBuffer.dispose has run.
A storage buffer the application owns (§82, §83) — created by
WebgpuRenderer.createComputeBuffer, disposed by whoever created it.Allocated
STORAGE | COPY_DST | COPY_SRC, so every compute buffer can be written (writeComputeBuffer), bound to a kernel, and read back (readComputeBufferBytes) — three usage bits against a per-usage-tier split that would complicate §82's tiny surface for no measured saving. (wgpu-particle-simulation.tsmints one withVERTEXadded, for a buffer the §36 draw also binds as an instance stream — its header owns the deviation.)The one implementor of
@fourjs/render's structural ComputeBuffer (the Q3 promotion): the brand plus the §83 trio are the cross-seam face; WgpuComputeBuffer.buffer stays this backend's own.