Whether WgpuComputeCache.dispose has run.
How many WGSL modules have been compiled. Diagnostics and tests.
How many distinct compute pipelines have been created. Diagnostics and tests.
Records and submits one §82 dispatch.
Returns false — having recorded nothing — when the device surface lacks
the compute entry points (createComputePipeline, beginComputePass);
the caller turns that into UNSUPPORTED_GPU_FEATURE. Throws
INVALID_APPLICATION_STATE for a descriptor this module refuses: a
workgroup count that is not a non-negative integer, a disposed buffer in
bindings, or a call on a disposed cache.
Drops every pipeline, layout and module. Nothing is destroyed — WebGPU
compute pipelines and modules have no destroy() (§83, the render
cache's rule). Idempotent.
Per-device store of compute pipelines, layouts and modules (§82) — the compute half of
WgpuPipelineCache, kept separate so the frame path never links it: the renderer creates one lazily on the firstcompute()call (the lazy-subsystem precedent), and drops it whole on device loss and disposal (nothing here has adestroy(); dropping the maps is the release, the render cache's §83 reading).