Module webgpu 

CubeCL WebGPU provider runtime and backend skeleton.

The WebGPU/Metal backend is experimental and implements a narrow operation subset: explicit transfers, F32/C32 dot_general (and the einsum paths that lower to it), F32/I32 transpose and compaction, and the Apple Metal FFT. Elementwise math, reductions, reshape/broadcast, indexing and linear algebra return typed unsupported errors instead of falling back to CPU. The devices-and-gpu guide compares this subset with CUDA.

Structs

AppleContext

Explicit Apple shared-allocation context for CPU and Metal backends.

AppleTransferStats

Explicit host/device transfer counters for one AppleContext.

WebGpuBackend

CubeCL WebGPU tensor backend.

WebGpuRuntime

CubeCL WebGPU runtime wrapper.

WebGpuRuntimeIdentity

Opaque identity of one exact WebGPU runtime instance.

Functions

download_webgpu_tensor

Download a CubeCL-managed WebGPU tensor back to host memory.

upload_webgpu_tensor

Upload a host tensor into a CubeCL-managed WebGPU allocation.

webgpu_available

Returns true if a WebGPU adapter is available for CubeCL.

webgpu_runtime_engine_id

Return the canonical WebGPU runtime engine identifier.

webgpu_runtime_engine_registration

Build a runtime engine registration for a WebGpuBackend.

webgpu_runtime_engine_registration_with_id

Build a WebGPU runtime engine registration with a caller-selected engine ID.

webgpu_runtime_hardware_class

Return the canonical WebGPU runtime hardware class.