Struct WebGpuBackend 

pub struct WebGpuBackend { /* private fields */ }

CubeCL WebGPU tensor backend.

Examples

use tenferro_gpu::webgpu::WebGpuBackend;

let _ctor: fn(usize) -> tenferro_tensor::Result<WebGpuBackend> = WebGpuBackend::new;

The backend is not an operation route: the operations live on [WebGpuExecSession], so the owner does not implement the operation traits.

fn requires_elementwise<B: tenferro_tensor::TensorElementwise>() {}
requires_elementwise::<tenferro_gpu::webgpu::WebGpuBackend>();

Implementations

impl WebGpuBackend

pub fn new(device_ordinal: usize) -> Result<Self>

Initialize a WebGPU backend for a discrete GPU ordinal.

Examples

use tenferro_gpu::webgpu::WebGpuBackend;

let _ctor: fn(usize) -> tenferro_tensor::Result<WebGpuBackend> = WebGpuBackend::new;

Errors

Returns crate::Error::RuntimeState when no adapter/device is available, or crate::Error::BackendSource when CubeCL initialization fails.

pub fn new_default() -> Result<Self>

Initialize a WebGPU backend using CubeCL’s default adapter selection.

Examples

use tenferro_gpu::webgpu::WebGpuBackend;

let _ctor: fn() -> tenferro_tensor::Result<WebGpuBackend> = WebGpuBackend::new_default;

Errors

Returns crate::Error::RuntimeState when default adapter selection is unavailable, or crate::Error::BackendSource when initialization fails.

pub fn from_runtime(runtime: WebGpuRuntime) -> Self

Build a WebGPU backend from an initialized runtime.

Examples

use tenferro_gpu::{webgpu::WebGpuBackend, webgpu::WebGpuRuntime};

let _from_runtime: fn(WebGpuRuntime) -> WebGpuBackend = WebGpuBackend::from_runtime;

pub fn runtime(&self) -> &WebGpuRuntime

Return this backend’s WebGPU runtime.

Examples

use tenferro_gpu::{webgpu::WebGpuBackend, webgpu::WebGpuRuntime};

let _runtime: fn(&WebGpuBackend) -> &WebGpuRuntime = WebGpuBackend::runtime;

pub fn runtime_identity(&self) -> WebGpuRuntimeIdentity

Return the opaque identity of this exact executable backend instance.

Clones of a backend return the same identity. Independently initialized backends return different identities even when they target the same WebGPU device ordinal. This also covers Apple-backed WebGPU runtimes.

Examples

use tenferro_gpu::webgpu::WebGpuBackend;

let _identity = WebGpuBackend::runtime_identity;

pub fn synchronize(&self) -> Result<()>

Block until queued WebGPU work completes.

Examples

use tenferro_gpu::webgpu::WebGpuBackend;

let _sync: fn(&WebGpuBackend) -> tenferro_tensor::Result<()> = WebGpuBackend::synchronize;

Errors

Returns crate::Error::BackendSource when queue flush or synchronization fails, or crate::Error::RuntimeState when the runtime has lost its device state.

Trait Implementations

impl BackendRuntimeCache for WebGpuBackend

impl BackendSessionHost for WebGpuBackend

fn with_backend_session<R: Send>( &mut self, f: impl FnOnce(&mut dyn BackendSession) -> R + Send, ) -> Result<R, SessionEntryError>

Open one backend session and run f inside it.

impl Clone for WebGpuBackend

fn clone(&self) -> WebGpuBackend

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for WebGpuBackend

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl DotGeneralPreparation for WebGpuBackend

fn prepare( &self, request: DotGeneralPrepareRequest<'_>, ) -> Result<PrepareCapability, PrepareError>

Prepare one dot-general operation.

impl TensorBackend for WebGpuBackend

impl TensorDeviceTransfer for WebGpuBackend

fn download_to_host(&mut self, tensor: TensorRead<'_>) -> Result<Tensor>

Explicitly copy a provider-owned read target into host storage.

fn upload_host_tensor(&mut self, tensor: TensorRead<'_>) -> Result<Tensor>

Explicitly copy a host read target into provider storage.