Module wgpu 

WebGPU backend: device context, buffer management, pipeline cache, and kernel dispatch. Production backend of Forge; the CPU backend is the numerical reference.

Modules

ops

GPU op wrappers: bind tensors, compute grid sizes, dispatch WGSL kernels. Semantics mirror backend::cpu exactly; see the shader sources for the kernel-side contracts.

Structs

DispatchScope

An open recording scope: every dispatch made while one is alive goes into a single command encoder and a single queue.submit.

PooledBuffer

A storage buffer that returns to its context’s free list when dropped rather than being destroyed.

Stats

Cumulative device counters, read with WgpuContext::stats.

WgpuContext

Owns the wgpu device/queue and a cache of compiled compute pipelines.

Constants

OFFSET_ALIGN_BYTES

Storage-buffer offsets must respect this alignment when creating views.

Functions

linear_grid

Split a linear element count into a (x, y, 1) workgroup grid of 256-thread groups, respecting the 65535 per-dimension dispatch limit.