Module gpu 

Vision compute device policy (programme A1).

Default builds use CPU oracles (ops/). Feature gpu will attach shared_gpu / Forge dispatch without inventing a second adapter.

Today: honest Cpu / Unavailable reporting so SR and CV can log device choice; GPU path lands when gpu feature wires Forge (A2).

Re-exports

pub use dispatch::avg_pool2d_dispatch;

pub use dispatch::conv2d_nchw_dispatch;

pub use dispatch::max_pool2d_dispatch;

pub use dispatch::resize_nearest_nchw_dispatch;

pub use dispatch::VisionComputeDevice;

pub use dispatch::VisionComputeReport;

pub use forge_bicubic::bicubic2d_cpu;

pub use forge_bicubic::bicubic2d_wgsl;

pub use forge_bicubic::try_resize_bicubic_shared_gpu;

pub use forge_bicubic::BICUBIC_ENTRY;

pub use forge_resize::try_resize_nearest_shared_gpu;

pub use policy::thermal_allows_gpu_tiles;

pub use policy::ThermalHint;

pub use policy::VisionVramBudget;

Modules

dispatch

Dispatch vision ops to CPU oracle (always) or GPU when feature lands (A1/A2).

forge_bicubic

Bicubic (Keys a=-0.5) NCHW f32 resize via Forge WGSL on the auxiliary GPU circuit.

forge_resize

B2 — nearest NCHW resize via Forge on the auxiliary GPU circuit when available.

policy

Thermal / VRAM policy stubs for vision (shared with SR plan Auto policy).