Struct ComputePolicy 

pub struct ComputePolicy { /* private fields */ }

The shared compute policy: a measured per-class matrix plus the host’s precision/VRAM envelope. Built once at startup; select is the hot, O(1) call.

Implementations

impl ComputePolicy

pub fn from_class_matrix( matrix: ClassMatrix, budget: PowerThermalBudget, host: HostCapabilities, ) -> Self

Build from an already-measured per-class matrix (e.g. one loaded from the passport, or a synthetic one in tests — no GPU required).

pub fn probe( registry: &BackendRegistry, panel: &KernelPanel, budget: PowerThermalBudget, host: HostCapabilities, ) -> Self

Probe the registry once (heavy) and build the policy. Call at startup; cache the matrix in the passport to avoid re-probing every boot.

pub fn matrix(&self) -> &ClassMatrix

The measured per-class matrix (for inspection / passport caching).

pub fn select(&self, class: KernelClass, problem_bytes: u64) -> Plan

Select the execution plan for a class kernel touching problem_bytes of data. O(1), zero-heap, never fails: returns a CPU plan when no accelerator wins, when no GPU was probed, or when the measured GPU win is within noise of CPU for a class that is not typically GPU-amenable (the §13 tie-break — a measured GPU win on an amenable class is always honoured).