Function cpu_kernel_tier 

pub fn cpu_kernel_tier() -> KernelTier

Best CPU kernel tier for the current machine, detected once and cached.

Unlike KernelDispatcher::auto_detect, this never selects the GPU tier and never logs, so it is cheap enough to call on every kernel invocation. It is used by the standard-quant (Q4_0 / Q8_0) free-function entry points to build a lightweight CPU dispatcher per call without repeating feature detection.