Struct CachedDispatcher 

pub struct CachedDispatcher { /* private fields */ }

Cached kernel dispatcher — singleton per process.

Returns Arc<dyn QuantKernel> instead of Box<dyn QuantKernel>, allowing zero-allocation kernel lookup after the first call for each type.

Implementations

impl CachedDispatcher

pub fn new() -> Self

Create a new cached dispatcher with runtime CPU feature detection.

pub fn get_kernel( &self, tensor_type: GgufTensorType, ) -> QuantResult<Arc<dyn QuantKernel>>

Get or create a cached kernel for the given tensor type.

The first call for each type allocates; subsequent calls return a clone of the Arc (cheap reference count bump).

pub fn capabilities(&self) -> &SimdCapabilities

Access the underlying capabilities.

pub fn is_supported(&self, tensor_type: GgufTensorType) -> bool

Check if a type is supported (delegates to inner).

pub fn supported_types(&self) -> Vec<GgufTensorType>

List supported types (delegates to inner).

Trait Implementations

impl Default for CachedDispatcher

fn default() -> Self

Returns the “default value” for a type.