Struct TunableSet

pub struct TunableSet<K, Inputs, Output>
where
    K: AutotuneKey,
    Inputs: Send + 'static,
    Output: 'static,
{ /* private fields */ }

Groups operations of the same type for autotune

Implementations

impl<K, Inputs, Output> TunableSet<K, Inputs, Output>

where K: AutotuneKey, Inputs: Clone + Send + 'static, Output: 'static,

pub fn len(&self) -> usize

The number of tunables in the set.

pub fn is_empty(&self) -> bool

If the tunable set is empty.

pub fn new<KMarker, IMarker>( key_gen: impl IntoKeyGenerator<K, Inputs, KMarker>, input_gen: impl IntoInputGenerator<K, Inputs, IMarker>, ) -> TunableSet<K, Inputs, Output>

Create a tunable set from a key generator and an input generator

pub fn with_tunable<Marker>( self, tunable: impl IntoTunable<Inputs, Output, Marker>, ) -> TunableSet<K, Inputs, Output>

Register a tunable with this tunable set

pub fn with_custom_checksum( self, checksum: impl Fn(&TunableSet<K, Inputs, Output>) -> String + Send + Sync + 'static, ) -> TunableSet<K, Inputs, Output>

Override the checksum algorithm

pub fn autotunables( &self, ) -> Vec<Arc<dyn Tunable<Inputs = Inputs, Output = Output>>>

All candidate operations for autotuning this operation type Operations can run on toy tensors of relevant size

pub fn fastest( &self, fastest_index: usize, ) -> Arc<dyn Tunable<Inputs = Inputs, Output = Output>>

Returns the operation for the given index, matching the order returned by autotunables. Operation obtained here runs on original tensors Nb: The 0 index is used a “good default”.

pub fn compute_checksum(&self) -> String

Compute a checksum that can invalidate outdated cached auto-tune results.

pub fn generate_key(&self, inputs: &Inputs) -> K

Generate a key from a set of inputs

pub fn generate_inputs(&self, key: &K, inputs: &Inputs) -> Inputs

Generate a set of test inputs from a key and reference inputs

Trait Implementations

impl<K, Inputs, Output> Clone for TunableSet<K, Inputs, Output>

where K: Clone + AutotuneKey, Inputs: Clone + Send + 'static, Output: Clone + 'static,

fn clone(&self) -> TunableSet<K, Inputs, Output>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.