Enum F16Evaluation 

pub enum F16Evaluation {
    PerOperation,
    Chain,
    Accumulators,
}

How far an f32 intermediate is allowed to travel before it is rounded back to f16.

Variants

PerOperation

Round after every operation, which is what a GPU does. Where f16 is not native it costs a convert pair per operation.

Chain

Round where a value is stored, a let mut included, or read by anything but arithmetic. The default where f16 is not native.

Accumulators

Also hold a private f16 variable in f32 where that removes more converts than it adds, so a running total read often enough inside its loop stays in f16. Costs vector registers, so a wide kernel may want a narrower line.

Implementations

impl F16Evaluation

pub fn for_native_f16(native: bool) -> Self

The mode for a host that does or does not compute in f16 directly. Where it does, a chain held in f32 only adds converts.

Trait Implementations

impl Clone for F16Evaluation

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for F16Evaluation

impl Debug for F16Evaluation

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for F16Evaluation

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for F16Evaluation

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for F16Evaluation

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Eq for F16Evaluation

impl PartialEq for F16Evaluation

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl Serialize for F16Evaluation

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for F16Evaluation