Enum CoExecution 

pub enum CoExecution {
    NativePrecision,
    MatchGpuPrecision,
}

What precision the CPU half of a split batch should run at.

Variants

NativePrecision

CPU lanes stay f64, GPU lanes are f32. Fastest and most accurate overall, but the output array is not uniform in accuracy — see the module docs.

MatchGpuPrecision

Round the CPU half’s results to f32 as well, so every lane in the output has the same accuracy grade.

This makes the array uniform, not better: it throws away good bits on the CPU lanes to remove the seam. Choose it when downstream code compares cells against each other and a discontinuity at the split point would be mistaken for physics.

Trait Implementations

impl Clone for CoExecution

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 CoExecution

impl Debug for CoExecution

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

Formats the value using the given formatter.

impl Default for CoExecution

fn default() -> Self

Returns the “default value” for a type.

impl Eq for CoExecution

impl PartialEq for CoExecution

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for CoExecution