Struct LInfNorm 

pub struct LInfNorm;

Evaluate the LInf norm of the argument.

Implementation

Closed implementation: Supported input types: f32, Half. f32 path: simple scalar loop using abs and max. Half path: widens each element with ‘diskann_wide::cast_f16_to_f32’ then applies abs and max.

Performance

The Half widening (cast_f16_to_f32) is per element and does not auto-vectorize well, so LInfNorm on large Half slices may become a throughput bottleneck compared to an explicit SIMD reduction (e.g. loading f16x8, converting once, then doing lane-wise abs & max in f32). Callers with large Half inputs should be aware of the potential bottleneck.

Current behavior is correct but potentially slower than expected for large Half slices.

Trait Implementations

impl Clone for LInfNorm

fn clone(&self) -> LInfNorm

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for LInfNorm

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

Formats the value using the given formatter.

impl Norm<&[f16]> for LInfNorm

fn evaluate(&self, x: &[Half]) -> f32

impl Norm<&[f32]> for LInfNorm

fn evaluate(&self, x: &[f32]) -> f32

impl<A> Target1<A, f32, &[f16]> for LInfNorm

where A: Architecture,

fn run(self, _: A, x: &[Half]) -> f32

Run the operation with the provided Architecture.

impl<A> Target1<A, f32, &[f32]> for LInfNorm

where A: Architecture,

fn run(self, _: A, x: &[f32]) -> f32

Run the operation with the provided Architecture.

impl Copy for LInfNorm