Enum HmmaKernel 

pub enum HmmaKernel {
    CpAsync {
        tile: WmmaTile,
        stages: usize,
    },
    Basic,
}

Which half-precision WMMA GEMM kernel is the FASTEST legal choice for a shape.

Variants

CpAsync

cp.async double-buffered — the measured 1.43× lever; needs sm_80+, the tile to divide the shape evenly, and the (operand-byte-sized) double-buffered footprint to fit static shared memory.

Fields

tile: WmmaTile

stages: usize

Basic

The basic one-warp-per-16×16-tile kernel — any 16-aligned shape (the fallback when cp.async is illegal).

Trait Implementations

impl Clone for HmmaKernel

fn clone(&self) -> HmmaKernel

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for HmmaKernel

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

Formats the value using the given formatter.

impl PartialEq for HmmaKernel

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for HmmaKernel

impl Eq for HmmaKernel

impl StructuralPartialEq for HmmaKernel