Struct MatmulGeometry 

pub struct MatmulGeometry {
    pub tile_x: u32,
    pub tile_y: u32,
    pub micro_m: u32,
    pub micro_n: u32,
    pub depth: u32,
}

Shape of one MATMUL workgroup: tile_x × tile_y invocations, each accumulating a micro_m × micro_n register block, over shared-memory slabs depth deep in k. Every dimension is a power of two and the two slabs stage evenly over the invocations.

Fields

tile_x: u32

tile_y: u32

micro_m: u32

micro_n: u32

depth: u32

Implementations

impl MatmulGeometry

pub const fn wide(tile: u32) -> Self

The square register-tiled geometry: tile × tile invocations, MATMUL_MICRO² each, tile deep — a 64 × 64 block at the preferred tile.

pub const fn invocations(self) -> u32

pub const fn block_m(self) -> u32

pub const fn block_n(self) -> u32

pub const fn local_size(self) -> [u32; 3]

pub const fn shared_bytes(self) -> u32

Two binary32 slabs: lhs block_m × depth (stored transposed) and rhs depth × block_n.

Trait Implementations

impl Clone for MatmulGeometry

fn clone(&self) -> MatmulGeometry

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for MatmulGeometry

impl Debug for MatmulGeometry

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

Formats the value using the given formatter.

impl Eq for MatmulGeometry

impl PartialEq for MatmulGeometry

fn eq(&self, other: &MatmulGeometry) -> 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 StructuralPartialEq for MatmulGeometry