Struct GemmWidths 

pub struct GemmWidths {
    pub a: u64,
    pub b: u64,
    pub a_buffer: u64,
    pub b_buffer: u64,
    pub out: u64,
}

The tile widths a GEMM’s movement is priced at. A transfer moves its tensor’s stored bytes, so each operand is read at its own tensor’s width and the result written at the output’s — an exact accumulator writes a wider result than the operands it reads (fp32 under bf16, int32 or int64 under int8) — and a multicast moves the sender’s buffer, at that buffer’s width.

Fields

a: u64

b: u64

a_buffer: u64

b_buffer: u64

out: u64

Implementations

impl GemmWidths

pub fn of(p: &SpatialProgram, t: &SpatialTarget) -> Option<GemmWidths>

The widths of p’s GEMM on t: the tensors its matrix stage’s operand buffers are loaded from (an operand no stage loads moves at its buffer’s width), those buffers, and the output. None for a program with no two-operand tile matmul or no output.

Trait Implementations

impl Clone for GemmWidths

fn clone(&self) -> GemmWidths

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GemmWidths

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

Formats the value using the given formatter.

impl PartialEq for GemmWidths

fn eq(&self, other: &GemmWidths) -> 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 GemmWidths

impl Eq for GemmWidths

impl StructuralPartialEq for GemmWidths