Enum TensorWasmOp 

pub enum TensorWasmOp {
    VecAdd {
        elem: ElemType,
        lanes: u32,
    },
    VecMul {
        elem: ElemType,
        lanes: u32,
    },
    VecFma {
        elem: ElemType,
        lanes: u32,
    },
    MatMul {
        m: u32,
        n: u32,
        k: u32,
    },
    LoadUnified {
        elem: ElemType,
        lanes: u32,
    },
    StoreUnified {
        elem: ElemType,
        lanes: u32,
    },
    Barrier,
}

One coarse-grained op in a TensorWasm-emitted GPU kernel.

Each op corresponds to a small group of PTX instructions; the emitter (S12) materialises them into a register-allocated assembly block.

Variants

VecAdd

c = a + b element-wise.

Fields

elem: ElemType

Per-lane element type.

lanes: u32

Lane count.

VecMul

c = a * b element-wise.

Fields

elem: ElemType

Per-lane element type.

lanes: u32

Lane count.

VecFma

d = a*b + c element-wise (single rounding for floats).

Fields

elem: ElemType

Per-lane element type.

lanes: u32

Lane count.

MatMul

16x16x16 matrix multiply-accumulate (wmma on sm_80).

Fields

m: u32

Tile size m (currently fixed at 16).

n: u32

Tile size n (currently fixed at 16).

k: u32

Tile size k (currently fixed at 16).

LoadUnified

Load lanes from unified-memory pointer + offset.

Fields

elem: ElemType

Per-lane element type.

lanes: u32

Lane count.

StoreUnified

Store lanes to unified-memory pointer + offset.

Fields

elem: ElemType

Per-lane element type.

lanes: u32

Lane count.

Barrier

CTA-wide synchronisation barrier.

Trait Implementations

impl Clone for TensorWasmOp

fn clone(&self) -> TensorWasmOp

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorWasmOp

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

Formats the value using the given formatter.

impl Display for TensorWasmOp

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

Formats the value using the given formatter.

impl PartialEq for TensorWasmOp

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