Module triton 

The Triton DSL trait and its supporting types.

Re-exports

pub use types::*;

Modules

llvm

LLVM-backend-facing DSL types (the compiled counterpart of this module’s Tensor/Pointer).

types

Dtype/numeric-kind trait hierarchy (Dtype, Num, Int, Float) used to bound the Triton trait’s generic methods.

Enums

Axis

A grid/program-ID axis.

CacheModifier

L1/L2 cache behaviour for load and store instructions.

DotFormat

Input format for scaled dot-product (dot_scaled).

EvictionPolicy

Cache eviction priority hint for load and store instructions.

FpDowncastRounding

Rounding mode used when down-casting floating-point types in cast.

InputPrecision

Tensor-core precision mode for dot on f32 × f32 inputs.

MemScope

Synchronization scope for atomic operations.

MemSem

Memory ordering semantics for atomic operations.

PaddingOption

Padding value applied to out-of-bounds lanes when using boundary_check in load.

Traits

BitAnd

The bitwise AND operator &.

BitOr

The bitwise OR operator |.

Triton

The Triton-like kernel DSL: tensor/pointer types and the operations (creation, shape manipulation, linear algebra, memory, math, reduction, scan/sort, atomics, RNG) available inside a #[kernel]-annotated function. See the module docs for how this compiles.