For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).
Mojo struct
StandardTileStorage
struct StandardTileStorage[a_type: DType, b_type: DType, a_shape: IndexList[2], b_shape: IndexList[2], num_pipeline_stages: Int]
Storage for standard matmul tiles (A and B).
This is the single source of truth for tile array types and storage. SMEM structs embed this rather than defining tile arrays separately.
All tiles use TileTensor natively. Convert to LayoutTensor at TMA/MMA boundaries using {ptr} syntax or explicit construction.
Parameters
- a_type (
DType): Data type for A matrix tiles. - b_type (
DType): Data type for B matrix tiles. - a_shape (
IndexList[2]): Tile shape for A tiles as (rows, cols). - b_shape (
IndexList[2]): Tile shape for B tiles as (rows, cols). - num_pipeline_stages (
Int): Number of pipeline stages (determines array depth).
Fields
- a_tiles_storage (
StandardTileStorage[a_type, b_type, a_shape, b_shape, num_pipeline_stages].ATileArray.Storage): - b_tiles_storage (
StandardTileStorage[a_type, b_type, a_shape, b_shape, num_pipeline_stages].BTileArray.Storage):
Implemented traits
AnyType,
ImplicitlyDestructible
comptime members
ATileArray
comptime ATileArray = SMemTileArray2D[a_type, a_shape[0], a_shape[1], num_pipeline_stages]
BTileArray
comptime BTileArray = SMemTileArray2D[b_type, b_shape[0], b_shape[1], num_pipeline_stages]
Methods
a_tiles
a_tiles(ref[AddressSpace._value] self) -> StandardTileStorage[a_type, b_type, a_shape, b_shape, num_pipeline_stages].ATileArray
Get A tile array accessor (TileTensor-based).
Returns:
StandardTileStorage[a_type, b_type, a_shape, b_shape, num_pipeline_stages].ATileArray
b_tiles
b_tiles(ref[AddressSpace._value] self) -> StandardTileStorage[a_type, b_type, a_shape, b_shape, num_pipeline_stages].BTileArray
Get B tile array accessor (TileTensor-based).
Returns:
StandardTileStorage[a_type, b_type, a_shape, b_shape, num_pipeline_stages].BTileArray
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!