IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /max/get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
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

SMemTileArray

struct SMemTileArray[dtype: DType, shape_types: TypeList[shape_types.values], stride_types: TypeList[stride_types.values], num_tiles: Int, alignment: Int = 128]

Array of TileTensor tiles with variadic shape/stride type parameters.

This is the TileTensor equivalent of the LayoutTensor-based SMemTileArray in structuring.mojo. By taking shape_types and stride_types directly as variadic type parameters, this preserves full compile-time type information including swizzle patterns.

Example: comptime a_layout = internal_k_major[dtype, BM, BK, 128] comptime ATileArray = SMemTileArray[ dtype, a_layout.shape_types, a_layout.stride_types, num_pipeline_stages, ]

var array = ATileArray.stack_allocation() var tile = array[0] # Returns TileTensor with correct swizzled layout

Parameters​

  • ​dtype (DType): Tile element data type.
  • ​shape_types (TypeList[shape_types.values]): List of shape types from Layout (preserves compile-time info).
  • ​stride_types (TypeList[stride_types.values]): List of stride types from Layout (preserves compile-time info).
  • ​num_tiles (Int): Number of tiles in the array.
  • ​alignment (Int): Memory alignment (default 128 for shared memory).

Fields​

  • ​ptr (UnsafePointer[Scalar[dtype], MutAnyOrigin, address_space=AddressSpace.SHARED]):

Implemented traits​

AnyType, Copyable, ImplicitlyCopyable, ImplicitlyDestructible, Movable, RegisterPassable, TrivialRegisterPassable

comptime members​

num_elements​

comptime num_elements = (SMemTileArray[dtype, shape_types, stride_types, num_tiles, alignment].tile_size * num_tiles)

Storage​

comptime Storage = InlineArray[Scalar[dtype], SMemTileArray[dtype, shape_types, stride_types, num_tiles, alignment].num_elements]

storage_size​

comptime storage_size = (SMemTileArray[dtype, shape_types, stride_types, num_tiles, alignment].num_elements * size_of[dtype]())

Tile​

comptime Tile = TileTensor[dtype, Layout[shape_types, stride_types], MutAnyOrigin, address_space=AddressSpace.SHARED]

tile_size​

comptime tile_size = Coord[*#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))[idx]._ParamListType))[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[#kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))], [idx: __mlir_type.index] #kgen.param_list.concat(#kgen.param_list.tabulate(TypeList[values], [idx: __mlir_type.index] values[idx]._ParamListType))[idx]._ParamListType))[idx]._ParamListType))[idx]._ParamListType))].static_product

TileLayout​

comptime TileLayout = Layout[shape_types, stride_types]

Methods​

__init__​

__init__(ref[AddressSpace._value] storage: InlineArray[Scalar[dtype], SMemTileArray[dtype, shape_types, stride_types, num_tiles, alignment].num_elements]) -> Self

Initialize from inline storage.

Args:

Returns:

Self: A new SMemTileArray pointing to the storage.

__init__(unsafe_ptr: UnsafePointer[Scalar[dtype], address_space=AddressSpace.SHARED]) -> Self

Initialize with a shared memory pointer.

Args:

__getitem__​

__getitem__[T: Intable](self, index: T) -> SMemTileArray[dtype, shape_types, stride_types, num_tiles, alignment].Tile

Get tile at the given index.

Args:

  • ​index (T): The tile index.

Returns:

SMemTileArray[dtype, shape_types, stride_types, num_tiles, alignment].Tile: A TileTensor with correct swizzled layout at the given index.

slice​

slice[length: Int](self, start: Int) -> SMemTileArray[dtype, shape_types, stride_types, length, alignment]

Get a slice of the array.

Parameters:

  • ​length (Int): The length of the slice.

Args:

  • ​start (Int): The starting index.

Returns:

SMemTileArray[dtype, shape_types, stride_types, length, alignment]: A new SMemTileArray representing the slice.

stack_allocation​

static stack_allocation() -> Self

Allocate the array on the stack (in shared memory).

Returns:

Self: A new SMemTileArray backed by stack-allocated shared memory.

Models

Cookbook

Puzzles

Blog

Community

Contact