For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo trait
TileLoader
Base trait for tile loading mechanisms in matrix multiplication.
This trait defines the interface for loading tiles from global memory to shared memory, abstracting over different hardware mechanisms.
Implemented traits
AnyType,
Copyable,
Deinitable,
ImplicitlyCopyable,
Movable,
RegisterPassable,
TrivialRegisterPassable
Required methods
load_tile
def load_tile(self, dst: TileTensor[address_space=AddressSpace.SHARED, linear_idx_type=dst.linear_idx_type], mem_barrier: Pointer[SharedMemBarrier, address_space=AddressSpace.SHARED], coords: Tuple[Int, Int])
Load a tile from global memory to shared memory.
Args:
- dst (
TileTensor[address_space=AddressSpace.SHARED, linear_idx_type=dst.linear_idx_type]): Destination tile in shared memory (must be 128-byte aligned). - mem_barrier (
Pointer[SharedMemBarrier, address_space=AddressSpace.SHARED]): Memory barrier for synchronization. - coords (
Tuple[Int, Int]): Tile coordinates (row, column) in the source matrix.