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 struct
PointerLoad
struct PointerLoad[dt: DType, width: Int, align: Int = Int((mul size_of[dt](), width))]
A vector load of width contiguous elements into registers.
Parameters
- dt (
DType): The element type. - width (
Int): The elements one load moves, a power of two. - align (
Int): The bytes of alignment the access claims.
Implemented traits
AnyType,
Atom,
Copyable,
Deinitable,
ImplicitlyCopyable,
LoadAtom,
Movable,
RegisterPassable,
TrivialRegisterPassable
comptime members
Completion
comptime Completion = Synchronous
Ready when the call returns.
dtype
comptime dtype = dt
The element type the instruction moves.
Engine
comptime Engine = DefaultEngine
The engine of the memory the instruction reads.
mem_map
comptime mem_map = vector_map(width, size_of[dt]())
Where each index reads memory.
reg_map
comptime reg_map = registers_map(width, Int(1))
Which register each index writes.
shape
comptime shape = ParameterList()
The instruction's shape, such as (32, 16) for tcgen05.ld.32x32b.x16 or (m, n, k) for an MMA.
shape_values
comptime shape_values = width
One issue moves width elements.
Methods
__call__
def __call__[T: DType, LR: TensorLayout, ER: TensorEngine, LM: TensorLayout, EM: TensorEngine](self, dst: MutTileTensor[T, LR, Engine=ER, address_space=dst.address_space, linear_idx_type=dst.linear_idx_type], src: ImmTileTensor[T, LM, Engine=EM, address_space=src.address_space, linear_idx_type=src.linear_idx_type])
Issues the load.
Parameters:
- T (
DType): The operands' element type. - LR (
TensorLayout): The register view's layout. - ER (
TensorEngine): The register view's engine. - LM (
TensorLayout): The memory view's layout. - EM (
TensorEngine): The memory view's engine.
Args:
- dst (
MutTileTensor[T, LR, Engine=ER, address_space=dst.address_space, linear_idx_type=dst.linear_idx_type]): This thread'swidthregisters. - src (
ImmTileTensor[T, LM, Engine=EM, address_space=src.address_space, linear_idx_type=src.linear_idx_type]): The memory, from the vector's first element.