triton.experimental.gluon.language.NVMMASharedLayout

class triton.experimental.gluon.language.NVMMASharedLayout(self, swizzle_byte_width: int, element_bitwidth: int, rank: int = 2, transposed: bool = False, fp4_padded: bool = False, cga_layout: List[List[int]] = <factory>)

Represents a layout for shared memory suitable for NVIDIA MMA operations.

Parameters:

• swizzle_byte_width (int) – Width in bytes for swizzling.

• element_bitwidth (int) – Bitwidth of element type.

• rank (int) – Rank of the tensor.

• transposed (bool) – Whether the layout is transposed.

• fp4_padded (bool) – Whether FP4 padding is used.

• cga_layout (Optional[List[List[int]]]) – Bases describing CTA tiling.

__init__(self, swizzle_byte_width: int, element_bitwidth: int, rank: int = 2, transposed: bool = False, fp4_padded: bool = False, cga_layout: List[List[int]] = <factory>) → None

Methods

__init__(self, swizzle_byte_width, ...)
format_hardware_view(self, shape)
format_tensor_view(self, shape)
mangle(self)

Attributes

fp4_padded
get_default_for	Returns an NVMMASharedLayout with default swizzling for a given shape.
rank
transposed
type
swizzle_byte_width
element_bitwidth
cga_layout