warp submodule

class cutlass.cute.nvgpu.warp.MmaF16BF16Op(

ab_dtype: Type[cutlass.cute.typing.Numeric],

acc_dtype: Type[cutlass.cute.typing.Numeric],

shape_mnk: cutlass.cute.typing.Shape,

)

Bases: WarpMmaOp

F16/BF16 tcgen05 MMA Operation.

See the PTX documentation. This Operation covers the instructions using the .f16 or .bf16 qualifiers for the input operands.

ab_dtype: Type[cutlass.cute.typing.Numeric]

acc_dtype: Type[cutlass.cute.typing.Numeric]

shape_mnk: cutlass.cute.typing.Shape

__init__(

ab_dtype: Type[cutlass.cute.typing.Numeric],

acc_dtype: Type[cutlass.cute.typing.Numeric],

shape_mnk: cutlass.cute.typing.Shape,

) → None

class cutlass.cute.nvgpu.warp.LdMatrix8x8x16bOp(transpose: bool = False, num_matrices: int = 1)

Bases: BaseOp

8x8 ldmatrix Operation.

See the PTX documentation. This operation corresponds to the .m8n8 qualifier.

__init__(

transpose: bool = False,

num_matrices: int = 1,

) → None

class cutlass.cute.nvgpu.warp.LdMatrix16x16x8bOp(num_matrices: int)

Bases: BaseOp

16x16 8-bit ldmatrix Operation.

See the PTX documentation. This operation corresponds to the .m16n16 and the .b16 qualifiers.

__init__(num_matrices: int) → None

class cutlass.cute.nvgpu.warp.StMatrix8x8x16bOp(transpose: bool = False, num_matrices: int = 1)

Bases: BaseOp

8x8 stmatrix Operation.

See the PTX documentation. This operation corresponds to the m8n8 qualifier.

__init__(

transpose: bool = False,

num_matrices: int = 1,

) → None

class cutlass.cute.nvgpu.warp.StMatrix16x8x8bOp(num_matrices: int)

Bases: BaseOp

16x8 stmatrix Operation.

See the PTX documentation. This operation corresponds to the m16n8 qualifier.

__init__(num_matrices: int) → None