Common#
- class cutlass.cute.nvgpu.OpError(*args: Any, **kwargs: Any)#
Bases:
DSLBaseErrorAn exception class for Op construction errors.
- class cutlass.cute.nvgpu.MmaUniversalOp(abacc_dtype: Type[cutlass.cute.typing.Numeric])#
Bases:
MmaOpThe universal MMA Operation.
This Operation currently expects the A/B operands as well as the accumulator to share the same data types.
- Parameters:
abacc_dtype (Type[Numeric]) – The data type for the A/B operands and the accumulator
- abacc_dtype: Type[cutlass.cute.typing.Numeric]#
- class cutlass.cute.nvgpu.CopyUniversalOp#
Bases:
CopyOpThe universal Copy Operation.
When creating a Copy Atom out of this operation, the expected usage pattern is
op = cute.nvgpu.CopyUniversalOp() atom = cute.make_copy_atom(op, tensor_dtype, num_bits_per_copy=64)
tensor_dtypeis the data type used to build the reference TV Layout (either the source or the destination TV Layout) in unit of tensor elements and is used for partitioning byTiledCopyfor examplenum_bits_per_copyis a kw argument specifying the number of bits to copy per Atom execution. This can be larger than the width of the above data type. When not provided, the compiler will do a best effort at auto-vectorizing.