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
TiledTensorCore
struct TiledTensorCore[out_type: DType, in_type: DType, shape: Coord[*shape.element_types], group_size: Int, transpose_b: Bool = False]
Describes a TensorCore MMA that is issued group_size times along K.
The AMD attention and matmul kernels pass this as a parameter carrier for
the MMA shape, operand types and K grouping; the K loop itself runs in
TiledMmaOp.
Parameters
- out_type (
DType): The data type for output/accumulation operations. - in_type (
DType): The data type for input matrix elements. - shape (
Coord[*element_types]): The shape parameters for individual MMA operations [M, N, K]. - group_size (
Int): Number of MMA operations along the K dimension. - transpose_b (
Bool): Whether to transpose the b matrix. Defaults to False.
Implemented traits
comptime members
mma_op
comptime mma_op = TensorCore()
The underlying TensorCore instance for MMA operations.