For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

swapAB_largeM_clustered

def swapAB_largeM_clustered[a_type: DType, b_type: DType, c_type: DType, transpose_b: Bool = True](m: Int, pdl_level: PDLLevel) -> MatmulConfig[a_type, b_type, c_type, transpose_b]

Config for m in [129, 240] range with cluster=(2,1,1).

Pattern: - BN = 72 + ((m - 129) // 16) * 8 - Stages: 12 for m<=160, 10 for m<=224, 8 otherwise - cluster = (2,1,1), k_group_size = 2, swapAB = True

Parameters:

a_type (DType): Element DType of the A input matrix.
b_type (DType): Element DType of the B input matrix.
c_type (DType): Element DType of the output C matrix.
transpose_b (Bool): Whether B is stored transposed (K-major layout) (defaults to True).

Args:

m (Int): The M dimension of the matmul, in the range [129, 240].
pdl_level (PDLLevel): Programmatic dependent launch level for grid controls.

Returns:

MatmulConfig[a_type, b_type, c_type, transpose_b]