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

tiled_matmul_run

def tiled_matmul_run[config: KernelConfig, transpose_b: Bool, b_packed: Bool, simd_size: Int, elementwise_epilogue_enabled: Bool, kernel_id: InnerKernelID, algorithm: InnerMatmulKernel, ElementwiseEpilogueFnType: ImplicitlyCopyable & def[.origin._mlir_origin2x: LITImmutOrigin, .dtype2x1: DType, .LayoutType2x2: TensorLayout, .origin2x3: ImmutOrigin, .linear_idx_type2x4: DType, .element_size`2x5: Int](GemmShape, GemmShape, TileTensor[dtype, LayoutType, origin, linear_idx_type=linear_idx_type, element_size=element_size]) -> None](alg: algorithm, c: TileTensor[linear_idx_type=c.linear_idx_type, element_size=c.element_size], a: TileTensor[linear_idx_type=a.linear_idx_type, element_size=a.element_size], b: TileTensor[linear_idx_type=b.linear_idx_type, element_size=b.element_size], elementwise_epilogue_fn: ElementwiseEpilogueFnType, global_tile_shape: GemmShape, global_tile_offset: GemmShape)``

Interface function to run tiled matmul on a given sub-tile.

Args:

alg (algorithm): InnerMatmulKernel algorithm for microkernel.
c (TileTensor[linear_idx_type=c.linear_idx_type, element_size=c.element_size]): Pre-allocated buffer space for result.
a (TileTensor[linear_idx_type=a.linear_idx_type, element_size=a.element_size]): Operand A of the matmul.
b (TileTensor[linear_idx_type=b.linear_idx_type, element_size=b.element_size]): Operand B of the mamtul.
elementwise_epilogue_fn (ElementwiseEpilogueFnType): The elementwise epilogue function.
global_tile_shape (GemmShape): Tile shape this call will process.
global_tile_offset (GemmShape): Tile offset on the original buffer.