Mojo function

blackwell_block_scaled_matmul_tma_umma_warp_specialized

blackwell_block_scaled_matmul_tma_umma_warp_specialized[c_type: DType, c_layout: Layout, a_type: DType, a_layout: Layout, b_type: DType, b_layout: Layout, sfa_dtype: DType, sfa_layout: Layout, sfb_dtype: DType, sfb_layout: Layout, transpose_b: Bool, *, config: BlockScaledMatmulConfig[a_type, b_type, c_type, sfa_dtype, sfb_dtype, transpose_b], elementwise_compute_lambda_fn: OptionalReg[fn[dtype: DType, width: Int, *, alignment: Int = 1](IndexList[2], SIMD[dtype, width]) capturing -> SIMD[dtype, width]] = None, register_based_epilogue: Bool = True, pdl_level: PDLLevel = PDLLevel(), max_profiled_tiles_per_SM: OptionalReg[UInt32] = None](c_tensor: LayoutTensor[c_type, c_layout, origin, address_space=address_space, element_layout=element_layout, layout_int_type=layout_int_type, linear_idx_type=linear_idx_type, masked=masked, alignment=alignment], a_tensor: LayoutTensor[a_type, a_layout, origin, address_space=address_space, element_layout=element_layout, layout_int_type=layout_int_type, linear_idx_type=linear_idx_type, masked=masked, alignment=alignment], b_tensor: LayoutTensor[b_type, b_layout, origin, address_space=address_space, element_layout=element_layout, layout_int_type=layout_int_type, linear_idx_type=linear_idx_type, masked=masked, alignment=alignment], a_scales_tensor: LayoutTensor[sfa_dtype, sfa_layout, MutAnyOrigin], b_scales_tensor: LayoutTensor[sfb_dtype, sfb_layout, MutAnyOrigin], ctx: DeviceContext)

Launch block-scaled FP8 matmul kernel on SM100.

Computes C = scale(A) @ scale(B) where A and B are FP8 matrices with per-block scaling factors following MXFP8 conventions.

Parameters:

c_type (DType): Output element type.
c_layout (Layout): Output tensor layout.
a_type (DType): A matrix element type (FP8).
a_layout (Layout): A matrix layout.
b_type (DType): B matrix element type (FP8).
b_layout (Layout): B matrix layout.
sfa_dtype (DType): A scaling factor type (F8-UE8M0).
sfa_layout (Layout): A scaling factor layout.
sfb_dtype (DType): B scaling factor type (F8-UE8M0).
sfb_layout (Layout): B scaling factor layout.
transpose_b (Bool): Whether B is transposed (must be True).
config (BlockScaledMatmulConfig): Block-scaled matmul configuration.
elementwise_compute_lambda_fn (OptionalReg): Optional epilogue lambda.
register_based_epilogue (Bool): Whether to use register-based epilogue.
pdl_level (PDLLevel): Programmatic dependent launch level.
max_profiled_tiles_per_SM (OptionalReg): Optional profiling tile count.

Args:

c_tensor (LayoutTensor): Output tensor.
a_tensor (LayoutTensor): A matrix tensor.
b_tensor (LayoutTensor): B matrix tensor.
a_scales_tensor (LayoutTensor): A scaling factors.
b_scales_tensor (LayoutTensor): B scaling factors.
ctx (DeviceContext): Device context for kernel launch.

Raises:

If configuration constraints are violated.