Skip to main content

Mojo function

tt_load_b

tt_load_b[mma_shape: IndexList[3], swizzle: Optional[Swizzle], num_mmas: Int, simd_width: Int](src: TileTensor[src.dtype, src.LayoutType, src.origin, address_space=AddressSpace.SHARED, linear_idx_type=src.linear_idx_type, element_size=src.element_size]) -> InlineArray[SIMD[src.dtype, simd_width], num_mmas]

Full B operand load from a SMEM warp tile.

Loads all MMA tiles from a WN×BK SMEM warp tile and returns them as an InlineArray of SIMD fragments (one per MMA tile).

Returns:

InlineArray

Was this page helpful?