tilelang.carver.roller.hint =========================== .. py:module:: tilelang.carver.roller.hint .. autoapi-nested-parse:: Hint definition for schedule Classes ------- .. autoapisummary:: tilelang.carver.roller.hint.TensorCoreExtraConfig tilelang.carver.roller.hint.Stride tilelang.carver.roller.hint.TileDict tilelang.carver.roller.hint.IntrinInfo tilelang.carver.roller.hint.Hint Module Contents --------------- .. py:class:: TensorCoreExtraConfig(AS_shape, BS_shape, AF_shape, BF_shape, tc_axis) This class is used to store extra information for tensorcore .. py:attribute:: AS_shape :type: Tuple[int] .. py:attribute:: BS_shape :type: Tuple[int] .. py:attribute:: AF_shape :type: Tuple[int] .. py:attribute:: BF_shape :type: Tuple[int] .. py:attribute:: tc_axis :type: Tuple[int] .. py:class:: Stride(stride = 1, ax = -1) Manages stride information for a given axis of a tensor. .. py:property:: ax :type: int .. py:property:: stride :type: int .. py:method:: compute_strides_from_shape(shape) .. py:method:: compute_elements_from_shape(shape) .. py:method:: is_valid() .. py:method:: __repr__() .. py:class:: TileDict(output_tile) Manages tiling information and configurations for computational tasks. .. py:attribute:: output_tile .. py:attribute:: tile_map .. py:attribute:: rstep_map .. py:attribute:: cached_tensors_map .. py:attribute:: output_strides_map .. py:attribute:: tensor_strides_map .. py:attribute:: traffic :value: -1 .. py:attribute:: smem_cost :value: -1 .. py:attribute:: block_per_SM :value: -1 .. py:attribute:: num_wave :value: -1 .. py:attribute:: grid_size :value: -1 .. py:attribute:: valid :value: True .. py:method:: get_tile(func) .. py:method:: get_rstep(node) .. py:method:: __hash__() .. py:class:: IntrinInfo(in_dtype, out_dtype, trans_b, input_transform_kind = 0, weight_transform_kind = 0) The information of tensorcore intrinsic related information .. py:attribute:: in_dtype .. py:attribute:: out_dtype .. py:attribute:: trans_a :value: False .. py:attribute:: trans_b .. py:attribute:: input_transform_kind :value: 0 .. py:attribute:: weight_transform_kind :value: 0 .. py:method:: __repr__() .. py:method:: is_input_8bit() .. py:property:: smooth_a :type: bool .. py:property:: smooth_b :type: bool .. py:property:: inter_transform_a :type: bool .. py:property:: inter_transform_b :type: bool .. py:class:: Hint Bases: :py:obj:`object` Central configuration class for managing various parameters of computational tasks. .. py:attribute:: arch :value: None .. py:attribute:: use_tc :value: None .. py:attribute:: block :value: [] .. py:attribute:: thread :value: [] .. py:attribute:: warp :value: [] .. py:attribute:: rstep :value: [] .. py:attribute:: reduce_thread :value: [] .. py:attribute:: rasterization_plan .. py:attribute:: cached_tensors :value: [] .. py:attribute:: output_strides .. py:attribute:: schedule_stages :value: None .. py:attribute:: block_reduction_depth :type: int :value: None .. py:attribute:: split_k_factor :type: int :value: 1 .. py:attribute:: vectorize :type: Dict[str, int] .. py:attribute:: pipeline_stage :value: 1 .. py:attribute:: use_async :value: False .. py:attribute:: opt_shapes :type: Dict[str, int] .. py:attribute:: intrin_info .. py:attribute:: shared_scope :type: str :value: 'shared' .. py:attribute:: pass_context :type: Dict .. py:method:: to_dict() .. py:method:: from_dict(dic) :classmethod: .. py:method:: tensorcore_legalization() .. py:property:: raxis_order :type: List[int] .. py:property:: step :type: List[int] .. py:method:: __repr__() .. py:method:: complete_config(node)