tilelang.carver.roller.shape_inference.tir module#

class tilelang.carver.roller.shape_inference.tir.DependencyAnalysis(deps)#

Bases: object

analyze()#

find_path_from_source(start_name, target_name)#: Finds the path (if it exists) from a starting node (source) to a target node. Returns the path as a list of nodes.

get_or_create_node(name)#

print_dependencies()#

traverse_dependencies(compute)#

class tilelang.carver.roller.shape_inference.tir.InputShapeInference(deps: List[Statement])#

Bases: object

construct_dependency_target(targets: Tuple[str])#

get_input_exprs(output_exprs)#

infer(shape: Dict[str, List[ConstIntBound]], rstep: Optional[Dict[str, int]] = None, targets=None)#

class tilelang.carver.roller.shape_inference.tir.Statement(block_analyzer, block: BlockRV)#

Bases: object

make_reverse(input_name: str, input_iter: List[PrimExpr])#

class tilelang.carver.roller.shape_inference.tir.TensorDepNode(name)#

Bases: object

For tensor dependency analysis.

add_next(node)#

add_prev(node)#

deduplicate(lst)#

tilelang.carver.roller.shape_inference.tir.get_analyzer_by_tir(block_analyzer, args) → InputShapeInference#

tilelang.carver.roller.shape_inference.tir.region_exist_in_list(a, list) → bool#

tilelang.carver.roller.shape_inference.tir.walk_indice(expr)#