tilelang.carver.roller.shape_inference.tir¶

類別¶

`Statement`
`TensorDepNode`	For tensor dependency analysis.
`DependencyAnalysis`
`InputShapeInference`

函式¶

`region_exist_in_list`(a, list)
`walk_indice`(expr)
`get_analyzer_by_tir`(block_analyzer, args)

Module Contents¶

class tilelang.carver.roller.shape_inference.tir.Statement(block_analyzer, block)¶

參數:: block (tvm.tir.schedule.schedule.BlockRV)

block_analyzer¶

block¶

dep_name¶

dependent_region¶

reverse_bound_inference¶

make_reverse(input_name, input_iter)¶

參數:

input_name (str)
input_iter (list[tvm.tir.PrimExpr])

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

For tensor dependency analysis.

name¶

add_next(node)¶

add_prev(node)¶

deduplicate(lst)¶

__str__()¶

__repr__()¶

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

deps¶

name2dep¶

mapping¶

get_or_create_node(name)¶

traverse_dependencies(compute)¶

analyze()¶

print_dependencies()¶

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.

class tilelang.carver.roller.shape_inference.tir.InputShapeInference(deps)¶

參數:: deps (list[Statement])

deps¶

target_mapping¶

buffer_mapping¶

reduce_axes = []¶

dep_analysis¶

construct_dependency_target(targets)¶

參數:: targets (tuple[str])

infer(shape, rstep=None, targets=None)¶

參數:

shape (dict[str, list[tvm.arith.ConstIntBound]])
rstep (dict[str, int])

get_input_exprs(output_exprs)¶

tilelang.carver.roller.shape_inference.tir.region_exist_in_list(a, list)¶

回傳型別:: bool

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

tilelang.carver.roller.shape_inference.tir.get_analyzer_by_tir(block_analyzer, args)¶

回傳型別:: InputShapeInference