from torch.fx.experimental.proxy_tensor import is_sym_node, py_sym_types from torch.fx.experimental.symbolic_shapes import hint_int import torch import torch.fx as fx import operator import math import torch.utils._pytree as pytree import copy import os from collections import defaultdict from torch.fx.passes import graph_drawer from typing import Tuple from .compile_utils import fx_graph_cse, get_aten_target from . import config import functools AOT_PARTITIONER_DEBUG = config.debug_partitioner class InvalidNodeBase: def __repr__(self): return "Invalid Node" InvalidNode = InvalidNodeBase() def _extract_graph_with_inputs_outputs(joint_graph, inputs, outputs): """ Given a graph, extracts out a subgraph that takes the specified nodes as inputs and returns the specified outputs. This includes specifying non-placeholder nodes as inputs. The general strategy is to initialize all inputs with proxies as we encounter them, and trace through the graph, only keeping values which take in valid proxies. Then, all dead code is eliminated. """ new_graph = fx.Graph() env = {} # Add new placeholder nodes in the order specified by the inputs for node in inputs: new_node = new_graph.placeholder(node.name) # Can't use node_copy here as we may be turning previous call_function into placeholders new_node.meta = node.meta env[node] = new_node for node in joint_graph.nodes: if node in inputs: continue elif node.op == 'placeholder': env[node] = InvalidNode elif node.op == 'call_function': all_args = pytree.tree_flatten((node.args, node.kwargs))[0] all_args = [isinstance(env[x], InvalidNodeBase) for x in all_args if isinstance(x, fx.Node)] if any(all_args): env[node] = InvalidNode continue env[node] = new_graph.node_copy(node, lambda x: env[x]) elif node.op == 'get_attr': env[node] = new_graph.node_copy(node, lambda x: env[x]) elif node.op == 'output': pass output_values = [] for x in outputs: if isinstance(x, fx.Node): if x not in env: raise RuntimeError(f"Node {x} couldn't be found in env") output_values.append(env[x]) else: output_values.append(x) new_graph.output(output_values) new_graph.eliminate_dead_code() new_graph.lint() return new_graph def _is_primal(node): return node.op == "placeholder" and "tangents" not in node.target def _is_tangent(node): return node.op == "placeholder" and "tangents" in node.target def _extract_fwd_bwd_outputs(joint_module: fx.GraphModule, *, num_fwd_outputs): outputs = pytree.tree_flatten([node.args for node in joint_module.graph.nodes if node.op == 'output'])[0] fwd_outputs = outputs[:num_fwd_outputs] bwd_outputs = outputs[num_fwd_outputs:] return fwd_outputs, bwd_outputs def _extract_fwd_bwd_modules(joint_module: fx.GraphModule, saved_values, saved_sym_nodes=(), *, num_fwd_outputs): fwd_outputs, bwd_outputs = _extract_fwd_bwd_outputs(joint_module, num_fwd_outputs=num_fwd_outputs) primal_inputs = list(filter(_is_primal, joint_module.graph.nodes)) tangent_inputs = list(filter(_is_tangent, joint_module.graph.nodes)) # Construct the forward module # Keep symints separate from tensors, passed between fwd/bwd graphs, and in the right order. fwd_graph = _extract_graph_with_inputs_outputs(joint_module.graph, primal_inputs, fwd_outputs + saved_values + saved_sym_nodes) bwd_graph = _extract_graph_with_inputs_outputs(joint_module.graph, saved_sym_nodes + saved_values + tangent_inputs, bwd_outputs) # This is to filter out saved values that don't actually end up being used by the backwards pass for node in bwd_graph.nodes: if node.op == 'placeholder' and not node.users: for saved_value in saved_values: if saved_value.name == node.name: saved_values.remove(saved_value) break for saved_sym in saved_sym_nodes: if saved_sym.name == node.name: saved_sym_nodes.remove(saved_sym) break # Now, we re-generate the fwd/bwd graphs. # NB: This might increase compilation time, but I doubt it matters fwd_graph = _extract_graph_with_inputs_outputs(joint_module.graph, primal_inputs, fwd_outputs + saved_values + saved_sym_nodes) bwd_graph = _extract_graph_with_inputs_outputs(joint_module.graph, saved_sym_nodes + saved_values + tangent_inputs, bwd_outputs) fwd_module = fx.GraphModule(joint_module, fwd_graph) bwd_module = fx.GraphModule(joint_module, bwd_graph) return fwd_module, bwd_module def default_partition( joint_module: fx.GraphModule, _joint_inputs, *, num_fwd_outputs ) -> Tuple[fx.GraphModule, fx.GraphModule]: """ Partitions the :attr:`joint_module` in a manner that closely resembles the behavior observed in the original ``.forward()`` and ``.backward()`` of the callable, i.e., the resulting forward graph contains those operators that are executed in the original ``.forward()`` callable passed to :func:`aot_function`. The default partitioner collects the operators that are between the forward inputs and the forward outputs. This helps in finding the tensors which have to be stashed for the backward pass. These stashed tensors become the output of the generated forward graph. The remaining operators are then placed in the backward graph. .. warning:: This API is experimental and likely to change. Args: joint_module(fx.GraphModule): The joint forward and backward graph. This is the result of AOT Autograd tracing. Returns: Returns the generated forward and backward Fx graph modules. """ primal_inputs = list(filter(_is_primal, joint_module.graph.nodes)) fwd_outputs, bwd_outputs = _extract_fwd_bwd_outputs(joint_module, num_fwd_outputs=num_fwd_outputs) forward_only_graph = _extract_graph_with_inputs_outputs(joint_module.graph, primal_inputs, fwd_outputs) forward_node_names = {node.name for node in forward_only_graph.nodes if node.op != 'output'} saved_values = [] saved_sym_nodes = [] for node in joint_module.graph.nodes: if node.name not in forward_node_names: continue if is_sym_node(node): # Symints must be kept separate from tensors so that PythonFunction only calls # save_for_backward on tensors and stashes symints in autograd .ctx saved_sym_nodes.append(node) elif ( 'tensor_meta' not in node.meta and node.op == 'call_function' ): # Since we can't save tuple of tensor values, we need to flatten out what we're saving users = node.users assert all(user.target == operator.getitem for user in users) for user in users: saved_values.append(user) else: backward_usages = [n for n in node.users if n.name not in forward_node_names] if 'tensor_meta' in node.meta and all(is_sym_node(n) for n in backward_usages): # If we have a tensor in the forward, where only its sizes/strides are needed in the backward, # and not the actual tensor data, # then it will be a lot cheaper to save only the sizes/strides, and not the actual tensor. # # Note that saving the tensor could also cause compilation problems: # If the user mutated an input in the forward and uses its sizes/strides in the backward, # then we would be obligated to clone the input before saving it to appease autograd. # (This is how we originally found this bug). for user in backward_usages: saved_sym_nodes.append(user) else: saved_values.append(node) saved_values = list(set(saved_values)) saved_sym_nodes = list(set(saved_sym_nodes)) return _extract_fwd_bwd_modules(joint_module, saved_values, saved_sym_nodes=saved_sym_nodes, num_fwd_outputs=num_fwd_outputs) def _prod(x): s = 1 for i in x: s *= i return s def _tensor_nbytes(numel, dtype): sizes = { torch.float: 4, torch.float16: 2, torch.bfloat16: 2, torch.float32: 4, torch.float64: 8, torch.int: 4, torch.int8: 1, torch.int16: 2, torch.int32: 4, torch.int64: 8, torch.uint8: 1, torch.bool: 1, } if dtype not in sizes: raise NotImplementedError("Don't know the size of dtype ", dtype) return numel * sizes[dtype] def _size_of(node: fx.Node) -> int: if 'val' in node.meta: val = node.meta['val'] if isinstance(val, py_sym_types): return 1 elif isinstance(val, (list, tuple)): return sum(_tensor_nbytes(hint_int(n.numel()), n.dtype) for n in val if isinstance(n, torch.Tensor)) elif isinstance(val, torch.Tensor): return _tensor_nbytes(hint_int(val.numel()), val.dtype) raise RuntimeError(f"Unknown metadata type {type(val)}") # Only needed since we don't always trace with fake tensors. if 'tensor_meta' in node.meta: metadata = node.meta['tensor_meta'] numel = _prod(map(to_size_hint, metadata.shape)) dtype = metadata.dtype else: return 0 return _tensor_nbytes(numel, dtype) # Used for some investigative purposes def _count_ops(graph): from collections import defaultdict cnt = defaultdict(int) for node in graph.nodes: if node.op == 'call_function': cnt[node.target.__name__] += 1 print(sorted(cnt.items(), key=lambda x: x[1], reverse=True)) @functools.lru_cache(None) def pointwise_ops(): ops = [] for attr_name in dir(torch.ops.aten): opoverloadpacket = getattr(torch.ops.aten, attr_name) if not isinstance(opoverloadpacket, torch._ops.OpOverloadPacket): continue for overload in opoverloadpacket.overloads(): op_overload = getattr(opoverloadpacket, overload) if torch.Tag.pointwise in op_overload.tags: # currently aot autograd uses packet not overload ops.append(opoverloadpacket) break return ops def min_cut_rematerialization_partition( joint_module: fx.GraphModule, _joint_inputs, compiler="nvfuser", recomputable_ops=None, *, num_fwd_outputs ) -> Tuple[fx.GraphModule, fx.GraphModule]: """ Partitions the joint graph such that the backward recomputes the forward. Recomputing helps in trading off memory bandwidth with computation. To create the fwd and bwd graph, we copy the joint graph, manually set the outputs to just original forward or backward outputs. And then we run the resulting graphs through dead code elimintation. .. warning:: This API is experimental and likely to change. Args: joint_module(fx.GraphModule): The joint forward and backward graph. This is the result of AOT Autograd tracing. _joint_inputs: The inputs to the joint graph. This is unused. compiler: This option determines the default set of recomputable ops. Currently, there are two options: ``nvfuser`` and ``inductor``. recomputable_ops: This is an optional set of recomputable ops. If this is not None, then this set of ops will be used instead of the default set of ops. num_fwd_outputs: The number of outputs from the forward graph. Returns: Returns the generated forward and backward Fx graph modules. """ try: import networkx as nx except ImportError as e: raise RuntimeError("Need networkx installed to perform smart recomputation " "heuristics") from e joint_module.graph.eliminate_dead_code() joint_module.recompile() fx_g = joint_module.graph # add the CSE pass if config.cse: cse_graph = fx_graph_cse(fx_g) joint_module.graph = cse_graph full_bw_graph = joint_module.graph name_to_node = {} for node in joint_module.graph.nodes: name_to_node[node.name] = node def classify_nodes(joint_module): required_bw_nodes = set() for node in joint_module.graph.nodes: if node.op == 'placeholder' and "tangents" in node.target: required_bw_nodes.add(node) if node in required_bw_nodes: for user in node.users: required_bw_nodes.add(user) primal_inputs = list(filter(_is_primal, joint_module.graph.nodes)) fwd_outputs, _ = _extract_fwd_bwd_outputs(joint_module, num_fwd_outputs=num_fwd_outputs) forward_only_graph = _extract_graph_with_inputs_outputs(joint_module.graph, primal_inputs, fwd_outputs) required_fw_nodes = {name_to_node[node.name] for node in forward_only_graph.nodes if node.op != 'output'} unclaimed_nodes = {node for node in joint_module.graph.nodes if node not in required_fw_nodes and node not in required_bw_nodes} return fwd_outputs, required_fw_nodes, required_bw_nodes, unclaimed_nodes orig_fw_outputs, required_fw_nodes, required_bw_nodes, unclaimed_nodes = classify_nodes(joint_module) def is_tensor_node(x): # When dynamic shapes are not enabled, fw outputs can be raw ints and not fx nodes if not isinstance(x, fx.Node): return False # It would be nice if we could guarantee that all fx nodes from make_fx get a 'val' # key in their meta dict, but that isn't always true today (see proxy_tensor.py) return 'tensor_meta' in x.meta or ('val' in x.meta and isinstance(x.meta['val'], torch.Tensor)) # networkx blows up on graphs with no required backward nodes # Since there's nothing to partition anyway, and the default partitioner can "handle" # this case, send our graph over to the default partitioner. if len(required_bw_nodes) == 0: return default_partition(joint_module, _joint_inputs, num_fwd_outputs=num_fwd_outputs) for node in reversed(joint_module.graph.nodes): if node not in required_fw_nodes: node.dist_from_bw = 0 else: node.dist_from_bw = int(1e9) for user in node.users: node.dist_from_bw = min(node.dist_from_bw, user.dist_from_bw + 1) aten = torch.ops.aten prims = torch.ops.prims # compiler == "nvfuser" is the default set of recomputable ops default_recomputable_ops = [aten.add, aten.sub, aten.div, aten.atan2, aten.mul, aten.max, aten.min, aten.pow, aten.remainder, aten.fmod, aten.__and__, aten.__or__, aten.__xor__, aten.__lshift__, aten.__rshift__, aten.eq, aten.ne, aten.ge, aten.gt, aten.le, aten.lt, aten.abs, aten.bitwise_not, aten.ceil, aten.floor, aten.frac, aten.neg, aten.relu, aten.round, aten.silu, aten.trunc, aten.log, aten.log10, aten.log1p, aten.log2, aten.lgamma, aten.exp, aten.expm1, aten.erf, aten.erfc, aten.cos, aten.acos, aten.cosh, aten.sin, aten.asin, aten.sinh, aten.tan, aten.atan, aten.tanh, aten.atanh, aten.sqrt, aten.rsqrt, aten.reciprocal, aten.sigmoid, aten.softplus, aten.threshold, aten.threshold_backward, aten.clamp, aten.where, aten.lerp, aten.addcmul, aten.gelu, aten.gelu_backward, aten.sum, aten.mean, aten._grad_sum_to_size, aten.sum_to_size, aten.amax, aten.to, aten.type_as, operator.getitem, aten.squeeze, aten.unsqueeze, aten.rsub, aten._to_copy] # noqa: E501 view_ops = [aten.squeeze, aten.unsqueeze, aten.alias] if compiler == "inductor": default_recomputable_ops += [prims.div, prims.convert_element_type, aten.clone, aten._to_copy, aten.full_like, prims.var, prims.sum, aten.var, aten.std, prims.broadcast_in_dim, aten.select, aten.permute, aten._unsafe_view, aten.view, aten.expand, aten.slice, aten.reshape, aten.broadcast_tensors, aten.scalar_tensor, aten.ones, aten.new_zeros, aten.lift_fresh_copy, aten.arange, aten.triu, aten.var_mean, aten.isinf, aten.any, aten.full, aten.as_strided, aten.zeros, aten.argmax, aten.maximum] # noqa: E501 view_ops += [aten.view, aten.slice, aten.permute, aten.t, prims.broadcast_in_dim, aten.expand, aten.as_strided] # Natalia said that we should allow recomputing indexing :) default_recomputable_ops += [aten.index] default_recomputable_ops += view_ops default_recomputable_ops += pointwise_ops() recomputable_ops = set(recomputable_ops) if recomputable_ops is not None else set(default_recomputable_ops) random_ops = [aten.native_dropout, aten.rand_like, aten.randn_like] compute_intensive_ops = [aten.mm, aten.convolution, aten.convolution_backward, aten.bmm, aten.addmm, aten.upsample_bilinear2d, aten._softmax, aten._softmax_backward_data, aten.native_layer_norm, aten.native_layer_norm_backward, aten.native_batch_norm, aten.native_batch_norm_backward, aten._native_batch_norm_legit] # noqa: E501 unrecomputable_ops = random_ops + compute_intensive_ops fusible_ops = recomputable_ops | set(random_ops) if AOT_PARTITIONER_DEBUG: joint_module_ops = { str(node.target._overloadpacket) for node in joint_module.graph.nodes if node.op == "call_function" and hasattr(node.target, "_overloadpacket") } ops_ignored = joint_module_ops - {str(i) for i in recomputable_ops} print("Ops banned from rematerialization: ", ops_ignored) print() AGGRESSIVE_RECOMPUTATION = False def is_materialized_backwards(node): cur_nodes = {node} while len(cur_nodes) > 0: cur = cur_nodes.pop() for user in cur.users: if user not in required_fw_nodes and not is_fusible(cur, user): return True if user not in required_fw_nodes and get_aten_target(user) in view_ops: cur_nodes.add(user) return False def ban_recomputation(node): if AGGRESSIVE_RECOMPUTATION: return (node.op == 'call_function' and get_aten_target(node) in unrecomputable_ops) else: if node.op != 'call_function': return False if get_aten_target(node) not in recomputable_ops: return True if node.target == operator.getitem: return False if node.target in [aten.lift_fresh_copy.default, aten.lift_fresh.default]: return False # If a node *must* be materialized in the backwards pass, then we # should never recompute it. This is a pretty subtle point. In # general, the assumption we make is that recomputing a node in the # backwards pass is "free". However, if a node must be materialized # in the backwards pass, then recomputing it is never free. if is_materialized_backwards(node): return True # Arbitrary hack that sometimes seems to help things. The above # modification appears to have made this heuristic a lot less critical # for performance. # TODO: Investigate why this hack helps. if compiler == "inductor" and node.dist_from_bw > config.max_dist_from_bw: return True # If the output of an op is 4x smaller (arbitrary choice), # then we don't allow recomputation. input_tensors_size = sum(_size_of(i) for i in node.args if isinstance(i, fx.Node)) output_size = _size_of(node) return (output_size * 4 < input_tensors_size) def is_fusible(a, b): return get_aten_target(a) in fusible_ops and get_aten_target(b) in fusible_ops def is_materialized(node): if node.op == 'placeholder': return True return not all(is_fusible(node, user) for user in node.users) def get_node_weight(node) -> int: mem_sz = _size_of(node) # Heuristic to bias towards nodes closer to the backwards pass # Complete guess about current value mem_sz = int(mem_sz * (1.1 ** max(min(node.dist_from_bw, 100), 1))) # mem_sz = int(mem_sz + node.dist_from_bw) if is_materialized(node): return mem_sz else: return mem_sz * 2 nx_graph = nx.DiGraph() for node in full_bw_graph.nodes: if node.op == 'output': continue if node in required_bw_nodes: nx_graph.add_edge(node.name + "_in", "sink", capacity=math.inf) continue if node.op == 'placeholder' and "primals" in node.target: nx_graph.add_edge("source", node.name + "_in", capacity=math.inf) # If a node can't be recomputed (too expensive or involves randomness), # we prevent it from being recomputed by adding an inf edge to the source # We only need to ban nodes in the fw pass, as those are the only ones that would be recomputed. if ban_recomputation(node) and node in required_fw_nodes: nx_graph.add_edge("source", node.name + "_in", capacity=math.inf) # Checks if a node is actually a tuple. Can be simplified to just an isisinstance check if we always use faketensors. is_non_tensor_node = (('val' not in node.meta and 'tensor_meta' not in node.meta) or ('val' in node.meta and not isinstance(node.meta['val'], torch.Tensor))) if is_sym_node(node): weight = 1 elif is_non_tensor_node: weight = math.inf else: weight = get_node_weight(node) # Creates the weights on the "node" edge nx_graph.add_edge(node.name + "_in", node.name + "_out", capacity=weight) for user in node.users: nx_graph.add_edge(node.name + "_out", user.name + "_in", capacity=math.inf) cut_value, partition = nx.minimum_cut(nx_graph, "source", "sink") reachable, non_reachable = partition cutset = set() for u, nbrs in ((n, nx_graph[n]) for n in reachable): cutset.update((u, v) for v in nbrs if v in non_reachable) cut_nodes = set() for node_in, node_out in cutset: assert node_in[:-3] == node_out[:-4] node_name = node_in[:-3] cut_nodes.add(node_name) # To make this stuff deterministic node_idx = {node: idx for idx, node in enumerate(joint_module.graph.nodes)} saved_values = sorted((name_to_node[node] for node in cut_nodes), key=lambda x: node_idx[x]) # Symints must be kept separate from tensors so that PythonFunction only calls # save_for_backward on tensors and stashes symints in autograd .ctx saved_sym_nodes = list(filter(lambda n: is_sym_node(n), saved_values)) saved_values = list(filter(lambda n: not is_sym_node(n), saved_values)) fw_module, bw_module = _extract_fwd_bwd_modules( joint_module, saved_values, saved_sym_nodes=saved_sym_nodes, num_fwd_outputs=num_fwd_outputs) if AOT_PARTITIONER_DEBUG: print("Theoretical Activations Stored: ", sum([_size_of(i) for i in saved_values]) / 1e9) fw_module_nodes = {node.name for node in fw_module.graph.nodes if node.op == 'call_function'} bw_module_nodes = {node.name for node in bw_module.graph.nodes if node.op == 'call_function'} remat_nodes = fw_module_nodes & bw_module_nodes counts = defaultdict(int) for node in fw_module.graph.nodes: if node.name in remat_nodes and hasattr(node.target, '_overloadpacket'): counts[str(node.target._overloadpacket)] += 1 print(f"# remat/fw/bw: {len(remat_nodes)}/{len(fw_module_nodes)}/{len(bw_module_nodes)}") print("Count of Ops Rematerialized: ", sorted(counts.items(), key=lambda x: x[1], reverse=True)) return fw_module, bw_module def draw_graph(traced: torch.fx.GraphModule, fname: str, figname: str = "fx_graph", clear_meta=True): if clear_meta: new_graph = copy.deepcopy(traced.graph) traced = fx.GraphModule(traced, new_graph) for node in traced.graph.nodes: node.meta = {} base, ext = os.path.splitext(fname) if not ext: ext = ".svg" print(f"Writing FX graph to file: {base}{ext}") g = graph_drawer.FxGraphDrawer(traced, figname) x = g.get_main_dot_graph() getattr(x, "write_" + ext.lstrip("."))(f"{base}{ext}") def draw_joint_graph(graph, joint_inputs, file_name="full_graph.png"): draw_graph(graph, file_name) return default_partition(graph, joint_inputs)