# Copyright (c) Meta Platforms, Inc. and affiliates import contextlib import logging import warnings from collections.abc import Sequence from typing import cast import torch import torch.distributed as dist import torch.distributed.tensor._api as dtensor import torch.distributed.tensor._random as random from torch._library.utils import fill_defaults from torch.distributed._functional_collectives import _are_we_tracing from torch.distributed.device_mesh import DeviceMesh from torch.distributed.tensor._argmin_argmax import argmin_argmax_handler from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta from torch.distributed.tensor._op_schema import ( OpInfo, OpSchema, OutputSharding, OutputSpecType, ) from torch.distributed.tensor._random import is_rng_supported_mesh from torch.distributed.tensor._redistribute import redistribute_local_tensor from torch.distributed.tensor._sharding_prop import ShardingPropagator from torch.distributed.tensor._tp_conv import ( convolution_backward_handler, convolution_handler, ) from torch.distributed.tensor._utils import ( ExplicitRedistributionContext, try_find_mesh_from_args, ) from torch.distributed.tensor.placement_types import Partial, Placement, Replicate from torch.utils._debug_mode import get_active_debug_mode from torch.utils._python_dispatch import return_and_correct_aliasing try: from torch.utils import _cxx_pytree as pytree except ImportError: from torch.utils import _pytree as pytree # type: ignore[no-redef] aten = torch.ops.aten logger = logging.getLogger(__name__) def as_strided_handler( op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], ): args, kwargs = fill_defaults(op_call._schema, args, kwargs) assert not kwargs tensor, size, stride, storage_offset = args if ( tensor.size() == tuple(size) and tensor.stride() == tuple(stride) and (storage_offset is None or tensor.storage_offset() == storage_offset) ): return torch.ops.aten.alias.default(tensor) raise RuntimeError("as_strided not supported with DTensor") def is_same_size_handler( op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], ) -> bool: lhs = cast(torch.Tensor, args[0]) rhs = cast(torch.Tensor, args[1]) return lhs.shape == rhs.shape def found_inf_reduce_handler( op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], ) -> None: op_info = dtensor.DTensor._op_dispatcher.unwrap_to_op_info(op_call, args, kwargs) local_tensor_args = pytree.tree_unflatten( cast(list[object], op_info.local_args), op_info.args_tree_spec, # type: ignore[arg-type] ) local_tensor_args = cast(tuple[object, ...], local_tensor_args) op_call(*local_tensor_args, **op_info.local_kwargs) grad_dtensor = cast(list[dtensor.DTensor], args[0])[0] grad_placements = grad_dtensor.placements mesh = grad_dtensor.device_mesh found_inf_placements: list[Placement] = [] for placement in grad_placements: if isinstance(placement, Replicate): found_inf_placements.append(placement) else: found_inf_placements.append(Partial("max")) target_tensor = cast(torch.Tensor, args[1]) spec = DTensorSpec( mesh=mesh, placements=tuple(found_inf_placements), tensor_meta=TensorMeta( shape=target_tensor.size(), stride=target_tensor.stride(), dtype=target_tensor.dtype, ), ) # pyrefly: ignore [bad-argument-type] found_inf_dtensor = dtensor.DTensor( local_tensor=target_tensor, # pyrefly: ignore [unexpected-keyword] spec=spec, # pyrefly: ignore [unexpected-keyword] requires_grad=False, # pyrefly: ignore [unexpected-keyword] ) found_inf = found_inf_dtensor.full_tensor() target_tensor.copy_(found_inf) class OpDispatcher: """ Op dispatching class instance to handle args/kwargs pre-processing (un-wrapping), sharding propagation, redistribute local args, local compute, and post-processing (re-wrapping). It also handles any op specific logic if necessary. NOTE: Given the runtime overhead of Tensor subclass (__torch_dispatch__), the OpDispatcher is designed to minimize the CPU overhead by using the tricks of proper unflattening, faster pytree if needed, and leveraging various caching mechanisms implemented in the sharding propagation and redistribute modules. The CPU overhead is critical to eager mode performance, one need to carefully measure the CPU overhead when making significant changes to the OpDispatcher and ShardingPropagator. """ def __init__(self) -> None: self.sharding_propagator = ShardingPropagator() # NOTE: must stay in sync with is_random_op in # torch/csrc/autograd/python_variable.cpp self._random_ops = { aten.native_dropout.default, aten.normal_.default, aten.rand.default, aten.rand_like.default, aten.randn.default, aten.randn_like.default, aten.randint_like.default, aten.randint_like.low_dtype, aten.randint_like.low_dtype_out, aten.uniform_.default, aten.bernoulli.default, aten.bernoulli_.float, } self._custom_op_handlers = { aten.is_same_size.default: is_same_size_handler, aten.convolution.default: convolution_handler, aten.convolution_backward.default: convolution_backward_handler, aten._amp_foreach_non_finite_check_and_unscale_.default: found_inf_reduce_handler, aten.as_strided.default: as_strided_handler, aten.argmin.default: argmin_argmax_handler, aten.argmax.default: argmin_argmax_handler, } # ******************************************************************************************** # def dispatch(...) # # NOTE: this class no longer contains the top-level dispatch entrypoint! # See #167051 for details # # The entrypoint has been moved to C++, and it handles common cases and then calls back into # OpDispatcher python to handle corner cases. # See dispatchDTensorOp() defined in python_variable.cpp and called from python_arg_parser.cpp # ******************************************************************************************** # This flag is used internally to control whether we treat the torch.Tensor(non-DTensor) # as implicitly replicated or we throw error to user. # NOTE: It is EXTREMELY UNSAFE to turn this flag on by default so we intentionally leave # it as False by default. @property def _allow_implicit_replication(self) -> bool: return torch._C._get_dtensor_allow_implicit_replication() @_allow_implicit_replication.setter def _allow_implicit_replication(self, value: bool) -> None: return torch._C._set_dtensor_allow_implicit_replication(value) def _propagate_op_sharding_dispatch_slow_path( self, op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], op_info: OpInfo, # The logic here is a bit messy. There are several reasons why the # C++ fastpath may have bailed out. If we just cache missed, we will # come here because we need to actually calculate the real thing. # There's no need to have a SECOND Python cache lookup; the C++ native # cache completely subsumes it. But sometimes, we will have failed # to compute the cache key in C++ entirely. In this case, we DO need # to do a cache lookup in Python, as the missing cache key in C++ # means we don't have access to it all. Furthermore, without duping # this function, we need to do the try_cache test inside of the # try-except block so that either case hits the inference mode / # exception rewrapping case. # # This should be cleaned up. First, ensuring the C++ codepath can # always compute a key will be a big help. Second, we should properly # fastpath inference mode composite implicit autograd so that you # don't have to throw an exception even in "fastpath". try_cache: bool, ) -> object: try: # We have basically inlined propagate() here, but WITHOUT the # output_sharding assignment if try_cache and not _are_we_tracing(): return self.sharding_propagator.propagate_op_sharding(op_info.schema) else: return self.sharding_propagator.propagate_op_sharding_non_cached( op_info.schema ) except NotImplementedError: if torch._C._dispatch_has_kernel_for_dispatch_key( op_call.name(), torch._C.DispatchKey.CompositeImplicitAutograd ): # When running under inference mode, CompositeImplicitAutograd ops show up in __torch_dispatch__, # so we manually decompose them, here out = op_call.decompose(*args, **kwargs) assert out is not NotImplemented return out else: raise except Exception as e: raise RuntimeError( f"{e}\n\nSharding propagation failed for {op_info.schema}" ) from e def _dispatch_get_local_results_slow_path( self, op_call: torch._ops.OpOverload, args: tuple[object, ...], op_info: OpInfo, ) -> object: output_sharding = op_info.output_sharding assert output_sharding is not None, "output sharding should not be None" mesh = op_info.compute_mesh participating = mesh.get_coordinate() is not None local_results = None if participating: # computation that happens in the current rank of the mesh, normal case if output_sharding.needs_redistribute: # If sharding propagation decision needs redistribute, perform redistribute # on args first, which could potentially modify args (i.e. allgather certain arg) assert output_sharding.redistribute_schema is not None self.redistribute_local_args( op_info, output_sharding.redistribute_schema, output_sharding.use_val_from_redistribute_schema, ) local_tensor_args = ( pytree.tree_unflatten( cast(list[object], op_info.local_args), # pyrefly: ignore [bad-argument-type] op_info.args_tree_spec, ) if op_info.args_tree_spec else op_info.local_args ) # run local op computation with potentially modified args/kwargs local_tensor_args = cast(tuple[object, ...], local_tensor_args) if op_call in self._random_ops: if not random._rng_tracker and is_rng_supported_mesh(mesh): # Default to `OffsetBasedRNGTracker` if the parallelism API # did not already construct one random._rng_tracker = random.OffsetBasedRNGTracker(mesh) first_arg, first_local_arg = ( cast(dtensor.DTensor, args[0]), cast(torch.Tensor, local_tensor_args[0]), ) # If the user provided a generator, we hook it up to our RNG manager, but we also pop it from kwargs # so the op_call does not directly use it (we want op_call to fall back to the 'default' which is # our RNG manager) maybe_user_generator = op_info.local_kwargs.pop("generator", None) assert maybe_user_generator is None or isinstance( maybe_user_generator, torch.Generator ) # maybe_user_generator = None rng_context = ( random._rng_tracker._distribute_region( first_arg._spec, generator=maybe_user_generator ) if random._rng_tracker and not first_local_arg.is_meta else contextlib.nullcontext() ) # For DTensor random operator, run it within a RNGTracker context to # ensure the random number generator is properly distributed. with rng_context: local_results = op_call(*local_tensor_args, **op_info.local_kwargs) else: # normal case, run local sharded op computation local_results = op_call(*local_tensor_args, **op_info.local_kwargs) else: # For a non-participating device (happens on rank that does not belong to # the device mesh), we do: # 1. if the return type is scalar, set the local result to None. # 2. if the return type is Tensor or List[Tensor], return empty # tensor(s) with correct dtype. spec = output_sharding.output_spec ret_list = op_call._schema.returns if spec is None: # For a scalar return type, the non-participating device has None # as its local result local_results = None else: def default_tensor(spec: DTensorSpec) -> torch.Tensor: if spec.tensor_meta is not None: shape = spec.tensor_meta.shape dtype = spec.tensor_meta.dtype if len(shape) == 0: # scalar tensor return torch.zeros((), dtype=dtype) else: # non-scalar tensor return torch.tensor([], dtype=dtype) else: raise RuntimeError(f"{spec} has no tensor metadata.") if isinstance(spec, DTensorSpec): # return a Tensor value local_results = default_tensor(spec) elif isinstance(spec, Sequence): # return a List[Tensor] value local_results = [ default_tensor(s) if s is not None else None for s in spec ] assert isinstance(local_results, list) if None in local_results: ret_type = str(ret_list[0].type) raise NotImplementedError( f"return type {ret_type} in DTensor op is not supported" ) return local_results def _dispatch_fast_path_python_tail( self, op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], compute_mesh: DeviceMesh, output_sharding: OutputSharding, local_results: object, participating: bool, is_inplace_op: bool, is_out_variant_op: bool, ) -> object: """ Tail of main dispatching logic, called from C++ fast path. """ if output_sharding.output_spec is None: if op_call == aten.equal.default: # The output of the equal op is a bool, by converting it into a # a single value tensor, we can use all-reduce with min reduce op # to simulate logical and. assert local_results is None or isinstance(local_results, bool) r = torch.tensor( int(local_results) if local_results is not None else 1, device=compute_mesh.device_type, ) dist.all_reduce(r, op=dist.ReduceOp.MIN) local_results = bool(r.item()) if is_inplace_op: # inplace op should return self instead of re-wrapping if output_sharding.output_spec is not None: output_spec = output_sharding.output_spec assert isinstance(output_spec, DTensorSpec) assert isinstance(args[0], dtensor.DTensor) # NOTE: aten.squeeze_.dim is an inplace op but it also may change # the inplace argument's tensor meta. Here we choose to special case # this op because as far as I know this is the only inplace op that # has such as behavior. We can extend this special case if necessary. if op_call == aten.squeeze_.dim: # update the spec to handle tensor meta changes args[0]._spec = output_spec # use return_and_correct_aliasing to match the outer and the inner # aliasing. See https://github.com/pytorch/pytorch/pull/158954 return return_and_correct_aliasing(op_call, args, kwargs, args[0]) else: # For all other inplace ops, check if placement changes are required # Inplace operations that change placement are not supported because # they would require redistribution, which breaks aliasing semantics. # If there are views into the tensor, the views would not be updated. if args[0]._spec.placements != output_spec.placements: raise RuntimeError( f"{op_call}: in-place operations that require placement changes " f"are not supported. The operation would change placement from " f"{args[0]._spec.placements} to {output_spec.placements}, " f"which requires redistribution and breaks aliasing semantics. " f"Please use the out-of-place version of this operation instead." ) # Most inplace ops don't change tensor meta, so no spec update needed return args[0] else: return None elif is_out_variant_op: # out variant could possibly have multiple out args (i.e. lu_unpack.out) output_specs = ( (output_sharding.output_spec,) if not isinstance(output_sharding.output_spec, tuple) else output_sharding.output_spec ) out_dts = [] spec_idx = 0 for argument in op_call._schema.arguments: if argument.is_out: out_dt = cast(dtensor.DTensor, kwargs[argument.name]) out_dt._spec = cast(DTensorSpec, output_specs[spec_idx]) out_dts.append(out_dt) spec_idx += 1 assert len(out_dts) >= 1, "out variant should have at least one out arg" return tuple(out_dts) if len(out_dts) > 1 else out_dts[0] else: assert op_call == aten.equal.default, op_call ret = self.wrap(local_results, output_sharding.output_spec) # type: ignore[possibly-undefined] if participating and op_call._schema._is_view_op(): return return_and_correct_aliasing(op_call, args, kwargs, ret) else: return ret @staticmethod def redistribute_local_args( op_info: OpInfo, suggested_input_schema: OpSchema, use_val_from_redistribute_schema: bool, ) -> None: debug_mode = get_active_debug_mode() # NOTE: it's very rare that we need to reshard kwargs so we intentionally skip it if op_info.args_tree_spec is not None: flatten_args_schema_to_reshard = tuple( pytree.tree_leaves(suggested_input_schema.args_schema) ) else: flatten_args_schema_to_reshard = suggested_input_schema.args_schema new_local_args: list[object] = [] for i, arg_spec in enumerate(op_info.flat_args_schema): reshard_arg_spec = flatten_args_schema_to_reshard[i] if isinstance(arg_spec, DTensorSpec): local_tensor = cast(torch.Tensor, op_info.local_args[i]) if arg_spec != reshard_arg_spec: redistribute_context = ( debug_mode.record_redistribute_calls( # type: ignore[union-attr] i, arg_spec, reshard_arg_spec ) if debug_mode is not None else contextlib.nullcontext() ) ExplicitRedistributionContext.observe_redistribution( arg_spec, # pyrefly: ignore [bad-argument-type] reshard_arg_spec, message=f"Implicit redistribution occurred for {op_info.schema} " "while ExplicitRedistributionContext was active", ) with redistribute_context: resharded_local_tensor = redistribute_local_tensor( local_tensor, arg_spec, # pyrefly: ignore [bad-argument-type] reshard_arg_spec, ) new_local_args.append(resharded_local_tensor) else: new_local_args.append(local_tensor) else: if use_val_from_redistribute_schema: # args can be updated for view related ops, we refer to the # update in redistribute_schema. new_local_args.append(reshard_arg_spec) else: new_local_args.append(arg_spec) op_info.local_args = tuple(new_local_args) def unwrap_to_op_info( self, op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], ) -> OpInfo: return self._unwrap_to_op_info_impl(op_call, args, kwargs, True) def _unwrap_to_op_info_impl( self, op_call: torch._ops.OpOverload, args: tuple[object, ...], kwargs: dict[str, object], create_schema: bool, ) -> OpInfo: # get runtime schema info to determine whether to use pytree to flatten inputs runtime_schema_info = self.sharding_propagator.op_to_schema_info.get( op_call, None ) if runtime_schema_info is not None and runtime_schema_info.needs_pytree: # flatten args/kwargs when op says necessary tree_args, args_spec = pytree.tree_flatten(args) args_list: Sequence[object] = tree_args else: args_list, args_spec = args, None args_schema: list[object] = [] kwargs_schema: dict[str, object] = {} local_args: list[object] = [] local_kwargs: dict[str, object] = {} compute_mesh: DeviceMesh | None = None for arg in args_list: if isinstance(arg, dtensor.DTensor): local_args.append(arg._local_tensor) args_schema.append(arg._spec) if compute_mesh is None: # record the first compute device mesh from args compute_mesh = arg.device_mesh elif isinstance(arg, torch.Tensor): compute_mesh = compute_mesh or try_find_mesh_from_args( op_call, args_list ) args_schema.append( self._try_replicate_spec_for_scalar_tensor( op_call, arg, compute_mesh ) ) local_args.append(arg) else: # non DTensor/Tensor args (i.e. int/float/bool), just add to args_schema/local_args args_schema.append(arg) local_args.append(arg) for k, v in kwargs.items(): if isinstance(v, dtensor.DTensor): local_kwargs[k] = v._local_tensor kwargs_schema[k] = v._spec elif isinstance(v, torch.Tensor): compute_mesh = compute_mesh or try_find_mesh_from_args( op_call, args_list ) kwargs_schema[k] = self._try_replicate_spec_for_scalar_tensor( op_call, v, # pyrefly: ignore [bad-argument-type] compute_mesh, ) local_kwargs[k] = v else: # non DTensor/Tensor args (i.e. int/float/bool), just add to args_schema/local_args kwargs_schema[k] = v local_kwargs[k] = v assert compute_mesh is not None, ( f"found no DeviceMesh from dtensor args for {op_call}!" ) op_info = OpInfo( compute_mesh, OpSchema( op_call, ( # pyrefly: ignore [bad-argument-type] pytree.tree_unflatten(args_schema, args_spec) if args_spec else tuple(args_schema) ), kwargs_schema, schema_info=runtime_schema_info, ) if create_schema else None, # type: ignore[arg-type] args_schema, tuple(local_args), local_kwargs, args_spec, ) return op_info @staticmethod def wrap(res: object, spec: OutputSpecType) -> object: if isinstance(res, torch.Tensor): if spec is not None: assert isinstance(spec, DTensorSpec), ( f"output spec does not match with output! Expected DTensorSpec, got {spec}." ) # pyrefly: ignore [bad-argument-type, bad-argument-count, unexpected-keyword] return dtensor.DTensor(res, spec, requires_grad=res.requires_grad) else: # if output does not have a DTensorSpec due to specific ops, it must be a scalar tensor assert res.ndim == 0, "output tensor should be scalar!" return res elif isinstance(res, (list, tuple)): assert spec is not None and isinstance(spec, (list, tuple)), ( f"output spec does not match with output! Expected list/tuple, got {spec}." ) res_list = [] for e, s in zip(res, spec): res_list.append(OpDispatcher.wrap(e, s)) return tuple(res_list) if isinstance(res, tuple) else res_list else: # if the res contains only non tensor values (i.e. int/float/none), we simply return it # without rewrapping to DTensor. return res def _try_replicate_spec_for_scalar_tensor( self, op_call: torch._ops.OpOverload, tensor_arg: torch.Tensor, compute_mesh: DeviceMesh, ) -> DTensorSpec: # util function to produce a replicate spec for a scalar tensor arg/kwarg if tensor_arg.numel() == 1 and tensor_arg.ndim == 1: warnings.warn( "Found a non-scalar tensor with numel=1 and ndim!=0, " "we are implicitly creating a replicated DTensor for it. " "However, please consider changing it to a scalar tensor " "or explicitly create a DTensor under distributed environment.", stacklevel=2, ) if tensor_arg.numel() == 1 or self._allow_implicit_replication: # scalar tensor can be safely treated as replicated replication_spec = DTensorSpec( compute_mesh, (Replicate(),) * compute_mesh.ndim, tensor_meta=TensorMeta( shape=tensor_arg.shape, stride=tensor_arg.stride(), dtype=tensor_arg.dtype, ), ) else: raise RuntimeError( f"{op_call}: got mixed torch.Tensor and DTensor, need to convert all" " torch.Tensor to DTensor before calling distributed operators!" " Please see https://docs.pytorch.org/docs/main/distributed.tensor.html#mixed-tensor-and-dtensor-operations" " for more details." ) return replication_spec