mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-06 12:20:52 +01:00
f68f76d8c7
These seem to have been costing us 5-10 usec per detach (out of ~~95 usec total). If they need to ship let's talk about requirements and how we can make this more efficient given that we would prefer if an entire DTensor op could finish in 10 usec. Differential Revision: [D81530106](https://our.internmc.facebook.com/intern/diff/D81530106) Pull Request resolved: https://github.com/pytorch/pytorch/pull/161596 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: #161591, #161595, #161633, #161634, #161692, #162219, #162220, #162218
510 lines
22 KiB
Python
510 lines
22 KiB
Python
# Copyright (c) Meta Platforms, Inc. and affiliates
|
|
import contextlib
|
|
import functools
|
|
import logging
|
|
import operator
|
|
import warnings
|
|
from collections.abc import Sequence
|
|
from typing import cast, Optional
|
|
|
|
import torch
|
|
import torch.distributed as dist
|
|
import torch.distributed.tensor._api as dtensor
|
|
import torch.distributed.tensor._random as random
|
|
from torch.distributed.device_mesh import DeviceMesh
|
|
from torch.distributed.tensor._dtensor_spec import DTensorSpec, TensorMeta
|
|
from torch.distributed.tensor._op_schema import OpInfo, OpSchema, 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 try_find_mesh_from_args
|
|
from torch.distributed.tensor.placement_types import Partial, Placement, Replicate
|
|
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 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,
|
|
),
|
|
)
|
|
found_inf_dtensor = dtensor.DTensor(
|
|
local_tensor=target_tensor, spec=spec, requires_grad=False
|
|
)
|
|
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()
|
|
self._random_ops = {
|
|
aten.native_dropout.default,
|
|
aten.normal_.default,
|
|
aten.rand_like.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,
|
|
}
|
|
|
|
# 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 dispatch(
|
|
self,
|
|
op_call: torch._ops.OpOverload,
|
|
args: tuple[object, ...],
|
|
kwargs: dict[str, object],
|
|
) -> object:
|
|
"""
|
|
Main dispatching logic. Follows precedence order:
|
|
(1) custom_op_handler
|
|
(2) registered sharding strategy, then rule
|
|
(3) composite implicit autograd decomposition
|
|
"""
|
|
if op_call in self._custom_op_handlers:
|
|
return self._custom_op_handlers[op_call](op_call, args, kwargs) # type: ignore[operator]
|
|
|
|
# extract local tensor and sharding infos to a OpInfo
|
|
op_info = self.unwrap_to_op_info(op_call, args, kwargs)
|
|
|
|
try:
|
|
self.sharding_propagator.propagate(op_info)
|
|
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"Sharding propagation failed for {op_info.schema}"
|
|
) from e
|
|
|
|
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
|
|
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), 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_info.schema.op._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"
|
|
)
|
|
|
|
if output_sharding.output_spec is None:
|
|
if op_call == aten.equal.default:
|
|
# For equal operator, The local results from all devices should be all-gathered
|
|
# and a reduce op (AND) will be performed on the list of results to ensure SPMD
|
|
# execution. We can extend this for more ops if necessary.
|
|
obj_list = [None for _ in range(dist.get_world_size())]
|
|
dist.all_gather_object(obj_list, local_results) # type: ignore[possibly-undefined]
|
|
obj_list = list(filter(lambda x: x is not None, obj_list))
|
|
# perform reduce on the collection with AND op
|
|
local_results = functools.reduce(operator.and_, obj_list, True)
|
|
|
|
if op_info.schema.is_inplace_op():
|
|
# inplace op should return self instead of re-wrapping
|
|
if output_sharding.output_spec is not None:
|
|
# 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:
|
|
output_spec = output_sharding.output_spec
|
|
assert isinstance(output_spec, DTensorSpec)
|
|
assert isinstance(args[0], dtensor.DTensor)
|
|
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:
|
|
return args[0]
|
|
else:
|
|
return None
|
|
elif op_info.schema.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:
|
|
ret = self.wrap(local_results, output_sharding.output_spec) # type: ignore[possibly-undefined]
|
|
if participating and op_info.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:
|
|
# 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:
|
|
resharded_local_tensor = redistribute_local_tensor(
|
|
local_tensor, arg_spec, 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:
|
|
# 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: Optional[DeviceMesh] = 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, 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,
|
|
(
|
|
pytree.tree_unflatten(args_schema, args_spec)
|
|
if args_spec
|
|
else tuple(args_schema)
|
|
),
|
|
kwargs_schema,
|
|
schema_info=runtime_schema_info,
|
|
),
|
|
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}."
|
|
)
|
|
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."
|
|
)
|
|
|
|
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!"
|
|
)
|
|
return replication_spec
|