_fsdp_collectives.py

from itertools import chain
from typing import Callable, cast, NamedTuple, Optional, Union

import torch
import torch.distributed as dist
from torch.distributed.device_mesh import _get_device_handle
from torch.distributed.distributed_c10d import _resolve_process_group, ReduceOp
from torch.distributed.tensor import DTensor

from ._fsdp_common import (
    _get_dim0_padded_size,
    _raise_assert_with_print,
    _to_dtype_if_needed,
    compiled_autograd_enabled,
)
from ._fsdp_param import FSDPParam, ShardedState


class AllGatherResult(NamedTuple):
    all_gather_output: torch.Tensor
    all_gather_event: Optional[torch.Event]
    all_gather_work: Optional[dist.distributed_c10d.Work]
    # For each parameter, the all-gather input dtype for each input
    param_all_gather_input_dtypes: list[list[torch.dtype]]
    # For each parameter, the all-gather input numel for each input
    param_all_gather_input_numels: list[list[int]]
    # 1D flattened version of `param_all_gather_input_numels` saved to avoid
    # CPU overhead from recomputing
    all_gather_input_split_sizes: list[int]


def allocate_memory(
    size: int,
    dtype: torch.dtype,
    device: torch.device,
    group: dist.ProcessGroup,
    from_process_group: bool,
) -> torch.Tensor:
    if from_process_group:
        backend = group._get_backend(device)
        if backend.supports_tensor_alloc(device):
            return backend.allocate_tensor(size, dtype=dtype, device=device)
    return torch.empty((size,), dtype=dtype, device=device)


lib = torch.library.Library("fsdp", "FRAGMENT")  # noqa: TOR901

lib.define(
    """
    all_gather_copy_in(
        Tensor[] all_gather_inputs,
        SymInt[] inp_split_sizes,
        SymInt all_gather_input_numel,
        SymInt world_size,
        SymInt rank,
        ScalarType dtype,
        Device device,
        str group_name,
        bool allocate_memory_from_process_group
    ) -> (Tensor, Tensor)
    """
)


@torch.library.impl(lib, "all_gather_copy_in", "Meta")
def all_gather_copy_in_meta(
    all_gather_inputs: list[torch.Tensor],
    inp_split_sizes: list[int],
    all_gather_input_numel: int,
    world_size: int,
    rank: int,
    dtype: torch.dtype,
    device: torch.device,
    group_name: str,
    allocate_memory_from_process_group: bool,
) -> tuple[torch.Tensor, torch.Tensor]:
    all_gather_output = torch.empty(
        (all_gather_input_numel * world_size,), dtype=dtype, device="meta"
    )
    all_gather_input = all_gather_output.narrow(
        0, all_gather_input_numel * rank, all_gather_input_numel
    )
    return all_gather_input, all_gather_output


@torch.library.impl(lib, "all_gather_copy_in", "CUDA")
@torch.library.impl(lib, "all_gather_copy_in", "XPU")
@torch.library.impl(lib, "all_gather_copy_in", "HPU")
@torch.library.impl(lib, "all_gather_copy_in", "CPU")
@torch.library.impl(lib, "all_gather_copy_in", "MTIA")
@torch.library.impl(lib, "all_gather_copy_in", "PrivateUse1")
def all_gather_copy_in_cuda(
    all_gather_inputs: list[torch.Tensor],
    inp_split_sizes: list[int],
    all_gather_input_numel: int,
    world_size: int,
    rank: int,
    dtype: torch.dtype,
    device: torch.device,
    group_name: str,
    allocate_memory_from_process_group: bool,
) -> tuple[torch.Tensor, torch.Tensor]:
    all_gather_output = allocate_memory(
        all_gather_input_numel * world_size,
        dtype=dtype,
        device=device,
        group=_resolve_process_group(group_name),
        from_process_group=allocate_memory_from_process_group,
    )
    all_gather_input = all_gather_output.narrow(
        0, all_gather_input_numel * rank, all_gather_input_numel
    )
    foreach_copy_dsts = torch.split(all_gather_input, inp_split_sizes)
    with torch.no_grad():
        torch._foreach_copy_(foreach_copy_dsts, all_gather_inputs)
    return all_gather_input, all_gather_output


lib.define(
    "split_with_sizes_copy(Tensor all_gather_output, SymInt[] all_gather_input_split_sizes, int dim=0, *, Tensor(a!)[] out) -> ()"
)


@torch.library.impl(lib, "split_with_sizes_copy", "Meta")
@torch.library.impl(lib, "split_with_sizes_copy", "CUDA")
@torch.library.impl(lib, "split_with_sizes_copy", "XPU")
@torch.library.impl(lib, "split_with_sizes_copy", "HPU")
@torch.library.impl(lib, "split_with_sizes_copy", "CPU")
@torch.library.impl(lib, "split_with_sizes_copy", "MTIA")
@torch.library.impl(lib, "split_with_sizes_copy", "PrivateUse1")
def split_with_sizes_copy(
    all_gather_output: torch.Tensor,
    all_gather_input_split_sizes: list[int],
    dim: int,
    out: list[torch.Tensor],
) -> None:
    torch.split_with_sizes_copy(
        all_gather_output, all_gather_input_split_sizes, dim=dim, out=out
    )


lib.define(
    "chunk_cat(Tensor[] tensors, int dim, int num_chunks, *, Tensor(a!) out) -> ()"
)


@torch.library.impl(lib, "chunk_cat", "Meta")
@torch.library.impl(lib, "chunk_cat", "CUDA")
@torch.library.impl(lib, "chunk_cat", "XPU")
@torch.library.impl(lib, "chunk_cat", "HPU")
@torch.library.impl(lib, "chunk_cat", "CPU")
@torch.library.impl(lib, "chunk_cat", "MTIA")
@torch.library.impl(lib, "chunk_cat", "PrivateUse1")
def chunk_cat(
    tensors: list[torch.Tensor],
    dim: int,
    num_chunks: int,
    out: torch.Tensor,
) -> None:
    torch._chunk_cat(tensors, dim, num_chunks, out=out)


@torch.no_grad()
def foreach_all_gather(
    fsdp_params: list[FSDPParam],
    group: dist.ProcessGroup,
    async_op: bool,
    all_gather_copy_in_stream: torch.Stream,
    all_gather_stream: torch.Stream,
    device: torch.device,
    allocate_memory_from_process_group: bool = False,
) -> Optional[AllGatherResult]:
    world_size, rank = group.size(), group.rank()
    device_handle = _get_device_handle(device.type)
    with device_handle.stream(all_gather_copy_in_stream):
        param_all_gather_inputs = _get_param_all_gather_inputs(fsdp_params)
        (
            param_all_gather_input_dtypes,
            param_all_gather_input_numels,
            dtype,
        ) = _get_all_gather_input_metadatas(param_all_gather_inputs)
        if dtype == torch.uint8:
            all_gather_inputs = [
                t.view(torch.uint8) for ts in param_all_gather_inputs for t in ts
            ]
        else:
            all_gather_inputs = [*chain.from_iterable(param_all_gather_inputs)]
        inp_split_sizes = [t.numel() for t in all_gather_inputs]
        all_gather_input_numel = sum(inp_split_sizes)
        all_gather_input, all_gather_output = torch.ops.fsdp.all_gather_copy_in(
            all_gather_inputs,
            inp_split_sizes,
            all_gather_input_numel,
            world_size,
            rank,
            dtype,
            device,
            group.group_name,
            allocate_memory_from_process_group,
        )
        del param_all_gather_inputs
    all_gather_stream.wait_stream(all_gather_copy_in_stream)
    with device_handle.stream(all_gather_stream):
        all_gather_work = dist.all_gather_into_tensor(
            output_tensor=all_gather_output,
            input_tensor=all_gather_input,
            group=group,
            async_op=async_op,
        )
        all_gather_event = all_gather_stream.record_event()
        return AllGatherResult(
            all_gather_output,
            all_gather_event,
            all_gather_work,
            param_all_gather_input_dtypes,
            param_all_gather_input_numels,
            inp_split_sizes,
        )


@torch.no_grad()
def _get_param_all_gather_inputs(
    fsdp_params: list[FSDPParam],
) -> list[list[torch.Tensor]]:
    if compiled_autograd_enabled():
        return [fsdp_param.all_gather_inputs for fsdp_param in fsdp_params]

    # Intentionally try to run a fast-path that bypasses abstractions for the
    # common FSDP case of bf16/fp32 mixed precision in order to use foreach
    # copy for lower CPU overhead and more efficient copying in eager
    def use_foreach_copy(fsdp_param: FSDPParam) -> bool:
        return (
            fsdp_param.param_dtype is not None
            and not fsdp_param.offload_to_cpu
            and not hasattr(fsdp_param._sharded_local_tensor, "fsdp_pre_all_gather")
        )

    param_all_gather_inputs: list[list[torch.Tensor]] = [[] for _ in fsdp_params]
    foreach_copy_indices: list[int] = []
    foreach_copy_inputs: list[torch.Tensor] = []
    foreach_copy_input_numels: list[int] = []

    # 1st pass: for foreach-copy parameters, get inputs and metadata for the
    # foreach copy, and for the others, actually get their all-gather inputs
    for i, fsdp_param in enumerate(fsdp_params):
        if use_foreach_copy(fsdp_param):
            foreach_copy_indices.append(i)
            all_gather_input = (
                fsdp_param._sharded_param_data
                if fsdp_param.sharded_state == ShardedState.SHARDED
                else cast(torch.Tensor, fsdp_param._sharded_post_forward_param_data)
            )
            foreach_copy_inputs.append(all_gather_input)
            foreach_copy_input_numels.append(all_gather_input.numel())
        else:
            param_all_gather_inputs[i] = fsdp_param.all_gather_inputs

    # 2nd pass: use foreach copy to compute the remaining all-gather inputs
    if foreach_copy_inputs:
        fsdp_param_0 = fsdp_params[foreach_copy_indices[0]]
        param_dtype, device = fsdp_param_0.param_dtype, fsdp_param_0.device
        flat_foreach_copy_input = torch.empty(
            (sum(foreach_copy_input_numels),), device=device, dtype=param_dtype
        )
        splits = torch.split(flat_foreach_copy_input, foreach_copy_input_numels)
        torch._foreach_copy_(splits, foreach_copy_inputs)
        for i, split in zip(foreach_copy_indices, splits):
            param_all_gather_inputs[i] = [split]

    return param_all_gather_inputs


@torch.no_grad()
def foreach_all_gather_copy_out(
    all_gather_result: AllGatherResult,
    fsdp_params: list[FSDPParam],
    group: dist.ProcessGroup,
) -> None:
    (
        all_gather_output,
        all_gather_event,
        all_gather_work,
        param_all_gather_input_dtypes,
        param_all_gather_input_numels,
        all_gather_input_split_sizes,
    ) = all_gather_result
    _dtype, device = all_gather_output.dtype, all_gather_output.device
    device_handle = _get_device_handle(device.type)
    if all_gather_event is not None:  # sync op
        device_handle.current_stream().wait_event(all_gather_event)
    if isinstance(all_gather_work, dist.distributed_c10d.Work):  # async op
        all_gather_work.wait()
    world_size, device = group.size(), all_gather_output.device

    split_with_sizes_out: list[torch.Tensor] = []
    shard_i_copy_infos: list[tuple[FSDPParam, list[torch.Tensor]]] = []
    for all_gather_input_numels, all_gather_input_dtypes, fsdp_param in zip(
        param_all_gather_input_numels, param_all_gather_input_dtypes, fsdp_params
    ):
        # NOTE: Under compile, make sure we always recreate all_gather_outputs
        # per AllGather. See [Note: Invariants for torch.compile Traceable FSDP2].
        force_recreate = compiled_autograd_enabled()
        fsdp_param.init_all_gather_outputs(
            all_gather_input_numels,
            all_gather_input_dtypes,
            world_size,
            device,
            force_recreate=force_recreate,
        )
        if not force_recreate:
            fsdp_param.alloc_all_gather_outputs()
        param_all_gather_outputs = fsdp_param.all_gather_outputs
        if fsdp_param.fsdp_placement.dim != 0:
            # Copy to a temporary and then chunk-cat into the final all-gather
            # output tensors
            param_all_gather_outputs = [
                torch.empty_like(t) for t in param_all_gather_outputs
            ]
            shard_i_copy_infos.append((fsdp_param, param_all_gather_outputs))
        split_with_sizes_out.extend(param_all_gather_outputs)

    all_gather_output = all_gather_output.view(world_size, -1)
    if all_gather_output.dtype == torch.uint8:
        out = [t.view(world_size, -1).view(torch.uint8) for t in split_with_sizes_out]
    else:
        out = [t.view(world_size, -1) for t in split_with_sizes_out]

    # only avoid VC bump if we are not in inference mode
    if torch._dynamo.is_compiling():
        # For torch.compile, we turn off inference_mode for fake tensor
        # propagation, and therefore graph break on is_inference. For `compile`,
        # we don't care about VCs, so just skip the optimization.
        non_inference_outs = []
    else:
        non_inference_outs = [o for o in out if not o.is_inference()]

    if len(non_inference_outs) > 0:
        with torch.autograd._unsafe_preserve_version_counter(tuple(non_inference_outs)):
            torch.ops.fsdp.split_with_sizes_copy(
                all_gather_output, all_gather_input_split_sizes, dim=1, out=out
            )
    else:
        torch.ops.fsdp.split_with_sizes_copy(
            all_gather_output, all_gather_input_split_sizes, dim=1, out=out
        )

    for fsdp_param, param_all_gather_outputs in shard_i_copy_infos:
        # Chunk-cat from the temporary to the final all-gather output tensors
        shard_dim = fsdp_param.fsdp_placement.dim

        with torch.autograd._unsafe_preserve_version_counter(
            tuple(fsdp_param.all_gather_outputs)
        ):
            for param_all_gather_output, target_all_gather_output in zip(
                param_all_gather_outputs, fsdp_param.all_gather_outputs
            ):
                padded_sharded_size = (
                    fsdp_param.padded_sharded_param_size
                    if fsdp_param.sharded_state == ShardedState.SHARDED
                    else cast(
                        torch.Tensor, fsdp_param._sharded_post_forward_param_data
                    ).size()
                )
                pre_param_size = list(padded_sharded_size)
                pre_param_size[0] *= world_size
                chunks = torch.chunk(
                    param_all_gather_output.view(pre_param_size), world_size, dim=0
                )
                post_param_size = list(padded_sharded_size)
                post_param_size[shard_dim] *= world_size
                cat_out = target_all_gather_output.view(post_param_size)
                torch.cat(chunks, dim=shard_dim, out=cat_out)


@torch.no_grad()
def foreach_reduce(
    fsdp_params: list[FSDPParam],
    unsharded_grads: list[torch.Tensor],
    reduce_scatter_group: dist.ProcessGroup,
    reduce_scatter_stream: torch.Stream,
    orig_dtype: Optional[torch.dtype],
    reduce_dtype: Optional[torch.dtype],
    device: torch.device,
    gradient_divide_factor: Optional[float],
    all_reduce_group: Optional[dist.ProcessGroup],  # not `None` iff HSDP
    all_reduce_stream: torch.Stream,
    all_reduce_grads: bool,
    partial_reduce_output: Optional[torch.Tensor],  # only used for HSDP
    all_reduce_hook: Optional[Callable[[torch.Tensor], None]],
    allocate_memory_from_process_group: bool = False,
    force_sum_reduction_for_comms: bool = False,
) -> tuple[
    torch.Tensor,
    torch.Event,
    torch.Event,
    Optional[torch.Tensor],
    Optional[torch.Event],
    Optional[torch.Tensor],
]:
    """
    ``unsharded_grads`` owns the references to the gradients computed by
    autograd, so clearing the list frees the gradients.
    """
    grad_dtypes = {grad.dtype for grad in unsharded_grads}
    if len(grad_dtypes) != 1:
        # Check this at runtime since it could be a real runtime error if e.g.
        # fp8 weights do not produce the correct higher precision gradients
        _raise_assert_with_print(
            f"FSDP reduce-scatter expects uniform gradient dtype but got {grad_dtypes}"
        )
    grad_dtype = unsharded_grads[0].dtype
    reduce_dtype = reduce_dtype or grad_dtype
    (predivide_factor, postdivide_factor, reduce_scatter_op, all_reduce_op) = (
        _get_gradient_divide_factors(
            reduce_scatter_group,
            all_reduce_group,
            reduce_dtype,
            device.type,
            gradient_divide_factor,
            force_sum_reduction_for_comms,
        )
    )
    world_size = reduce_scatter_group.size()
    for i, (fsdp_param, unsharded_grad) in enumerate(zip(fsdp_params, unsharded_grads)):
        if (shard_dim := fsdp_param.fsdp_placement.dim) == 0:
            continue
        assert unsharded_grad.size(shard_dim) % world_size == 0, (
            f"Shard({shard_dim}) requires even sharding: {unsharded_grad.size()=} {world_size=}"
        )
        chunks = torch.chunk(unsharded_grad, world_size, dim=shard_dim)
        unsharded_grads[i] = torch.cat(chunks, dim=0)
    padded_unsharded_sizes = tuple(
        _get_dim0_padded_size(grad.size(), world_size) for grad in unsharded_grads
    )
    reduce_scatter_input_numel = sum(s.numel() for s in padded_unsharded_sizes)
    reduce_scatter_output_numel = reduce_scatter_input_numel // world_size
    reduce_scatter_input = allocate_memory(
        reduce_scatter_input_numel,
        dtype=reduce_dtype,
        device=device,
        group=reduce_scatter_group,
        from_process_group=allocate_memory_from_process_group,
    )
    device_handle = _get_device_handle(device.type)
    foreach_reduce_scatter_copy_in(unsharded_grads, reduce_scatter_input, world_size)
    current_stream = device_handle.current_stream()
    # Only after the copy-in finishes can we free the gradients
    unsharded_grads.clear()
    reduce_scatter_stream.wait_stream(current_stream)
    all_reduce_input = None
    all_reduce_event = None
    with device_handle.stream(reduce_scatter_stream):
        reduce_output = allocate_memory(
            reduce_scatter_output_numel,
            dtype=reduce_dtype,
            device=device,
            group=reduce_scatter_group,
            from_process_group=allocate_memory_from_process_group,
        )
        _div_if_needed(reduce_scatter_input, predivide_factor)
        dist.reduce_scatter_tensor(
            output=reduce_output,
            input=reduce_scatter_input,
            group=reduce_scatter_group,
            op=reduce_scatter_op,
        )
        reduce_scatter_event = reduce_scatter_stream.record_event()
        post_reduce_stream = reduce_scatter_stream
        if all_reduce_group is not None:  # HSDP
            # Accumulations must run in the reduce-scatter stream
            if not all_reduce_grads:
                if partial_reduce_output is not None:
                    partial_reduce_output += reduce_output
                else:
                    partial_reduce_output = reduce_output
                return (
                    reduce_scatter_input,
                    reduce_scatter_event,
                    post_reduce_stream.record_event(),
                    all_reduce_input,
                    all_reduce_event,
                    partial_reduce_output,
                )
            if partial_reduce_output is not None:
                reduce_output += partial_reduce_output
            post_reduce_stream = all_reduce_stream
            all_reduce_stream.wait_stream(reduce_scatter_stream)
            with device_handle.stream(all_reduce_stream):
                dist.all_reduce(
                    reduce_output,
                    group=all_reduce_group,
                    op=all_reduce_op,
                )
                all_reduce_input = reduce_output
                all_reduce_event = all_reduce_stream.record_event()
    # -- END: ops in reduce_scatter stream

    if all_reduce_hook is not None:
        # Execute user-specified all reduce hook.
        # If native HSDP is used, this is executed after the HSDP all reduce.
        # If 1-d FSDP is used, this is executed post reduce-scatter.
        post_reduce_stream = all_reduce_stream
        all_reduce_stream.wait_stream(reduce_scatter_stream)
        with device_handle.stream(all_reduce_stream):
            all_reduce_hook(reduce_output)
    # -- END: ops post reduce_scatter

    with device_handle.stream(post_reduce_stream):
        _div_if_needed(reduce_output, postdivide_factor)
        reduce_output = _to_dtype_if_needed(reduce_output, orig_dtype)
        # View out and accumulate sharded gradients
        flat_grad_offset = 0  # [0, reduce_scatter_output_numel - 1]
        for padded_unsharded_size, fsdp_param in zip(
            padded_unsharded_sizes, fsdp_params
        ):
            # Assume even sharding for Shard(i), i > 0; otherwise would require
            # copy-out for contiguous strides
            new_sharded_grad = torch.as_strided(
                reduce_output,
                size=fsdp_param.sharded_size,
                stride=fsdp_param.contiguous_sharded_stride,
                storage_offset=flat_grad_offset,
            )
            to_accumulate_grad = fsdp_param.sharded_param.grad is not None
            if fsdp_param.offload_to_cpu:
                # Only overlap the D2H copy (copying to pinned memory) if not
                # accumulating gradients since the CPU add kernel depends on
                # the copy result and we cannot run the add as a callback
                non_blocking = fsdp_param.pin_memory and not to_accumulate_grad
                # Since the GPU sharded gradient is allocated in the RS stream,
                # we can free it here by not keeping a ref without waiting for
                # the D2H copy since future RS-stream ops run after the copy
                new_sharded_grad = new_sharded_grad.to(
                    torch.device("cpu"), non_blocking=non_blocking
                )
                if non_blocking:
                    # Record an event on which to block the CPU thread to
                    # ensure that the D2H copy finishes before the optimizer
                    fsdp_param.grad_offload_event = reduce_scatter_stream.record_event()
            if to_accumulate_grad:
                assert isinstance(fsdp_param.sharded_param.grad, DTensor)
                fsdp_param.sharded_param.grad._local_tensor += new_sharded_grad
            else:
                new_sharded_dtensor_grad = fsdp_param.to_sharded_dtensor(
                    new_sharded_grad
                )
                fsdp_param.sharded_param.grad = new_sharded_dtensor_grad
            if not compiled_autograd_enabled():
                for hook in (
                    getattr(fsdp_param.sharded_param, "_post_accumulate_grad_hooks", {})
                    or {}
                ).values():
                    hook(fsdp_param.sharded_param)
            padded_sharded_numel = padded_unsharded_size.numel() // world_size
            flat_grad_offset += padded_sharded_numel
        post_reduce_event = post_reduce_stream.record_event()
    # The RS output is allocated in the RS stream and used in the default
    # stream (for optimizer). To ensure its memory is not reused for later
    # RSs, we do not need extra synchronization since the sharded parameters
    # hold refs through the end of backward.
    return (
        reduce_scatter_input,
        reduce_scatter_event,
        post_reduce_event,
        all_reduce_input,
        all_reduce_event,
        None,
    )


def foreach_reduce_scatter_copy_in(
    unsharded_grads: list[torch.Tensor],
    reduce_scatter_input: torch.Tensor,
    world_size: int,
) -> None:
    reduce_scatter_input = reduce_scatter_input.view(world_size, -1)
    torch.ops.fsdp.chunk_cat(
        unsharded_grads, dim=0, num_chunks=world_size, out=reduce_scatter_input
    )


def _get_all_gather_input_metadatas(
    param_all_gather_inputs: list[list[torch.Tensor]],
) -> tuple[list[list[torch.dtype]], list[list[int]], torch.dtype]:
    param_all_gather_input_dtypes: list[list[torch.dtype]] = []
    param_all_gather_input_numels: list[list[int]] = []
    all_gather_dtype = param_all_gather_inputs[0][0].dtype
    for all_gather_inputs in param_all_gather_inputs:
        input_dtypes: list[torch.dtype] = []
        input_numels: list[int] = []
        for all_gather_input in all_gather_inputs:
            if all_gather_input.dtype != all_gather_dtype:
                all_gather_dtype = torch.uint8
            input_dtypes.append(all_gather_input.dtype)
            input_numels.append(all_gather_input.numel())
        param_all_gather_input_dtypes.append(input_dtypes)
        param_all_gather_input_numels.append(input_numels)
    return (
        param_all_gather_input_dtypes,
        param_all_gather_input_numels,
        all_gather_dtype,
    )


def _get_gradient_divide_factors(
    reduce_scatter_group: dist.ProcessGroup,
    all_reduce_group: Optional[dist.ProcessGroup],
    reduce_dtype: torch.dtype,
    device_type: str = "",
    factor: Optional[float] = None,
    force_sum_reduction_for_comms: bool = False,
) -> tuple[
    Optional[float],
    Optional[float],
    Union[dist.ReduceOp, dist.ReduceOp.RedOpType],
    Union[dist.ReduceOp, dist.ReduceOp.RedOpType],
]:
    # MTIA appears to only support SUM reduction, hence we force it implicitly
    if device_type == "mtia":
        force_sum_reduction_for_comms = True

    # For fp32/bf16, we do not need to worry about overflow/underflow, so we
    # use NCCL's built-in division to avoid separate div kernels
    overflow_risk = reduce_dtype not in (torch.float32, torch.bfloat16)

    data_parallel_size = reduce_scatter_group.size()
    if all_reduce_group is not None:
        data_parallel_size *= all_reduce_group.size()

    if factor is None:
        factor = float(data_parallel_size)

    if not overflow_risk and not force_sum_reduction_for_comms:
        if factor == data_parallel_size:
            # Warning: NCCL ReduceOp.AVG may produce incorrect results with
            # world size 1.
            return None, None, ReduceOp.AVG, ReduceOp.AVG
        else:
            reduce_scatter_op = torch.distributed._make_nccl_premul_sum(1 / factor)
            return None, None, reduce_scatter_op, ReduceOp.SUM

    pre_factor: Optional[float]
    if overflow_risk:
        # Since fp16 has smaller dynamic range than fp32/bf16, we want to avoid
        # overflow/underflow. For N data parallel workers, each worker computes
        # g_i, and they collectively reduce (g_1 + ... + g_N) / N. To avoid
        # overflow/underflow, we divide by ~sqrt(N) before/after the reduction.
        pre_factor = 1
        while factor % pre_factor == 0 and factor / pre_factor > pre_factor:
            pre_factor *= 2
        post_factor = factor / pre_factor
    else:
        # Prefer post-multiplying as it operates on less data and is thus faster
        pre_factor, post_factor = None, factor

    return pre_factor, post_factor, ReduceOp.SUM, ReduceOp.SUM


def _div_if_needed(tensor: torch.Tensor, div_factor: Optional[float]) -> None:
    if div_factor is not None and div_factor != 1:
        tensor.div_(div_factor)