torch/csrc/distributed/c10d/reducer.h

#pragma once

#include <atomic>
#include <memory>
#include <mutex>
#include <tuple>
#include <unordered_map>
#include <vector>

#include <c10d/ProcessGroup.hpp>
#include <torch/csrc/autograd/function.h>
#include <torch/csrc/autograd/variable.h>
#include <torch/csrc/distributed/autograd/context/context.h>
#include <torch/csrc/distributed/c10d/comm.h>

namespace c10d {

constexpr int kDefaultFirstBucketBytes = int(1024 * 1024);
constexpr int kDefaultBucketBytesCap = int(25 * 1024 * 1024);

class Reducer {
 public:
  // The constructor takes a list of variables for every model replica.
  // The bucket assignment for this reducer is specified as a list of
  // buckets, each of which is specified as a list of indices into the
  // variables list for **a single replica** (i.e. `variables[0]`).
  explicit Reducer(
      std::vector<std::vector<torch::autograd::Variable>> replicas,
      std::vector<std::vector<size_t>> bucket_indices,
      std::shared_ptr<c10d::ProcessGroup> process_group,
      std::vector<std::vector<bool>> expect_sparse_gradients,
      int64_t bucket_bytes_cap,
      bool find_unused_parameters);

  ~Reducer() noexcept(false);

  // To (re-)initialize bucket assignment, pass a list of buckets, each
  // of which is specified by a list of indices in the variables list.
  // This function performs validation that the variables within a bucket
  // all live on the same device and have the same dimensionality.
  void initialize_buckets(std::vector<std::vector<size_t>> bucket_indices);

  // This function is called when the forward function has produced an output,
  // and the user wishes to reduce gradients in the backwards pass.
  // If they don't, and wish to accumulate gradients before reducing them,
  // a call to this function can simply be omitted.
  void prepare_for_backward(
      const std::vector<torch::autograd::Variable>& outputs);

  // Returns the relative time in nanoseconds when gradients were ready,
  // with respect to the time `prepare_for_backward` was called. The outer
  // vector is for model replicas and the inner vector is for parameters.
  std::vector<std::vector<int64_t>> get_backward_stats() const {
    return backward_stats_;
  }

  // Registeres a hook to the reducer. The hook is `CommHookInterface`
  // type to allow both Python and CPP hooks. This function can only
  // be called once before calling backward.
  void register_comm_hook(std::unique_ptr<CommHookInterface> iface);

 protected:
  // Forward declaration.
  struct Bucket;

  // Locates a specific variable by replica index and variable index.
  struct VariableIndex {
    size_t replica_index;
    size_t variable_index;
  };

  std::mutex mutex_;
  std::vector<std::vector<torch::autograd::Variable>> replicas_;
  std::shared_ptr<c10d::ProcessGroup> process_group_;
  std::vector<std::vector<bool>> expect_sparse_gradients_;

  std::vector<std::vector<std::shared_ptr<torch::autograd::Node>>>
      grad_accumulators_;
  std::unordered_map<torch::autograd::Node*, VariableIndex> func_;
  std::vector<std::pair<uintptr_t, std::shared_ptr<torch::autograd::Node>>>
      hooks_;

  bool expect_autograd_hooks_;
  bool require_finalize_;
  size_t next_bucket_;

  bool has_marked_unused_parameters_;
  const bool find_unused_parameters_;
  std::vector<VariableIndex> unused_parameters_;
  // Locally used parameter maps indicating if parameters are used locally
  // during the current iteration or no_sync session if no_sync is on. One
  // tensor for each model replica and each tensor is one-dim int32 tensor of
  // number of parameters. These tensors are marked in autograd_hook to indicate
  // the corresponding param has been used, and get allreduced in the end of
  // backward of current iteration or no_sync session for figuring out the
  // globally unused parameters.
  //
  // local_used_maps_:     CPU tensors for bookkeeping locally used params
  // local_used_maps_dev_: dev tensors for reducing globally unused params
  std::vector<at::Tensor> local_used_maps_;
  std::vector<at::Tensor> local_used_maps_dev_;
  // Indicate that reduction is done and D2H copy is done as well.
  bool local_used_maps_reduced_;

  // Work handle for allreduce on local_used_maps_
  std::shared_ptr<c10d::ProcessGroup::Work> local_used_work_;

  void verify_replicas_within_process();

  void verify_replica0_across_processes();

  void mark_variable_ready_dense(VariableIndex index);

  void mark_variable_ready_sparse(VariableIndex index);

  void mark_variable_ready(VariableIndex index);

  void autograd_hook(VariableIndex index);

  void mark_bucket_ready(size_t bucket_index);

  void finalize_bucket_dense(Bucket& replica);

  void finalize_backward();

  // Broadcast rebuilt buckets from rank 0 to other ranks before initializing
  // the buckets
  void sync_bucket_indices(std::vector<std::vector<size_t>>& bucket_indices);
  // Rebuild buckets based on rebuilt_params_ and rebuilt_param_indices_
  // TODO this function makes broadcast communication call and
  // could be overlapped with next forward() call, thus
  // it could be async. Will make it async when rebuilding buckets for
  // find_unused_parameters = true case, as we could rebuild buckets more than
  // once for find_unused_parameters = true case, where subgraphs are trained
  // and parameter indices order may change more frequently.
  // For find_unused_parameters = false case, buckets are only rebuilt once,
  // the performance cost is negligible.
  std::vector<std::vector<size_t>> rebuildBuckets();

  using GradCallback =
      torch::distributed::autograd::DistAutogradContext::GradCallback;
  void runGradCallbackForVariable(
      torch::autograd::Variable& variable,
      GradCallback&& cb);

  // A bucket replica represents [1..N] gradients to be reduced,
  // with the same dtype, on the same device.
  //
  // Batching gradients together before reducing them can result in lower
  // overhead and/or faster time to completion. Only gradients of the same type
  // and on the same device can be batched. The tensor that represents the
  // flattened gradient uses the same type and is placed on the same device.
  // Buckets are filled as the gradients they hold are computed (triggered by
  // autograd hooks). Buckets are reduced in a predetemined order that is
  // identical across processes.
  //
  struct BucketReplica {
    // Flattened (1 dimensional) contents of bucket.
    at::Tensor contents;

    // Views into contents for each grad.  Each view will be created with
    // layout (sizes + strides) matching the grad's expected layout
    // ("Gradient Layout Contract" in torch/csrc/autograd/AccumulateGrad.h).
    // grad.copy_(bucket_views[i]) and
    // bucket_views[i].copy_(grad)
    // provide convenient ways to move grad data in/out of contents.
    std::vector<at::Tensor> bucket_views;

    // Variables that contribute to this bucket replica. Use refcounted value
    // here so that we can easily unflatten the bucket contents into the
    // participating variables after reduction has completed.
    std::vector<torch::autograd::Variable> variables;

    // Per-variable offset/length into the flat bucket contents tensor.
    std::vector<size_t> offsets;
    std::vector<size_t> lengths;

    // Number of tensors to be added before this bucket is complete.
    // This is reset to `variables.size()` every iteration.
    size_t pending;

    // TODO(@pietern)
    // Memory copies from gradient tensors into the bucket are potentially
    // done on different CUDA streams. We record an event for every copy
    // so that we can synchronize with them prior to kicking off the reduction.
    // std::vector<at::cuda::CUDAEvent> events;
  };

  // This function is called inside `initialize_buckets` and
  // `finalize_backward`. The function call in `initialize_bucket` creates views
  // into the contents tensor for each variable's grad. Views serve as entry
  // points to copy_ each grad's data in/out of the flat contents tensor. The
  // function call in `finalize_backward` happens only if DDP communication hook
  // was registered to recrate views with the result of `future_work`. Before
  // `finalize_backward` call, views must be cleared.
  void initialize_bucketviews(BucketReplica& replica, at::Tensor& contents);

  // A bucket holds N bucket replicas (1 per model replica).
  //
  // If every bucket in this struct is ready, the reduction can be kicked off.
  // One bucket per replica. Reduction is kicked off when every bucket is ready.
  //
  struct Bucket {
    std::vector<BucketReplica> replicas;

    // Global indices of participating variables in the bucket
    std::vector<size_t> variable_indices;

    // Number of replicas to be marked done before this bucket is ready.
    size_t pending;

    // Keep work handle around when this set of buckets is being reduced.
    std::shared_ptr<c10d::ProcessGroup::Work> work;

    // Keep future work handle around if DDP comm hook is registered.
    c10::intrusive_ptr<torch::jit::Future> future_work;

    // If this bucket should expect a single sparse gradient.
    // Implies: replicas[i].variables.size() == 1.
    bool expect_sparse_gradient = false;
  };

  std::vector<Bucket> buckets_;

  // A variable locator locates a particular variable in the bucket
  // structure. The `bucket_index` field points to the bucket in the `buckets_`
  // vector. The `intra_bucket_index` field points to the index of the variable
  // in any of the vector fields in the bucket replica.
  struct VariableLocator {
    // Index into the `buckets_` variable.
    size_t bucket_index;
    // Index of parameter in single bucket replica.
    size_t intra_bucket_index;
  };

  // Map the index of a variable to its location in the bucket structure.
  std::vector<VariableLocator> variable_locators_;

  // We collect the relative timestamp of every gradient being ready
  // when executing autograd. This can be used to derive a timeline of
  // the point in time buckets were ready, or ideal bucket assignment/ordering.
  int64_t backward_stats_base_;
  std::vector<std::vector<int64_t>> backward_stats_;

  // Following variables are to help build dynamic bucket order
  bool has_rebuilt_bucket_;
  std::vector<at::Tensor> rebuilt_params_;
  std::vector<int64_t> rebuilt_param_indices_;
  const int64_t bucket_bytes_cap_;

  struct RpcContext {
    using ContextPtr = torch::distributed::autograd::ContextPtr;
    // The shared_ptr is to hold the context instance.
    ContextPtr context_ptr_holder;
    std::atomic<ContextPtr::element_type*> context_ptr{nullptr};

    void set(ContextPtr&& new_context_ptr);
  };
  RpcContext rpc_context_;

 private:
  // comm_hook_ is used to access the DDP communication hook if registered.
  std::unique_ptr<CommHookInterface> comm_hook_;
};

std::vector<std::vector<size_t>> compute_bucket_assignment_by_size(
    const std::vector<at::Tensor>& tensors,
    const std::vector<size_t>& bucket_size,
    const std::vector<bool>& expect_sparse_gradient = {},
    const std::vector<int64_t>& tensor_indices = {});

} // namespace c10d