tensor_queue.cc

// Copyright 2019 Uber Technologies, Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// =============================================================================

#include "tensor_queue.h"

#include <assert.h>

#include "logging.h"

namespace horovod {
namespace common {

// Add a TensorTableEntry as well as its message to the queue.
Status TensorQueue::AddToTensorQueue(TensorTableEntry& e, Request& message) {
  std::lock_guard<std::mutex> guard(mutex_);
  if (tensor_table_.find(e.tensor_name) != tensor_table_.end()) {
    return DUPLICATE_NAME_ERROR;
  }
  tensor_table_.emplace(e.tensor_name, std::move(e));
  message_queue_.push(message);
  return Status::OK();
}

// Put callbacks for each tensor in the callback buffer and clear tensor queue
void TensorQueue::FinalizeTensorQueue(
    std::vector<StatusCallback>& callbacks_buffer) {
  std::lock_guard<std::mutex> guard(mutex_);
  for (auto& e : tensor_table_) {
    callbacks_buffer.emplace_back(e.second.callback);
  }
  tensor_table_.clear();
  while (!message_queue_.empty()) {
    message_queue_.pop();
  }
}

// Helper function to get list of allreduced tensor names and total size for
// use with the autotuner.
int64_t
TensorQueue::GetTensorDataForAutotuner(const ResponseList& response_list,
                                       std::vector<std::string>& tensor_names) {
  int64_t total_tensor_size = 0;
  for (auto& response : response_list.responses()) {
    if (response.response_type() == Response::ResponseType::ALLREDUCE) {
      for (auto& tensor_name : response.tensor_names()) {
        tensor_names.push_back(tensor_name);
        LOG(TRACE) << "Looking for tensor with name " << tensor_name;
        auto& entry = tensor_table_.at(tensor_name);
        LOG(TRACE) << "Found tensor with name " << tensor_name;
        total_tensor_size += entry.tensor->size();
      }
    }
  }
  return total_tensor_size;
}

// Parse tensor names from response and generate a vector of corresponding
// tensor entries.
void TensorQueue::GetTensorEntriesFromResponse(
    Response& response, std::vector<TensorTableEntry>& entries) {
  // Reserve to save re-allocation costs, as we know the size before.
  entries.reserve(response.tensor_names().size());
  {
    // Lock on the tensor table.
    std::lock_guard<std::mutex> guard(mutex_);
    for (auto& name : response.tensor_names()) {
      // We should never fail at finding this key in the tensor table.
      auto iter = tensor_table_.find(name);
      assert(iter != tensor_table_.end());

      assert(response.response_type() == Response::ALLREDUCE ||
             response.response_type() == Response::ALLGATHER ||
             response.response_type() == Response::BROADCAST ||
             response.response_type() == Response::ERROR);

      entries.push_back(std::move(iter->second));

      // Clear the tensor table of this tensor.
      tensor_table_.erase(iter);
    }
  }
}

// Get tensor entry given a tensor name
const TensorTableEntry&
TensorQueue::GetTensorEntry(const std::string& tensor_name) const{
  // Lock on the tensor table.
  std::lock_guard<std::mutex> guard(mutex_);
  auto& iter = tensor_table_.at(tensor_name);

  return iter;
}

// Pop out all the messages from the queue
void TensorQueue::PopMessagesFromQueue(
    std::deque<Request>& message_queue_buffer) {
  std::lock_guard<std::mutex> guard(mutex_);
  while (!message_queue_.empty()) {
    Request message = message_queue_.front();
    message_queue_.pop();
    message_queue_buffer.push_back(std::move(message));
  }
}

// Push a message to massage queue
void TensorQueue::PushMessageToQueue(Request& message) {
  std::lock_guard<std::mutex> guard(mutex_);
  message_queue_.push(std::move(message));
}

} // namespace common
} // namespace horovod