kfac_block.hpp

Go to the documentation of this file.

// Copyright (c) 2014-2023, Lawrence Livermore National Security, LLC.
// Produced at the Lawrence Livermore National Laboratory.
// Written by the LBANN Research Team (B. Van Essen, et al.) listed in
// the CONTRIBUTORS file. <lbann-dev@llnl.gov>
//
// LLNL-CODE-697807.
// All rights reserved.
//
// This file is part of LBANN: Livermore Big Artificial Neural Network
// Toolkit. For details, see http://software.llnl.gov/LBANN or
// https://github.com/LLNL/LBANN.
//
// Licensed under the Apache License, Version 2.0 (the "Licensee"); 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.

#ifndef LBANN_EXECUTION_ALGORITHMS_KFAC_KFAC_BLOCK_HPP_INCLUDED
#define LBANN_EXECUTION_ALGORITHMS_KFAC_KFAC_BLOCK_HPP_INCLUDED

#include "lbann/execution_algorithms/kfac/execution_context.hpp"
#include "lbann/layers/layer.hpp"

namespace lbann {

// Forward declaration
namespace kfac {
class KFACExecutionContext;
}

namespace kfac {
#if defined AL_HAS_NCCL
using BackendT = ::Al::NCCLBackend;
#elif defined AL_HAS_HOST_TRANSFER
using BackendT = ::Al::HostTransferBackend;
#else
using BackendT = ::Al::MPIBackend;
#endif

using ReqT = typename BackendT::req_type;
} // namespace kfac

template <El::Device Device>
class kfac_block
{
public:
  kfac_block(Layer* layer,
             kfac::KFACExecutionContext* context,
             size_t layer_id,
             size_t inverse_proc_rank,
             bool enable_copy_errors,
             bool enable_copy_activations,
             int input_size,
             int output_size)
    : m_layer(layer),
      m_layer_id(layer_id),
      m_inverse_proc_rank(inverse_proc_rank),
      m_input_size(input_size),
      m_output_size(output_size),
      m_enable_copy_errors(enable_copy_errors),
      m_enable_copy_activations(enable_copy_activations),
      m_context(context)
  {
    m_has_kronecker_inverse = false;
  }
  virtual ~kfac_block() = default;

  virtual void on_forward_prop_end(lbann_comm* comm) {}

  virtual int get_local_memory_consumption() = 0;

  virtual void compute_local_kronecker_factors(lbann_comm* comm,
                                               bool print_matrix,
                                               bool print_matrix_summary);

  virtual const std::vector<El::AbstractMatrix<DataType>*>
  get_local_kronecker_buffers();

  virtual void update_kronecker_average(lbann_comm* comm,
                                        DataType kronecker_decay,
                                        bool print_matrix,
                                        bool print_matrix_summary);

  virtual void update_kronecker_inverse(lbann_comm* comm,
                                        bool use_pi,
                                        DataType damping_act,
                                        DataType damping_err,
                                        DataType learning_rate_factor,
                                        bool use_eigen_decomposition,
                                        bool print_matrix,
                                        bool print_matrix_summary,
                                        bool print_time);

  virtual void compute_preconditioned_gradients(lbann_comm* comm,
                                                DataType learning_rate_factor,
                                                bool print_matrix,
                                                bool print_matrix_summary,
                                                bool print_time);

  virtual void initialize_activations_and_errors(lbann_comm* comm,
                                                 int num_local_activations,
                                                 int num_local_errors,
                                                 int num_weights);

  virtual void start_communication_forward_end(lbann_comm* comm) = 0;
  virtual void end_communication_forward_end(lbann_comm* comm) = 0;
  virtual void start_communication_backward_end(lbann_comm* comm) = 0;
  virtual void end_communication_backward_end(lbann_comm* comm) = 0;

  virtual const std::vector<El::AbstractMatrix<DataType>*>
  get_preconditioned_grad_buffers();

  virtual int get_inverse_matrices(El::Matrix<DataType, Device>& output,
                                   int offset) = 0;

  virtual int get_inverse_matrices_size(lbann_comm* comm) = 0;

  virtual std::vector<int>
  get_inverse_matrices_size_vector(lbann_comm* comm) = 0;

  virtual void resize_inverse_matrices_size(
    El::Matrix<double, El::Device::CPU>& inverse_matrices_size,
    int block_number) = 0;

  virtual int set_inverse_matrices(El::Matrix<DataType, Device>& workspace,
                                   int offset,
                                   lbann_comm* comm) = 0;

  void set_current_batch_size(El::Int batch_size) { m_batch_size = batch_size; }

  virtual std::string get_info() const
  {
    std::ostringstream oss;
    oss << "name=" << m_layer->get_name() << ", id=" << m_layer_id
        << ", type=" << m_layer->get_type()
        << ", inverse_proc_rank=" << m_inverse_proc_rank;
    return oss.str();
  }

  std::string get_name() const { return m_layer->get_name(); }

  size_t get_inverse_proc_rank() const { return m_inverse_proc_rank; }

  DataType* get_local_activation_buffer(int index)
  {
    return m_parent_local_activations[index]->Buffer();
  }

  DataType* get_local_error_buffer(int index)
  {
    return m_child_local_errors[index]->Buffer();
  }

  DataType* get_weight_buffer(int index)
  {
    return m_weight_values[index]->Buffer();
  }

  DataType* get_gradient_wrt_weight_buffer(int index)
  {
    return m_weight_gradients[index]->Buffer();
  }

  El::Int get_current_batch_size() { return m_batch_size; }

  El::Int get_input_size() { return m_input_size; }

  El::Int get_output_size() { return m_output_size; }

  virtual std::vector<std::tuple<std::string, size_t, size_t>>
  get_internal_matrix_info() const;

protected:
  El::Matrix<DataType, Device>&
  get_workspace_matrix(const std::string& key, size_t height, size_t width);

  El::SyncInfo<Device> get_sync_info();

  Layer* m_layer;

  const size_t m_layer_id;

  const int m_inverse_proc_rank;

  std::vector<std::unique_ptr<AbsDistMat>> m_parent_local_activations,
    m_child_local_errors, m_weight_gradients, m_subset_matrix, m_errors_copy,
    m_activations_copy;

  std::vector<std::unique_ptr<AbsDistMat>> m_weight_values;

  std::vector<kfac::ReqT> m_requests_forward_end, m_requests_backward_end;

  int m_input_size, m_output_size, m_batch_size;

  bool m_enable_copy_errors;

  bool m_enable_copy_activations;

  bool m_has_kronecker_inverse;

private:
  kfac::KFACExecutionContext* m_context;
};

} // namespace lbann

#endif // LBANN_EXECUTION_ALGORITHMS_KFAC_KFAC_BLOCK_HPP_INCLUDED