// Ceres Solver - A fast non-linear least squares minimizer
// Copyright 2022 Google Inc. All rights reserved.
// http://ceres-solver.org/
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// * Redistributions of source code must retain the above copyright notice,
//   this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above copyright notice,
//   this list of conditions and the following disclaimer in the documentation
//   and/or other materials provided with the distribution.
// * Neither the name of Google Inc. nor the names of its contributors may be
//   used to endorse or promote products derived from this software without
//   specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
// Author: sameeragarwal@google.com (Sameer Agarwal)

#include "ceres/block_random_access_sparse_matrix.h"

#include <algorithm>
#include <memory>
#include <set>
#include <utility>
#include <vector>

#include "ceres/internal/export.h"
#include "ceres/triplet_sparse_matrix.h"
#include "ceres/types.h"
#include "glog/logging.h"

namespace ceres::internal {

BlockRandomAccessSparseMatrix::BlockRandomAccessSparseMatrix(
    const std::vector<Block>& blocks,
    const std::set<std::pair<int, int>>& block_pairs)
    : kMaxRowBlocks(10 * 1000 * 1000), blocks_(blocks) {
  CHECK_LT(blocks.size(), kMaxRowBlocks);

  const int num_cols = NumScalarEntries(blocks);

  // Count the number of scalar non-zero entries and build the layout
  // object for looking into the values array of the
  // TripletSparseMatrix.
  int num_nonzeros = 0;
  for (const auto& block_pair : block_pairs) {
    const int row_block_size = blocks_[block_pair.first].size;
    const int col_block_size = blocks_[block_pair.second].size;
    num_nonzeros += row_block_size * col_block_size;
  }

  VLOG(1) << "Matrix Size [" << num_cols << "," << num_cols << "] "
          << num_nonzeros;

  tsm_ =
      std::make_unique<TripletSparseMatrix>(num_cols, num_cols, num_nonzeros);
  tsm_->set_num_nonzeros(num_nonzeros);
  int* rows = tsm_->mutable_rows();
  int* cols = tsm_->mutable_cols();
  double* values = tsm_->mutable_values();

  int pos = 0;
  for (const auto& block_pair : block_pairs) {
    const int row_block_size = blocks_[block_pair.first].size;
    const int col_block_size = blocks_[block_pair.second].size;
    cell_values_.emplace_back(block_pair, values + pos);
    layout_[IntPairToLong(block_pair.first, block_pair.second)] =
        new CellInfo(values + pos);
    pos += row_block_size * col_block_size;
  }

  // Fill the sparsity pattern of the underlying matrix.
  for (const auto& block_pair : block_pairs) {
    const int row_block_id = block_pair.first;
    const int col_block_id = block_pair.second;
    const int row_block_size = blocks_[row_block_id].size;
    const int col_block_size = blocks_[col_block_id].size;
    int pos =
        layout_[IntPairToLong(row_block_id, col_block_id)]->values - values;
    for (int r = 0; r < row_block_size; ++r) {
      for (int c = 0; c < col_block_size; ++c, ++pos) {
        rows[pos] = blocks_[row_block_id].position + r;
        cols[pos] = blocks_[col_block_id].position + c;
        values[pos] = 1.0;
        DCHECK_LT(rows[pos], tsm_->num_rows());
        DCHECK_LT(cols[pos], tsm_->num_rows());
      }
    }
  }
}

// Assume that the user does not hold any locks on any cell blocks
// when they are calling SetZero.
BlockRandomAccessSparseMatrix::~BlockRandomAccessSparseMatrix() {
  for (const auto& entry : layout_) {
    delete entry.second;
  }
}

CellInfo* BlockRandomAccessSparseMatrix::GetCell(int row_block_id,
                                                 int col_block_id,
                                                 int* row,
                                                 int* col,
                                                 int* row_stride,
                                                 int* col_stride) {
  const auto it = layout_.find(IntPairToLong(row_block_id, col_block_id));
  if (it == layout_.end()) {
    return nullptr;
  }

  // Each cell is stored contiguously as its own little dense matrix.
  *row = 0;
  *col = 0;
  *row_stride = blocks_[row_block_id].size;
  *col_stride = blocks_[col_block_id].size;
  return it->second;
}

// Assume that the user does not hold any locks on any cell blocks
// when they are calling SetZero.
void BlockRandomAccessSparseMatrix::SetZero() {
  if (tsm_->num_nonzeros()) {
    VectorRef(tsm_->mutable_values(), tsm_->num_nonzeros()).setZero();
  }
}

void BlockRandomAccessSparseMatrix::SymmetricRightMultiplyAndAccumulate(
    const double* x, double* y) const {
  for (const auto& cell_position_and_data : cell_values_) {
    const int row = cell_position_and_data.first.first;
    const int row_block_size = blocks_[row].size;
    const int row_block_pos = blocks_[row].position;

    const int col = cell_position_and_data.first.second;
    const int col_block_size = blocks_[col].size;
    const int col_block_pos = blocks_[col].position;

    MatrixVectorMultiply<Eigen::Dynamic, Eigen::Dynamic, 1>(
        cell_position_and_data.second,
        row_block_size,
        col_block_size,
        x + col_block_pos,
        y + row_block_pos);

    // Since the matrix is symmetric, but only the upper triangular
    // part is stored, if the block being accessed is not a diagonal
    // block, then use the same block to do the corresponding lower
    // triangular multiply also.
    if (row != col) {
      MatrixTransposeVectorMultiply<Eigen::Dynamic, Eigen::Dynamic, 1>(
          cell_position_and_data.second,
          row_block_size,
          col_block_size,
          x + row_block_pos,
          y + col_block_pos);
    }
  }
}

}  // namespace ceres::internal