ceres-solver/internal/ceres/sparse_normal_cholesky_solver.cc

// Ceres Solver - A fast non-linear least squares minimizer
// Copyright 2010, 2011, 2012 Google Inc. All rights reserved.
// http://code.google.com/p/ceres-solver/
//
// 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)

// This include must come before any #ifndef check on Ceres compile options.
#include "ceres/internal/port.h"

#include "ceres/sparse_normal_cholesky_solver.h"

#include <algorithm>
#include <cstring>
#include <ctime>

#include "ceres/compressed_row_sparse_matrix.h"
#include "ceres/cxsparse.h"
#include "ceres/internal/eigen.h"
#include "ceres/internal/scoped_ptr.h"
#include "ceres/linear_solver.h"
#include "ceres/suitesparse.h"
#include "ceres/triplet_sparse_matrix.h"
#include "ceres/types.h"
#include "ceres/wall_time.h"
#include "Eigen/SparseCore"


namespace ceres {
namespace internal {

SparseNormalCholeskySolver::SparseNormalCholeskySolver(
    const LinearSolver::Options& options)
    : factor_(NULL),
      cxsparse_factor_(NULL),
      options_(options){
}

void SparseNormalCholeskySolver::FreeFactorization() {
  if (factor_ != NULL) {
    ss_.Free(factor_);
    factor_ = NULL;
  }

  if (cxsparse_factor_ != NULL) {
    cxsparse_.Free(cxsparse_factor_);
    cxsparse_factor_ = NULL;
  }
}

SparseNormalCholeskySolver::~SparseNormalCholeskySolver() {
  FreeFactorization();
}

LinearSolver::Summary SparseNormalCholeskySolver::SolveImpl(
    CompressedRowSparseMatrix* A,
    const double* b,
    const LinearSolver::PerSolveOptions& per_solve_options,
    double * x) {

  const int num_cols = A->num_cols();
  VectorRef(x, num_cols).setZero();
  A->LeftMultiply(b, x);

  if (per_solve_options.D != NULL) {
    // Temporarily append a diagonal block to the A matrix, but undo
    // it before returning the matrix to the user.
    scoped_ptr<CompressedRowSparseMatrix> regularizer;
    if (A->col_blocks().size() > 0) {
      regularizer.reset(CompressedRowSparseMatrix::CreateBlockDiagonalMatrix(
                            per_solve_options.D, A->col_blocks()));
    } else {
      regularizer.reset(new CompressedRowSparseMatrix(
                            per_solve_options.D, num_cols));
    }
    A->AppendRows(*regularizer);
  }

  LinearSolver::Summary summary;
  switch (options_.sparse_linear_algebra_library_type) {
    case SUITE_SPARSE:
      summary = SolveImplUsingSuiteSparse(A, per_solve_options, x);
      break;
    case CX_SPARSE:
      summary = SolveImplUsingCXSparse(A, per_solve_options, x);
      break;
    case EIGEN_SPARSE:
      summary = SolveImplUsingEigen(A, per_solve_options, x);
      break;
    default:
      LOG(FATAL) << "Unknown sparse linear algebra library : "
                 << options_.sparse_linear_algebra_library_type;
  }

  if (per_solve_options.D != NULL) {
    A->DeleteRows(num_cols);
  }

  return summary;
}

LinearSolver::Summary SparseNormalCholeskySolver::SolveImplUsingEigen(
    CompressedRowSparseMatrix* A,
    const LinearSolver::PerSolveOptions& per_solve_options,
    double * rhs_and_solution) {
#ifndef CERES_USE_EIGEN_SPARSE

  LinearSolver::Summary summary;
  summary.num_iterations = 0;
  summary.termination_type = LINEAR_SOLVER_FATAL_ERROR;
   summary.message =
      "SPARSE_NORMAL_CHOLESKY cannot be used with EIGEN_SPARSE"
      "because Ceres was not built with support for"
      "Eigen's SimplicialLDLT decomposition."
      "This requires enabling building with -DEIGENSPARSE=ON.";
  return summary;

#else

  EventLogger event_logger("SparseNormalCholeskySolver::Eigen::Solve");

  LinearSolver::Summary summary;
  summary.num_iterations = 1;
  summary.termination_type = LINEAR_SOLVER_SUCCESS;
  summary.message = "Success.";

  // Compute the normal equations. J'J delta = J'f and solve them
  // using a sparse Cholesky factorization. Notice that when compared
  // to SuiteSparse we have to explicitly compute the normal equations
  // before they can be factorized. CHOLMOD/SuiteSparse on the other
  // hand can just work off of Jt to compute the Cholesky
  // factorization of the normal equations.
  //
  // TODO(sameeragarwal): See note about how this maybe a bad idea for
  // dynamic sparsity.
  if (outer_product_.get() == NULL) {
    outer_product_.reset(
        CompressedRowSparseMatrix::CreateOuterProductMatrixAndProgram(
            *A, &pattern_));
  }

  CompressedRowSparseMatrix::ComputeOuterProduct(
      *A, pattern_, outer_product_.get());

  // Map to an upper triangular column major matrix.
  //
  // outer_product_ is a compressed row sparse matrix and in lower
  // triangular form, when mapped to a compressed column sparse
  // matrix, it becomes an upper triangular matrix.
  //
  // TODO(sameeragarwal): It is not clear to me if an upper triangular
  // column major matrix is the way to go here, or if a lower
  // triangular matrix is better. This will require some testing. If
  // it turns out that the lower triangular is better, then the logic
  // used to compute the outer product needs to be updated.
  Eigen::MappedSparseMatrix<double, Eigen::ColMajor> AtA(
      outer_product_->num_rows(),
      outer_product_->num_rows(),
      outer_product_->num_nonzeros(),
      outer_product_->mutable_rows(),
      outer_product_->mutable_cols(),
      outer_product_->mutable_values());

  const Vector b = VectorRef(rhs_and_solution, outer_product_->num_rows());
  if (simplicial_ldlt_.get() == NULL || options_.dynamic_sparsity) {
    typedef Eigen::SimplicialLDLT<Eigen::SparseMatrix<double, Eigen::ColMajor>,
                                  Eigen::Upper> SimplicialLDLT;
    simplicial_ldlt_.reset(new SimplicialLDLT);
    // This is a crappy way to be doing this. But right now Eigen does
    // not expose a way to do symbolic analysis with a given
    // permutation pattern, so we cannot use a block analysis of the
    // Jacobian.
    simplicial_ldlt_->analyzePattern(AtA.selfadjointView<Eigen::Upper>());
    if (simplicial_ldlt_->info() != Eigen::Success) {
      summary.termination_type = LINEAR_SOLVER_FATAL_ERROR;
      summary.message =
          "Eigen failure. Unable to find symbolic factorization.";
      return summary;
    }
  }
  event_logger.AddEvent("Analysis");

  simplicial_ldlt_->factorize(AtA.selfadjointView<Eigen::Upper>());
  if(simplicial_ldlt_->info() != Eigen::Success) {
    summary.termination_type = LINEAR_SOLVER_FAILURE;
    summary.message =
        "Eigen failure. Unable to find numeric factorization.";
    return summary;
  }

  VectorRef(rhs_and_solution, outer_product_->num_rows()) =
      simplicial_ldlt_->solve(b);
  if(simplicial_ldlt_->info() != Eigen::Success) {
    summary.termination_type = LINEAR_SOLVER_FAILURE;
    summary.message =
        "Eigen failure. Unable to do triangular solve.";
    return summary;
  }

  event_logger.AddEvent("Solve");
  return summary;
#endif  // EIGEN_USE_EIGEN_SPARSE
}



LinearSolver::Summary SparseNormalCholeskySolver::SolveImplUsingCXSparse(
    CompressedRowSparseMatrix* A,
    const LinearSolver::PerSolveOptions& per_solve_options,
    double * rhs_and_solution) {
#ifdef CERES_NO_CXSPARSE

  LinearSolver::Summary summary;
  summary.num_iterations = 0;
  summary.termination_type = LINEAR_SOLVER_FATAL_ERROR;
  summary.message =
      "SPARSE_NORMAL_CHOLESKY cannot be used with CX_SPARSE"
      "because Ceres was not built with support for CXSparse."
      "This requires enabling building with -DCXSPARSE=ON.";

  return summary;

#else

  EventLogger event_logger("SparseNormalCholeskySolver::CXSparse::Solve");

  LinearSolver::Summary summary;
  summary.num_iterations = 1;
  summary.termination_type = LINEAR_SOLVER_SUCCESS;
  summary.message = "Success.";

  // Compute the normal equations. J'J delta = J'f and solve them
  // using a sparse Cholesky factorization. Notice that when compared
  // to SuiteSparse we have to explicitly compute the normal equations
  // before they can be factorized. CHOLMOD/SuiteSparse on the other
  // hand can just work off of Jt to compute the Cholesky
  // factorization of the normal equations.
  //
  // TODO(sameeragarwal): If dynamic sparsity is enabled, then this is
  // not a good idea performance wise, since the jacobian has far too
  // many entries and the program will go crazy with memory.
  if (outer_product_.get() == NULL) {
    outer_product_.reset(
        CompressedRowSparseMatrix::CreateOuterProductMatrixAndProgram(
            *A, &pattern_));
  }

  CompressedRowSparseMatrix::ComputeOuterProduct(
      *A, pattern_, outer_product_.get());
  cs_di AtA_view =
      cxsparse_.CreateSparseMatrixTransposeView(outer_product_.get());
  cs_di* AtA = &AtA_view;

  event_logger.AddEvent("Setup");

  // Compute symbolic factorization if not available.
  if (options_.dynamic_sparsity) {
    FreeFactorization();
  }
  if (cxsparse_factor_ == NULL) {
    if (options_.use_postordering) {
      cxsparse_factor_ = cxsparse_.BlockAnalyzeCholesky(AtA,
                                                        A->col_blocks(),
                                                        A->col_blocks());
    } else {
      if (options_.dynamic_sparsity) {
        cxsparse_factor_ = cxsparse_.AnalyzeCholesky(AtA);
      } else {
        cxsparse_factor_ = cxsparse_.AnalyzeCholeskyWithNaturalOrdering(AtA);
      }
    }
  }
  event_logger.AddEvent("Analysis");

  if (cxsparse_factor_ == NULL) {
    summary.termination_type = LINEAR_SOLVER_FATAL_ERROR;
    summary.message =
        "CXSparse failure. Unable to find symbolic factorization.";
  } else if (!cxsparse_.SolveCholesky(AtA, cxsparse_factor_, rhs_and_solution)) {
    summary.termination_type = LINEAR_SOLVER_FAILURE;
    summary.message = "CXSparse::SolveCholesky failed.";
  }
  event_logger.AddEvent("Solve");

  return summary;
#endif
}

LinearSolver::Summary SparseNormalCholeskySolver::SolveImplUsingSuiteSparse(
    CompressedRowSparseMatrix* A,
    const LinearSolver::PerSolveOptions& per_solve_options,
    double * rhs_and_solution) {
#ifdef CERES_NO_SUITESPARSE

  LinearSolver::Summary summary;
  summary.num_iterations = 0;
  summary.termination_type = LINEAR_SOLVER_FATAL_ERROR;
  summary.message =
      "SPARSE_NORMAL_CHOLESKY cannot be used with SUITE_SPARSE"
      "because Ceres was not built with support for SuiteSparse."
      "This requires enabling building with -DSUITESPARSE=ON.";
  return summary;

#else

  EventLogger event_logger("SparseNormalCholeskySolver::SuiteSparse::Solve");
  LinearSolver::Summary summary;
  summary.termination_type = LINEAR_SOLVER_SUCCESS;
  summary.num_iterations = 1;
  summary.message = "Success.";

  const int num_cols = A->num_cols();
  cholmod_sparse lhs = ss_.CreateSparseMatrixTransposeView(A);
  event_logger.AddEvent("Setup");

  if (options_.dynamic_sparsity) {
    FreeFactorization();
  }
  if (factor_ == NULL) {
    if (options_.use_postordering) {
      factor_ = ss_.BlockAnalyzeCholesky(&lhs,
                                         A->col_blocks(),
                                         A->row_blocks(),
                                         &summary.message);
    } else {
      if (options_.dynamic_sparsity) {
        factor_ = ss_.AnalyzeCholesky(&lhs, &summary.message);
      } else {
        factor_ = ss_.AnalyzeCholeskyWithNaturalOrdering(&lhs, &summary.message);
      }
    }
  }
  event_logger.AddEvent("Analysis");

  if (factor_ == NULL) {
    summary.termination_type = LINEAR_SOLVER_FATAL_ERROR;
    // No need to set message as it has already been set by the
    // symbolic analysis routines above.
    return summary;
  }

  summary.termination_type = ss_.Cholesky(&lhs, factor_, &summary.message);
  if (summary.termination_type != LINEAR_SOLVER_SUCCESS) {
    return summary;
  }

  cholmod_dense* rhs = ss_.CreateDenseVector(rhs_and_solution, num_cols, num_cols);
  cholmod_dense* solution = ss_.Solve(factor_, rhs, &summary.message);
  event_logger.AddEvent("Solve");

  ss_.Free(rhs);
  if (solution != NULL) {
    memcpy(rhs_and_solution, solution->x, num_cols * sizeof(*rhs_and_solution));
    ss_.Free(solution);
  } else {
    // No need to set message as it has already been set by the
    // numeric factorization routine above.
    summary.termination_type = LINEAR_SOLVER_FAILURE;
  }

  event_logger.AddEvent("Teardown");
  return summary;
#endif
}

}   // namespace internal
}   // namespace ceres