Files
ceres-solver/internal/ceres/iterative_schur_complement_solver.cc
T
Joydeep Biswas 829089053e CUDA CGNR, Part 4: CudaCgnrSolver
* Added CudaCgnrSolver, a new CUDA-accelerated CGNR.
* To use CudaCgnrSolver, the user must select CGNR as the linear_solver
  and CUDA_SPARSE as the sparse_linear_algebra_library.
* Updated ConjugateGradientSolver to work with an array of pointers to
  scratch to support CudaVectors as scratch.
* Moved CUDA initialization to run in Solver::Solve as needed.

Some performance comparisons on an Ubuntu 20.04 desktop with an
Intel i9-9940X CPU @ 3.30GHz, and an nVidia Quadro RTX 6000,
all configurations run with 24 threads, and 10 iterations.

=================================================
CGNR + CUDA_SPARSE + IDENTITY Preconditioner
problem-1778-993923-pre.txt
=================================================
Cost:
Initial                          2.563973e+08
Final                            1.724755e+06
Change                           2.546725e+08

Minimizer iterations                       11
Successful steps                            7
Unsuccessful steps                          4

Time (in seconds):
Preprocessor                         4.020158

  Residual only evaluation           1.567092 (10)
  Jacobian & residual evaluation     7.847130 (7)
  Linear solver                     31.688898 (10)
Minimizer                           46.834987

Postprocessor                        0.353974
Total                               51.209120

=================================================
SPARSE_SCHUR (CPU) + SUITE_SPARSE + AMD
problem-1778-993923-pre.txt
=================================================
Cost:
Initial                          2.563973e+08
Final                            1.651617e+06
Change                           2.547457e+08

Minimizer iterations                       11
Successful steps                           11
Unsuccessful steps                          0

Time (in seconds):
Preprocessor                        35.812003

  Residual only evaluation           1.658980 (10)
  Jacobian & residual evaluation    12.218799 (11)
  Linear solver                     76.409992 (10)
Minimizer                           98.809773

Postprocessor                        0.372712
Total                              134.994489

=================================================
ITERATIVE_SCHUR (CPU) + JACOBI Preconditioner
problem-1778-993923-pre.txt
=================================================
Cost:
Initial                          2.563973e+08
Final                            1.684447e+06
Change                           2.547128e+08

Minimizer iterations                       11
Successful steps                            8
Unsuccessful steps                          3

Time (in seconds):
Preprocessor                        15.331614

  Residual only evaluation           1.606114 (10)
  Jacobian & residual evaluation     8.502166 (8)
  Linear solver                    351.910080 (10)
Minimizer                          368.797327

Postprocessor                        0.363536
Total                              384.492478

=================================================
CGNR + CUDA_SPARSE + IDENTITY Preconditioner
problem-13682-4456117-pre.txt
=================================================
Cost:
Initial                          1.126372e+09
Final                            2.269329e+07
Change                           1.103678e+09

Minimizer iterations                       11
Successful steps                            7
Unsuccessful steps                          4

Time (in seconds):
Preprocessor                        19.140087

  Residual only evaluation           8.721920 (10)
  Jacobian & residual evaluation    41.955923 (7)
  Linear solver                    214.121861 (10)
Minimizer                          296.636890

Postprocessor                        1.971827
Total                              317.748804

Change-Id: I3a09f31aa6903f661e91f595afd39d427583e856
2022-08-15 23:44:34 -05:00

205 lines
8.0 KiB
C++

// Ceres Solver - A fast non-linear least squares minimizer
// Copyright 2015 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/iterative_schur_complement_solver.h"
#include <algorithm>
#include <cstring>
#include <utility>
#include <vector>
#include "Eigen/Dense"
#include "ceres/block_sparse_matrix.h"
#include "ceres/block_structure.h"
#include "ceres/conjugate_gradients_solver.h"
#include "ceres/detect_structure.h"
#include "ceres/implicit_schur_complement.h"
#include "ceres/internal/eigen.h"
#include "ceres/linear_solver.h"
#include "ceres/power_series_expansion_preconditioner.h"
#include "ceres/preconditioner.h"
#include "ceres/schur_jacobi_preconditioner.h"
#include "ceres/triplet_sparse_matrix.h"
#include "ceres/types.h"
#include "ceres/visibility_based_preconditioner.h"
#include "ceres/wall_time.h"
#include "glog/logging.h"
namespace ceres::internal {
IterativeSchurComplementSolver::IterativeSchurComplementSolver(
LinearSolver::Options options)
: options_(std::move(options)) {}
IterativeSchurComplementSolver::~IterativeSchurComplementSolver() = default;
LinearSolver::Summary IterativeSchurComplementSolver::SolveImpl(
BlockSparseMatrix* A,
const double* b,
const LinearSolver::PerSolveOptions& per_solve_options,
double* x) {
EventLogger event_logger("IterativeSchurComplementSolver::Solve");
CHECK(A->block_structure() != nullptr);
const int num_eliminate_blocks = options_.elimination_groups[0];
// Initialize a ImplicitSchurComplement object.
if (schur_complement_ == nullptr) {
DetectStructure(*(A->block_structure()),
num_eliminate_blocks,
&options_.row_block_size,
&options_.e_block_size,
&options_.f_block_size);
schur_complement_ = std::make_unique<ImplicitSchurComplement>(options_);
}
schur_complement_->Init(*A, per_solve_options.D, b);
const int num_schur_complement_blocks =
A->block_structure()->cols.size() - num_eliminate_blocks;
if (num_schur_complement_blocks == 0) {
VLOG(2) << "No parameter blocks left in the schur complement.";
LinearSolver::Summary summary;
summary.num_iterations = 0;
summary.termination_type = LinearSolverTerminationType::SUCCESS;
schur_complement_->BackSubstitute(nullptr, x);
return summary;
}
// Initialize the solution to the Schur complement system.
reduced_linear_system_solution_.resize(schur_complement_->num_rows());
reduced_linear_system_solution_.setZero();
if (options_.use_spse_initialization) {
PowerSeriesExpansionPreconditioner pse_solver(
schur_complement_.get(),
options_.max_num_spse_iterations,
options_.spse_tolerance);
pse_solver.RightMultiplyAndAccumulate(
schur_complement_->rhs().data(),
reduced_linear_system_solution_.data());
}
CreatePreconditioner(A);
if (preconditioner_ != nullptr) {
if (!preconditioner_->Update(*A, per_solve_options.D)) {
LinearSolver::Summary summary;
summary.num_iterations = 0;
summary.termination_type = LinearSolverTerminationType::FAILURE;
summary.message = "Preconditioner update failed.";
return summary;
}
}
ConjugateGradientsSolverOptions cg_options;
cg_options.min_num_iterations = options_.min_num_iterations;
cg_options.max_num_iterations = options_.max_num_iterations;
cg_options.residual_reset_period = options_.residual_reset_period;
cg_options.q_tolerance = per_solve_options.q_tolerance;
cg_options.r_tolerance = per_solve_options.r_tolerance;
LinearOperatorAdapter lhs(*schur_complement_);
LinearOperatorAdapter preconditioner(*preconditioner_);
Vector scratch[4];
for (int i = 0; i < 4; ++i) {
scratch[i] = Vector::Zero(schur_complement_->num_cols());
}
Vector* scratch_ptr[4] = {
&scratch[0], &scratch[1], &scratch[2], &scratch[3]
};
event_logger.AddEvent("Setup");
LinearSolver::Summary summary =
ConjugateGradientsSolver(cg_options,
lhs,
schur_complement_->rhs(),
preconditioner,
scratch_ptr,
reduced_linear_system_solution_);
if (summary.termination_type != LinearSolverTerminationType::FAILURE &&
summary.termination_type != LinearSolverTerminationType::FATAL_ERROR) {
schur_complement_->BackSubstitute(reduced_linear_system_solution_.data(),
x);
}
event_logger.AddEvent("Solve");
return summary;
}
void IterativeSchurComplementSolver::CreatePreconditioner(
BlockSparseMatrix* A) {
if (preconditioner_ != nullptr) {
return;
}
Preconditioner::Options preconditioner_options;
preconditioner_options.type = options_.preconditioner_type;
preconditioner_options.visibility_clustering_type =
options_.visibility_clustering_type;
preconditioner_options.sparse_linear_algebra_library_type =
options_.sparse_linear_algebra_library_type;
preconditioner_options.num_threads = options_.num_threads;
preconditioner_options.row_block_size = options_.row_block_size;
preconditioner_options.e_block_size = options_.e_block_size;
preconditioner_options.f_block_size = options_.f_block_size;
preconditioner_options.elimination_groups = options_.elimination_groups;
CHECK(options_.context != nullptr);
preconditioner_options.context = options_.context;
switch (options_.preconditioner_type) {
case IDENTITY:
preconditioner_ = std::make_unique<IdentityPreconditioner>(
schur_complement_->num_cols());
break;
case JACOBI:
preconditioner_ = std::make_unique<SparseMatrixPreconditionerWrapper>(
schur_complement_->block_diagonal_FtF_inverse());
break;
case SCHUR_POWER_SERIES_EXPANSION:
// Ignoring the value of spse_tolerance to ensure preconditioner stays
// fixed during the iterations of cg.
preconditioner_ = std::make_unique<PowerSeriesExpansionPreconditioner>(
schur_complement_.get(), options_.max_num_spse_iterations, 0);
break;
case SCHUR_JACOBI:
preconditioner_ = std::make_unique<SchurJacobiPreconditioner>(
*A->block_structure(), preconditioner_options);
break;
case CLUSTER_JACOBI:
case CLUSTER_TRIDIAGONAL:
preconditioner_ = std::make_unique<VisibilityBasedPreconditioner>(
*A->block_structure(), preconditioner_options);
break;
default:
LOG(FATAL) << "Unknown Preconditioner Type";
}
};
} // namespace ceres::internal