Skip to content

Improve lidar odometry performance by replacing mutexes with thread-local reduction #321

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
JanuszBedkowski merged 1 commit into from
Jan 20, 2026
Merged

Improve lidar odometry performance by replacing mutexes with thread-local reduction #321

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
181 changes: 123 additions & 58 deletions apps/lidar_odometry_step_1/lidar_odometry_utils_optimizers.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,8 @@
#include "lidar_odometry_utils.h"
#include <hash_utils.h>
#include <mutex>
#include <tbb/blocked_range.h>
#include <tbb/combinable.h>
#include <tbb/parallel_for.h>

#include <export_laz.h>

Expand Down Expand Up @@ -1451,7 +1453,6 @@ static void compute_hessian_indoor(
bool ablation_study_use_view_point_and_normal_vectors,
bool ablation_study_use_planarity,
bool ablation_study_use_norm,
std::vector<std::mutex>& mutexes,
Eigen::MatrixXd& AtPAndt,
Eigen::MatrixXd& AtPBndt)
{
Expand Down Expand Up @@ -1577,8 +1578,6 @@ static void compute_hessian_indoor(
AtPB *= w;
}

std::mutex& m = mutexes[intermediate_points_i.index_pose];
std::unique_lock lck(m);
AtPAndt.block<6, 6>(c, c) += AtPA;
AtPBndt.block<6, 1>(c, 0) -= AtPB;
}
Expand All @@ -1592,7 +1591,6 @@ static void compute_hessian_outdoor(
double max_distance,
bool ablation_study_use_view_point_and_normal_vectors,
bool ablation_study_use_planarity,
std::vector<std::mutex>& mutexes,
Eigen::MatrixXd& AtPAndt,
Eigen::MatrixXd& AtPBndt)
{
Expand Down Expand Up @@ -1713,8 +1711,6 @@ static void compute_hessian_outdoor(
AtPB *= planarity;
}

std::mutex& m = mutexes[intermediate_points_i.index_pose];
std::unique_lock lck(m);
AtPAndt.block<6, 6>(c, c) += AtPA;
AtPBndt.block<6, 1>(c, 0) -= AtPB;
}
Expand Down Expand Up @@ -1777,73 +1773,142 @@ void optimize_lidar_odometry(
for (const auto& pose : intermediate_trajectory)
tait_bryan_poses.emplace_back(pose_tait_bryan_from_affine_matrix(pose));

std::vector<std::mutex> mutexes(intermediate_trajectory.size());

auto hessian_fun_indoor = [&intermediate_trajectory,
&tait_bryan_poses,
&buckets_indoor,
&b_indoor,
&mutexes,
&AtPAndt,
&AtPBndt,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
ablation_study_use_norm](const Point3Di& pt)
if (multithread)
{
compute_hessian_indoor(
pt,
intermediate_trajectory,
tait_bryan_poses,
buckets_indoor,
b_indoor,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
ablation_study_use_norm,
mutexes,
AtPAndt,
AtPBndt);
};
using MatrixPair = std::pair<Eigen::MatrixXd, Eigen::MatrixXd>;
tbb::combinable<MatrixPair> thread_local_matrices(
[&intermediate_trajectory]()
{
return std::make_pair(
Eigen::MatrixXd::Zero(intermediate_trajectory.size() * 6, intermediate_trajectory.size() * 6),
Eigen::MatrixXd::Zero(intermediate_trajectory.size() * 6, 1));
});

tbb::parallel_for(
tbb::blocked_range<size_t>(0, intermediate_points.size()),
[&intermediate_trajectory,
&intermediate_points,
&tait_bryan_poses,
&buckets_indoor,
&b_indoor,
&thread_local_matrices,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
ablation_study_use_norm](const tbb::blocked_range<size_t>& range)
{
auto& [local_AtPA, local_AtPB] = thread_local_matrices.local();
for (size_t i = range.begin(); i != range.end(); ++i)
{
compute_hessian_indoor(
intermediate_points[i],
intermediate_trajectory,
tait_bryan_poses,
buckets_indoor,
b_indoor,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
ablation_study_use_norm,
local_AtPA,
local_AtPB);
}
});

if (multithread)
std::for_each(std::execution::par_unseq, std::begin(intermediate_points), std::end(intermediate_points), hessian_fun_indoor);
thread_local_matrices.combine_each(
[&AtPAndt, &AtPBndt](const MatrixPair& local)
{
AtPAndt += local.first;
AtPBndt += local.second;
});
}
else
std::for_each(std::begin(intermediate_points), std::end(intermediate_points), hessian_fun_indoor);

if (ablation_study_use_hierarchical_rgd)
{
Eigen::Vector3d b_outdoor(rgd_params_outdoor.resolution_X, rgd_params_outdoor.resolution_Y, rgd_params_outdoor.resolution_Z);

auto hessian_fun_outdoor = [&intermediate_trajectory,
&tait_bryan_poses,
&buckets_outdoor,
&b_outdoor,
&mutexes,
&AtPAndt,
&AtPBndt,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity](const Point3Di& pt)
for (const auto& pt : intermediate_points)
{
compute_hessian_outdoor(
compute_hessian_indoor(
pt,
intermediate_trajectory,
tait_bryan_poses,
buckets_outdoor,
b_outdoor,
buckets_indoor,
b_indoor,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
mutexes,
ablation_study_use_norm,
AtPAndt,
AtPBndt);
};
}
}

if (ablation_study_use_hierarchical_rgd)
{
Eigen::Vector3d b_outdoor(rgd_params_outdoor.resolution_X, rgd_params_outdoor.resolution_Y, rgd_params_outdoor.resolution_Z);

if (multithread)
std::for_each(std::execution::par_unseq, std::begin(intermediate_points), std::end(intermediate_points), hessian_fun_outdoor);
{
using MatrixPair = std::pair<Eigen::MatrixXd, Eigen::MatrixXd>;
tbb::combinable<MatrixPair> thread_local_matrices(
[&intermediate_trajectory]()
{
return std::make_pair(
Eigen::MatrixXd::Zero(intermediate_trajectory.size() * 6, intermediate_trajectory.size() * 6),
Eigen::MatrixXd::Zero(intermediate_trajectory.size() * 6, 1));
});

tbb::parallel_for(
tbb::blocked_range<size_t>(0, intermediate_points.size()),
[&intermediate_trajectory,
&intermediate_points,
&tait_bryan_poses,
&buckets_outdoor,
&b_outdoor,
&thread_local_matrices,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity](const tbb::blocked_range<size_t>& range)
{
auto& [local_AtPA, local_AtPB] = thread_local_matrices.local();
for (size_t i = range.begin(); i != range.end(); ++i)
{
compute_hessian_outdoor(
intermediate_points[i],
intermediate_trajectory,
tait_bryan_poses,
buckets_outdoor,
b_outdoor,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
local_AtPA,
local_AtPB);
}
});

thread_local_matrices.combine_each(
[&AtPAndt, &AtPBndt](const MatrixPair& local)
{
AtPAndt += local.first;
AtPBndt += local.second;
});
}
else
std::for_each(std::begin(intermediate_points), std::end(intermediate_points), hessian_fun_outdoor);
{
for (const auto& pt : intermediate_points)
{
compute_hessian_outdoor(
pt,
intermediate_trajectory,
tait_bryan_poses,
buckets_outdoor,
b_outdoor,
max_distance,
ablation_study_use_view_point_and_normal_vectors,
ablation_study_use_planarity,
AtPAndt,
AtPBndt);
}
}
}

std::vector<std::pair<int, int>> odo_edges;
Expand Down