Skip to content

perf: parallelize tune_kernel epsilon sweep #12

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

Open
malik672 wants to merge 2 commits into
base: main
Choose a base branch
from
Open

perf: parallelize tune_kernel epsilon sweep #12

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
16d6a23
perf: parallelize tune_kernel epsilon sweep
malik672 Feb 15, 2026
116b99f
fix: use std::exp for portability
malik672 Feb 15, 2026
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
47 changes: 30 additions & 17 deletions include/igneous/data/topology.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -404,12 +404,15 @@ struct DiffusionTopology {
nanoflann::L2_Simple_Adaptor<float, PointCloudAdaptor>, PointCloudAdaptor,
3>;

static std::vector<float> build_epsilons() {
std::vector<float> epsilons;
epsilons.reserve(80);
for (float exponent = -10.0f; exponent < 10.0f; exponent += 0.25f) {
epsilons.push_back(std::pow(2.0f, exponent));
}
static const std::vector<float> &build_epsilons() {
static const std::vector<float> epsilons = [] {
std::vector<float> values;
values.reserve(80);
for (float exponent = -10.0f; exponent < 10.0f; exponent += 0.25f) {
values.push_back(std::pow(2.0f, exponent));
}
return values;
}();
return epsilons;
}

Expand All @@ -418,25 +421,35 @@ struct DiffusionTopology {
const std::vector<float> &epsilons) {
const int e_count = static_cast<int>(epsilons.size());
std::vector<float> averages(static_cast<size_t>(e_count), 0.0f);
std::vector<float> inv_eps(static_cast<size_t>(e_count), 0.0f);
std::vector<float> log_eps(static_cast<size_t>(e_count), 0.0f);
const float inv_nk = 1.0f / static_cast<float>(std::max(1, n * k));
const float *entry_data = entries.data();
const size_t entry_count = entries.size();

for (int e = 0; e < e_count; ++e) {
const float epsilon = std::max(epsilons[static_cast<size_t>(e)], 1e-12f);
float sum = 0.0f;
for (float value : entries) {
sum += std::exp(-value / epsilon);
}
averages[static_cast<size_t>(e)] = sum * inv_nk;
}
core::parallel_for_index(
0, e_count,
[&](int e) {
const float epsilon =
std::max(epsilons[static_cast<size_t>(e)], 1e-12f);
const float inv_eps_e = 1.0f / epsilon;
inv_eps[static_cast<size_t>(e)] = inv_eps_e;
log_eps[static_cast<size_t>(e)] = std::log(epsilon);
float sum = 0.0f;
for (size_t idx = 0; idx < entry_count; ++idx) {
sum += std::exp(-entry_data[idx] * inv_eps_e);
}
averages[static_cast<size_t>(e)] = sum * inv_nk;
},
4);

int best_idx = 0;
float best_criterion = -std::numeric_limits<float>::infinity();
for (int i = 0; i + 1 < e_count; ++i) {
const float avg0 = std::max(averages[static_cast<size_t>(i)], 1e-30f);
const float avg1 = std::max(averages[static_cast<size_t>(i + 1)], 1e-30f);
const float eps0 = std::max(epsilons[static_cast<size_t>(i)], 1e-12f);
const float eps1 = std::max(epsilons[static_cast<size_t>(i + 1)], 1e-12f);
const float denom = std::log(eps1) - std::log(eps0);
const float denom =
log_eps[static_cast<size_t>(i + 1)] - log_eps[static_cast<size_t>(i)];
if (std::abs(denom) < 1e-12f) {
continue;
}
Expand Down
Loading