Skip to content

Parallelization of Weight Window Update #3467

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
jtramm merged 8 commits into from
Jun 25, 2025
Merged

Parallelization of Weight Window Update #3467

Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
f597be6
parallelized ww update function
jtramm Jun 24, 2025
bc640be
more parallelization
jtramm Jun 24, 2025
f0a547e
more parallel
jtramm Jun 24, 2025
7a0bc7f
combined loops
jtramm Jun 24, 2025
0e6876f
changed to static, and more loops parallelized
jtramm Jun 24, 2025
097dfdd
refactored loop index variable names and mesh to be int64_t
jtramm Jun 24, 2025
52139a7
Removal of unneeded eval causing a copy
jtramm Jun 24, 2025
1a61952
cleanup
jtramm Jun 24, 2025
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
Prev Previous commit
Next Next commit
more parallelization
  • Loading branch information
@jtramm
jtramm committed Jun 24, 2025
commit bc640be88e5f658bfb1dc2c4c82deb37805ba96f
35 changes: 21 additions & 14 deletions src/weight_windows.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -626,21 +626,26 @@ void WeightWindows::update_weights(const Tally* tally, const std::string& value,
auto& new_bounds = this->lower_ww_;
auto& rel_err = this->upper_ww_;

// Use original xtensor operations for initial calculations - these are already optimized
// get mesh volumes and dimensions
auto mesh_vols = this->mesh()->volumes();
int e_bins = new_bounds.shape()[0];
int mesh_bins = new_bounds.shape()[1];

// Use xtensor operations for the initial calculations but evaluate them immediately
xt::noalias(new_bounds) = sum / n;

xt::noalias(rel_err) =
xt::sqrt(((sum_sq / n) - xt::square(new_bounds)) / (n - 1)) / new_bounds;
xt::filter(rel_err, sum <= 0.0).fill(INFTY);

if (value == "rel_err")
xt::noalias(new_bounds) = 1 / rel_err;

// get mesh volumes
auto mesh_vols = this->mesh()->volumes();

int e_bins = new_bounds.shape()[0];
int mesh_bins = new_bounds.shape()[1];
if (value == "rel_err") {
#pragma omp parallel for collapse(2) schedule(runtime)
for (int e = 0; e < e_bins; e++) {
for (int m = 0; m < mesh_bins; m++) {
new_bounds(e, m) = 1.0 / rel_err(e, m);
}
}
Comment thread
jtramm marked this conversation as resolved.
Resolved
}

if (method == WeightWindowUpdateMethod::MAGIC) {
// If we are computing weight windows with forward fluxes derived from a
Expand All @@ -654,11 +659,12 @@ void WeightWindows::update_weights(const Tally* tally, const std::string& value,
}
}

// Second pass: find group maximum and normalize (this needs to be sequential per energy group)
// Second pass: find group maximum and normalize (per energy group)
for (int e = 0; e < e_bins; e++) {
double group_max = 0.0;

// Find maximum in this energy group
// Find maximum in this energy group using parallel reduction
#pragma omp parallel for schedule(runtime) reduction(max:group_max)
for (int i = 0; i < mesh_bins; i++) {
if (new_bounds(e, i) > group_max) {
group_max = new_bounds(e, i);
Expand Down Expand Up @@ -698,8 +704,9 @@ void WeightWindows::update_weights(const Tally* tally, const std::string& value,
}
}

// Find the maximum value across all elements
// Find the maximum value across all elements using parallel reduction
double max_val = 0.0;
#pragma omp parallel for collapse(2) schedule(runtime) reduction(max:max_val)
for (int e = 0; e < e_bins; e++) {
for (int i = 0; i < mesh_bins; i++) {
if (new_bounds(e, i) > max_val) {
Expand All @@ -719,8 +726,9 @@ void WeightWindows::update_weights(const Tally* tally, const std::string& value,
}
}

// Find minimum value for missed bins
// Find minimum value for missed bins using parallel reduction
double min_val = INFTY;
#pragma omp parallel for collapse(2) schedule(runtime) reduction(min:min_val)
for (int e = 0; e < e_bins; e++) {
for (int i = 0; i < mesh_bins; i++) {
if (new_bounds(e, i) != 0.0 && new_bounds(e, i) < min_val) {
Expand All @@ -742,7 +750,6 @@ void WeightWindows::update_weights(const Tally* tally, const std::string& value,

// make sure that values where the mean is zero are set s.t. the weight window
// value will be ignored
// Note: We'll use the original xtensor filtering for this since it involves the sum view
xt::filter(new_bounds, sum <= 0.0).fill(-1.0);

// make sure the weight windows are ignored for any locations where the
Expand Down