Skip to content

Fix potential false sharing in ReduceOps.eval with OMP #4618

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
merged 2 commits into from
Aug 19, 2025
Merged

Fix potential false sharing in ReduceOps.eval with OMP #4618

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fbef867
Fix false sharing in ReduceOps.eval with OMP
AlexanderSinn Aug 18, 2025
becf1fb
apply suggestions
AlexanderSinn Aug 19, 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
72 changes: 46 additions & 26 deletions Src/Base/AMReX_Reduce.H
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1059,17 +1059,22 @@ public:
#ifdef AMREX_USE_OMP
#pragma omp parallel
#endif
for (MFIter mfi(mf,true); mfi.isValid(); ++mfi) {
Box const& b = mfi.growntilebox(nghost);
const int li = mfi.LocalIndex();
auto& rr = reduce_data.reference(OpenMP::get_thread_num());
const auto lo = amrex::lbound(b);
const auto hi = amrex::ubound(b);
for (int k = lo.z; k <= hi.z; ++k) {
for (int j = lo.y; j <= hi.y; ++j) {
for (int i = lo.x; i <= hi.x; ++i) {
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(rr, f(li,i,j,k));
}}}
{
ReduceTuple rr;
Reduce::detail::for_each_init<0, ReduceTuple, Ps...>(rr);
for (MFIter mfi(mf,true); mfi.isValid(); ++mfi) {
Box const& b = mfi.growntilebox(nghost);
const int li = mfi.LocalIndex();
const auto lo = amrex::lbound(b);
const auto hi = amrex::ubound(b);
for (int k = lo.z; k <= hi.z; ++k) {
for (int j = lo.y; j <= hi.y; ++j) {
for (int i = lo.x; i <= hi.x; ++i) {
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(rr, f(li,i,j,k));
}}}
}
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(
reduce_data.reference(OpenMP::get_thread_num()), rr);
}
}

Expand All @@ -1081,34 +1086,44 @@ public:
#ifdef AMREX_USE_OMP
#pragma omp parallel
#endif
for (MFIter mfi(mf,true); mfi.isValid(); ++mfi) {
Box const& b = mfi.growntilebox(nghost);
const int li = mfi.LocalIndex();
auto& rr = reduce_data.reference(OpenMP::get_thread_num());
const auto lo = amrex::lbound(b);
const auto hi = amrex::ubound(b);
for (int n = 0; n < ncomp; ++n) {
for (int k = lo.z; k <= hi.z; ++k) {
for (int j = lo.y; j <= hi.y; ++j) {
for (int i = lo.x; i <= hi.x; ++i) {
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(rr, f(li,i,j,k,n));
}}}}
{
ReduceTuple rr;
Reduce::detail::for_each_init<0, ReduceTuple, Ps...>(rr);
for (MFIter mfi(mf,true); mfi.isValid(); ++mfi) {
Box const& b = mfi.growntilebox(nghost);
const int li = mfi.LocalIndex();
const auto lo = amrex::lbound(b);
const auto hi = amrex::ubound(b);
for (int n = 0; n < ncomp; ++n) {
for (int k = lo.z; k <= hi.z; ++k) {
for (int j = lo.y; j <= hi.y; ++j) {
for (int i = lo.x; i <= hi.x; ++i) {
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(rr, f(li,i,j,k,n));
}}}}
}
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(
reduce_data.reference(OpenMP::get_thread_num()), rr);
}
}

template <typename D, typename F>
void eval (Box const& box, D & reduce_data, F&& f)
{
auto& rr = reduce_data.reference(OpenMP::get_thread_num());
using ReduceTuple = typename D::Type;
ReduceTuple rr;
Reduce::detail::for_each_init<0, ReduceTuple, Ps...>(rr);
call_f<D>(box, rr, std::forward<F>(f));
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(
reduce_data.reference(OpenMP::get_thread_num()), rr);
}

template <typename N, typename D, typename F,
typename M=std::enable_if_t<std::is_integral_v<N>> >
void eval (Box const& box, N ncomp, D & reduce_data, F const& f)
{
using ReduceTuple = typename D::Type;
auto& rr = reduce_data.reference(OpenMP::get_thread_num());
ReduceTuple rr;
Reduce::detail::for_each_init<0, ReduceTuple, Ps...>(rr);
const auto lo = amrex::lbound(box);
const auto hi = amrex::ubound(box);
for (N n = 0; n < ncomp; ++n) {
Expand All @@ -1117,17 +1132,22 @@ public:
for (int i = lo.x; i <= hi.x; ++i) {
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(rr, f(i,j,k,n));
}}}}
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(
reduce_data.reference(OpenMP::get_thread_num()), rr);
}

template <typename N, typename D, typename F,
typename M=std::enable_if_t<std::is_integral_v<N>> >
void eval (N n, D & reduce_data, F const& f)
{
using ReduceTuple = typename D::Type;
auto& rr = reduce_data.reference(OpenMP::get_thread_num());
ReduceTuple rr;
Reduce::detail::for_each_init<0, ReduceTuple, Ps...>(rr);
for (N i = 0; i < n; ++i) {
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(rr, f(i));
}
Reduce::detail::for_each_local<0, ReduceTuple, Ps...>(
reduce_data.reference(OpenMP::get_thread_num()), rr);
}

template <typename D>
Expand Down
Loading