Skip to content

Use Kokkos::Parallel in Vector Operations #240

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
julianlitz merged 6 commits into from
May 4, 2026
Merged

Use Kokkos::Parallel in Vector Operations #240

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
6057acb
Use Kokkos::Parallel in Vector Operations
julianlitz May 2, 2026
c270cbc
Better Kokkos name
julianlitz May 2, 2026
317aa7d
Update vector_operations.h
julianlitz May 4, 2026
5770a01
Update vector_operations.h
julianlitz May 4, 2026
848a621
Update vector_operations.h
julianlitz May 4, 2026
42481a3
Update vector_operations.h
julianlitz May 4, 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
143 changes: 70 additions & 73 deletions include/LinearAlgebra/Vector/vector_operations.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,24 +18,21 @@ bool equals(ConstVector<T> lhs, ConstVector<T> rhs)
if (lhs.size() != rhs.size()) {
return false;
}

const std::size_t n = lhs.size();
for (std::size_t i = 0; i < n; ++i) {
if (!equals(lhs(i), rhs(i))) {
return false;
}
}
return true;
bool is_equal = true;
Kokkos::parallel_reduce(
"Vector: equals", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i, bool& local_is_equal) {
local_is_equal = local_is_equal && equals(lhs(i), rhs(i));
},
Kokkos::LAnd<bool>(is_equal));
return is_equal;
}

template <typename T>
void assign(Vector<T> lhs, const T& value)
{
std::size_t n = lhs.size();
#pragma omp parallel for if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
lhs(i) = value;
}
Kokkos::deep_copy(lhs, value);
}

template <typename T>
Expand All @@ -44,11 +41,9 @@ void add(Vector<T> result, ConstVector<T> x)
if (result.size() != x.size()) {
throw std::invalid_argument("Vectors must be of the same size.");
}
std::size_t n = result.size();
#pragma omp parallel for if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
result(i) += x(i);
}
const std::size_t n = result.size();
Kokkos::parallel_for(
"Vector: add", Kokkos::RangePolicy<>(0, n), KOKKOS_LAMBDA(const std::size_t i) { result(i) += x(i); });
}

template <typename T>
Expand All @@ -57,11 +52,9 @@ void subtract(Vector<T> result, ConstVector<T> x)
if (result.size() != x.size()) {
throw std::invalid_argument("Vectors must be of the same size.");
}
std::size_t n = result.size();
#pragma omp parallel for if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
result(i) -= x(i);
}
const std::size_t n = result.size();
Kokkos::parallel_for(
"Vector: subtract", Kokkos::RangePolicy<>(0, n), KOKKOS_LAMBDA(const std::size_t i) { result(i) -= x(i); });
}

template <typename T>
Expand All @@ -70,21 +63,18 @@ void linear_combination(Vector<T> x, const T& alpha, ConstVector<T> y, const T&
if (x.size() != y.size()) {
throw std::invalid_argument("Vectors must be of the same size.");
}
std::size_t n = x.size();
#pragma omp parallel for if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
x(i) = alpha * x(i) + beta * y(i);
}
const std::size_t n = x.size();
Kokkos::parallel_for(
"Vector: linear_combination", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i) { x(i) = alpha * x(i) + beta * y(i); });
}

template <typename T>
void multiply(Vector<T> x, const T& alpha)
{
std::size_t n = x.size();
#pragma omp parallel for if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
x(i) *= alpha;
}
const std::size_t n = x.size();
Kokkos::parallel_for(
"Vector: multiply", Kokkos::RangePolicy<>(0, n), KOKKOS_LAMBDA(const std::size_t i) { x(i) *= alpha; });
}

template <typename T>
Expand All @@ -93,25 +83,22 @@ T dot_product(ConstVector<T> lhs, ConstVector<T> rhs)
if (lhs.size() != rhs.size()) {
throw std::invalid_argument("Vectors must be of the same size.");
}

T result = 0.0;
std::size_t n = lhs.size();
#pragma omp parallel for reduction(+ : result) if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
result += lhs(i) * rhs(i);
}
const std::size_t n = lhs.size();
T result = T{0};
Kokkos::parallel_reduce(
"Vector: dot_product", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i, T& local_sum) { local_sum += lhs(i) * rhs(i); }, result);
return result;
}

template <typename T>
T l1_norm(ConstVector<T> x)
{
T result = 0.0;
std::size_t n = x.size();
#pragma omp parallel for reduction(+ : result) if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
result += std::abs(x(i));
}
const std::size_t n = x.size();
T result = T{0};
Kokkos::parallel_reduce(
"Vector: l1_norm", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i, T& local_sum) { local_sum += Kokkos::abs(x(i)); }, result);
return result;
}

Expand All @@ -120,41 +107,51 @@ template <typename T>
T l2_norm(ConstVector<T> x)
{
const std::size_t n = x.size();
// 1) find the largest absolute value
T scale = 0.0;
#pragma omp parallel for reduction(max : scale) if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
T abs_val = std::abs(x(i));
if (abs_val > scale) {
scale = abs_val;
}
}
// 2) if the largest absolute value is zero, the norm is zero

// 1) Find the largest absolute value
T scale = T{0};
Kokkos::parallel_reduce(
"Vector: l2_norm_scale", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i, T& local_max) {
const T abs_val = Kokkos::abs(x(i));
if (abs_val > local_max) {
local_max = abs_val;
}
},
Kokkos::Max<T>(scale));

// 2) If the largest absolute value is zero, the norm is zero
if (scale == T{0})
return T{0};
// 3) accumulate sum of squares of scaled entries
T sum = 0.0;
#pragma omp parallel for reduction(+ : sum) if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
T value = x(i) / scale;
sum += value * value;
}
// 4) rescale
return scale * std::sqrt(sum);

// 3) Accumulate sum of squares of scaled entries
T sum = T{0};
Kokkos::parallel_reduce(
"Vector: l2_norm_sum", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i, T& local_sum) {
const T value = x(i) / scale;
local_sum += value * value;
},
sum);

// 4) Rescale
return scale * Kokkos::sqrt(sum);
}

template <typename T>
T infinity_norm(ConstVector<T> x)
{
T result = 0.0;
std::size_t n = x.size();
#pragma omp parallel for reduction(max : result) if (n > 10'000)
for (std::size_t i = 0; i < n; ++i) {
T abs_value = std::abs(x(i));
if (abs_value > result) {
result = abs_value;
}
}
const std::size_t n = x.size();
T result = T{0};
Kokkos::parallel_reduce(
"Vector: infinity_norm", Kokkos::RangePolicy<>(0, n),
KOKKOS_LAMBDA(const std::size_t i, T& local_max) {
const T abs_value = Kokkos::abs(x(i));
if (abs_value > local_max) {
local_max = abs_value;
}
},
Kokkos::Max<T>(result));
return result;
}

Expand Down
Loading