Skip to content
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

Jump to bottom

Update OpenMP Kernels #211

Merged
merged 1 commit into from
Sep 23, 2024
Merged

Update OpenMP Kernels #211

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
82 changes: 38 additions & 44 deletions src/Spatter/Configuration.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,18 +219,6 @@ void ConfigurationBase::setup() {
if (dense_size < dense_size_)
dense_size = dense_size_;

// #ifdef USE_OPENMP
// if (kernel.compare("gather") == 0 || kernel.compare("multigather") == 0) {
// dense_perthread.resize(omp_threads);
// for (int j = 0; j < omp_threads; ++j) {
// dense_perthread[j].resize(dense_size);
// for (size_t i = 0; i < dense_perthread[j].size(); ++i) {
// dense_perthread[j][i] = rand();
// }
// }
// }
// #endif

if (sparse_size < sparse_size_)
sparse_size = sparse_size_;

Expand Down Expand Up @@ -570,20 +558,16 @@ void Configuration<Spatter::OpenMP>::gather(bool timed, unsigned long run_id) {

#pragma omp for
for (size_t i = 0; i < count; ++i) {
double *sl = &sparse[delta*i];
double *tl = &(dense_perthread[t][pattern_length*(i%wrap)]);
double *sl = sparse.data() + delta * i;
double *tl = dense_perthread[t].data() + pattern_length * (i % wrap);

#pragma omp simd
for (size_t j = 0; j < pattern_length; ++j) {
dense_perthread[t][j] = sparse[pattern[j] + delta * i];
tl[j] = sl[pattern[j]];
}
}
}

assert(dense_perthread[rand()%omp_threads][rand()%pattern_length]!=0);

std::atomic_thread_fence(std::memory_order_release);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @radelja - do we need to remove the atomic thread fence op here? I think this was added as a new "feature" for 2.0.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This was added in PR #12 to Jered's fork, and I assumed it was included as part of testing the gather kernel. Should I add this to the other kernels using the dense_perthread buffer?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should remove it for now, I think. It needs to be behind a flag when it is added. And should be on every kernel

if (timed) {
timer.stop();
time_seconds[run_id] = timer.seconds();
Expand All @@ -608,24 +592,16 @@ void Configuration<Spatter::OpenMP>::scatter(bool timed, unsigned long run_id) {

#pragma omp for
for (size_t i = 0; i < count; ++i) {
double *sl = &sparse[delta*i];
//double *tl = &(dense_perthread[t][0]);
//double *tl = &dense[pattern_length*i];
double *tl = sparse.data() + delta * i;
double *sl = dense_perthread[t].data() + pattern_length * (i % wrap);

#pragma omp simd
for (size_t j = 0; j < pattern_length; ++j) {
sl[pattern[j]] = dense[j];
tl[pattern[j]] = sl[j];
}
}
}

/*
#pragma omp parallel for simd
for (size_t i = 0; i < count; ++i)
for (size_t j = 0; j < pattern_length; ++j)
sparse[pattern[j] + delta * i] = dense[j + pattern_length * (i % wrap)];
*/

if (timed) {
timer.stop();
time_seconds[run_id] = timer.seconds();
Expand All @@ -645,11 +621,16 @@ void Configuration<Spatter::OpenMP>::scatter_gather(
if (timed)
timer.start();

#pragma omp parallel for simd
for (size_t i = 0; i < count; ++i)
for (size_t j = 0; j < pattern_length; ++j)
sparse_scatter[pattern_scatter[j] + delta_scatter * i] =
sparse_gather[pattern_gather[j] + delta_gather * i];
#pragma omp parallel for
for (size_t i = 0; i < count; ++i) {
double *tl = sparse_scatter.data() + delta_scatter * i;
double *sl = sparse_gather.data() + delta_gather * i;

#pragma omp simd
for (size_t j = 0; j < pattern_length; ++j) {
tl[pattern_scatter[j]] = sl[pattern_gather[j]];
}
}

if (timed) {
timer.stop();
Expand All @@ -671,15 +652,18 @@ void Configuration<Spatter::OpenMP>::multi_gather(

#pragma omp parallel
{

int t = omp_get_thread_num();

#pragma omp for
for (size_t i = 0; i < count; ++i)
for (size_t j = 0; j < pattern_length; ++j)
dense_perthread[t][j + pattern_length * (i % wrap)] =
sparse[pattern[pattern_gather[j]] + delta * i];
for (size_t i = 0; i < count; ++i) {
double *sl = sparse.data() + delta * i;
double *tl = dense_perthread[t].data() + pattern_length * (i % wrap);

#pragma omp simd
for (size_t j = 0; j < pattern_length; ++j) {
tl[j] = sl[pattern[pattern_gather[j]]];
}
}
}

if (timed) {
Expand All @@ -701,11 +685,21 @@ void Configuration<Spatter::OpenMP>::multi_scatter(
if (timed)
timer.start();

#pragma omp parallel for simd
for (size_t i = 0; i < count; ++i)
for (size_t j = 0; j < pattern_length; ++j)
sparse[pattern[pattern_scatter[j]] + delta * i] =
dense[j + pattern_length * (i % wrap)];
#pragma omp parallel
{
int t = omp_get_thread_num();

#pragma omp for
for (size_t i = 0; i < count; ++i) {
double *tl = sparse.data() + delta * i;
double *sl = dense_perthread[t].data() + pattern_length * (i % wrap);

#pragma omp simd
for (size_t j = 0; j < pattern_length; ++j) {
tl[pattern[pattern_scatter[j]]] = sl[j];
}
}
}

if (timed) {
timer.stop();
Expand Down