Skip to content

[SYCL][E2E] Resolve reqd_sub_group_size issue for get_coordinate tests #16429

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 4 commits into from
Dec 20, 2024
Merged

[SYCL][E2E] Resolve reqd_sub_group_size issue for get_coordinate tests #16429

Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
a78efa5
fix issue in get_coordinate_ops_impl.hpp
YixingZhang007 Dec 19, 2024
35b3a34
formate clean up
YixingZhang007 Dec 19, 2024
4c76c35
formate clean up
YixingZhang007 Dec 19, 2024
72951ea
code clean up
YixingZhang007 Dec 19, 2024
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
52 changes: 23 additions & 29 deletions sycl/test-e2e/Matrix/get_coordinate_ops_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,23 @@
template <typename T, size_t Rows, size_t Cols, layout Layout, use Use>
class matrix_process;

template <typename TResult, typename AccessorType>
void reduce_and_accumulate(sub_group sg, size_t sg_size, size_t global_idy,
AccessorType &global_acc, TResult *local_sums,
size_t count) {
for (size_t i = 0; i < count; i++) {
local_sums[i] = reduce_over_group(sg, local_sums[i], sycl::plus<>());

// Only the subgroup leader performs the global accumulation
if (global_idy % sg_size == 0) {
sycl::atomic_ref<TResult, sycl::memory_order::relaxed,
sycl::memory_scope::device>
aref(global_acc[i]);
aref.fetch_add(local_sums[i]);
}
}
}

template <typename T, typename TResult, size_t NUM_ROWS, size_t NUM_COLS,
size_t SROWS, size_t SCOLS, use Use, layout Layout, size_t VF>
void matrix_sum(big_matrix<T, NUM_ROWS / VF, NUM_COLS * VF> &M,
Expand All @@ -32,7 +49,7 @@ void matrix_sum(big_matrix<T, NUM_ROWS / VF, NUM_COLS * VF> &M,
{1, 1 * sg_size}),
[=](nd_item<2> spmd_item)
#ifdef SG_SZ
[[intel::reqd_sub_group_size(SG_SZ)]]
[[sycl::reqd_sub_group_size(SG_SZ)]]
#endif
{
// The submatrix API has to be accessed by all the workitems in a
Expand Down Expand Up @@ -83,29 +100,10 @@ void matrix_sum(big_matrix<T, NUM_ROWS / VF, NUM_COLS * VF> &M,
});
}

for (int i = 0; i < NUM_ROWS; i++) {
sum_local_rows[i] =
reduce_over_group(sg, sum_local_rows[i], sycl::plus<>());
// only Groups leader perform the global reduction
if (global_idy % sg_size == 0) {
sycl::atomic_ref<TResult, sycl::memory_order::relaxed,
sycl::memory_scope::device>
aref(v_rows[i]);
aref.fetch_add(sum_local_rows[i]);
}
}

for (int i = 0; i < NUM_COLS; i++) {
sum_local_cols[i] =
reduce_over_group(sg, sum_local_cols[i], sycl::plus<>());
// only Groups leader perform the global reduction
if (global_idy % sg_size == 0) {
sycl::atomic_ref<TResult, sycl::memory_order::relaxed,
sycl::memory_scope::device>
aref(v_cols[i]);
aref.fetch_add(sum_local_cols[i]);
}
}
reduce_and_accumulate(sg, sg_size, global_idy, v_rows,
sum_local_rows, NUM_ROWS);
reduce_and_accumulate(sg, sg_size, global_idy, v_cols,
sum_local_cols, NUM_COLS);
}); // parallel for
}).wait();
}
Expand All @@ -124,11 +122,7 @@ void test_get_coord_op() {
TResult sum_cols[Cols] = {0};
TResult sum_cols_ref[Cols] = {0};

for (int i = 0; i < Rows; i++) {
for (int j = 0; j < Cols; j++) {
M[i][j] = i + j;
}
}
matrix_fill(Rows, Cols, (T *)M, [](int i, int j) { return T(1) * (i + j); });

matrix_vnni<T>(Rows, Cols, *M, *Mvnni, VF);
big_matrix<T, Rows / VF, Cols * VF> MM((T *)&Mvnni);
Expand Down
Loading