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

Improve SYCL kernels in CSR table #2794

Merged
merged 4 commits into from
May 24, 2024
Merged

Improve SYCL kernels in CSR table #2794

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
e48e3ae
Improve SYCL kernels in CSR table
Vika-F May 21, 2024
db43d08
Enable sparse primitives testing on GPU
Vika-F May 21, 2024
afa97ac
Remove unneeded include
Vika-F May 21, 2024
feade7c
clang-format; revert changes in tests
Vika-F May 22, 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
62 changes: 26 additions & 36 deletions cpp/oneapi/dal/table/backend/csr_kernels.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -408,6 +408,7 @@ bool is_sorted(sycl::queue& queue,
// number of pairs of the subsequent elements in the data array that are sorted in desccending order,
// i.e. for which data[i] > data[i + 1] is true.
std::int64_t count_descending_pairs = 0L;

sycl::buffer<std::int64_t, 1> count_buf(&count_descending_pairs, sycl::range<1>(1));

// count the number of pairs of the subsequent elements in the data array that are sorted
Expand All @@ -418,10 +419,9 @@ bool is_sorted(sycl::queue& queue,
auto count_descending_reduction =
sycl::reduction(count_buf, cgh, sycl::ext::oneapi::plus<std::int64_t>());

cgh.parallel_for(sycl::nd_range<1>{ count - 1, 1 },
cgh.parallel_for(sycl::range<1>{ dal::detail::integral_cast<std::size_t>(count - 1) },
Copy link
Contributor

@avolkov-intel avolkov-intel May 24, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

Not sure, that this was necessary

count_descending_reduction,
[=](sycl::nd_item<1> idx, auto& count_descending) {
const auto i = idx.get_global_id(0);
[=](sycl::id<1> i, auto& count_descending) {
if (data[i] > data[i + 1])
count_descending.combine(1);
});
Expand Down Expand Up @@ -485,39 +485,29 @@ out_of_bound_type check_bounds(const array<T>& arr,
sycl::buffer<std::int64_t, 1> count_lt_buf(&count_lt_min, sycl::range<1>(1));
sycl::buffer<std::int64_t, 1> count_gt_buf(&count_gt_max, sycl::range<1>(1));

// count the number of elements which are less than min_vaule using sycl::reduction
auto event_count_lt_min = queue.submit([&](sycl::handler& cgh) {
cgh.depends_on(dependencies);
auto count_lt_reduction =
sycl::reduction(count_lt_buf, cgh, sycl::ext::oneapi::plus<std::int64_t>());

cgh.parallel_for(sycl::nd_range<1>{ count, 1 },
count_lt_reduction,
[=](sycl::nd_item<1> idx, auto& count_lt) {
const auto i = idx.get_global_id(0);
if (data[i] < min_value) {
count_lt.combine(1);
}
});
});

// count the number of elements which are greater than max_vaule using sycl::reduction
auto event_count_gt_max = queue.submit([&](sycl::handler& cgh) {
cgh.depends_on(dependencies);
auto count_gt_reduction =
sycl::reduction(count_gt_buf, cgh, sycl::ext::oneapi::plus<std::int64_t>());

cgh.parallel_for(sycl::nd_range<1>{ count, 1 },
count_gt_reduction,
[=](sycl::nd_item<1> idx, auto& count_gt) {
const auto i = idx.get_global_id(0);
if (data[i] > max_value) {
count_gt.combine(1);
}
});
});

sycl::event::wait_and_throw({ event_count_lt_min, event_count_gt_max });
// count the number of elements which are less than min_vaule and
// the the number of elements which are greater than max_value using sycl::reduction
Copy link
Contributor

@avolkov-intel avolkov-intel May 24, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

typo

queue
.submit([&](sycl::handler& cgh) {
cgh.depends_on(dependencies);
auto count_lt_reduction =
sycl::reduction(count_lt_buf, cgh, sycl::ext::oneapi::plus<std::int64_t>());
auto count_gt_reduction =
sycl::reduction(count_gt_buf, cgh, sycl::ext::oneapi::plus<std::int64_t>());

cgh.parallel_for(sycl::range<1>{ dal::detail::integral_cast<std::size_t>(count) },
count_lt_reduction,
count_gt_reduction,
[=](sycl::id<1> i, auto& count_lt, auto& count_gt) {
if (data[i] < min_value) {
count_lt.combine(1);
}
if (data[i] > max_value) {
count_gt.combine(1);
}
});
})
.wait_and_throw();

out_of_bound_type result{ out_of_bound_type::within_bounds };
if (count_lt_min > 0)
Expand Down