Skip to content

[GPU] Add event completion for optimized out instance #33122

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

Open
wilson-seok wants to merge 3 commits into
base: master
Choose a base branch
from
+5 −4
Open

[GPU] Add event completion for optimized out instance #33122

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e4ba248
add event completion for optimized out instance
wilson-seok Dec 4, 2025
6546b3f
add return when need_completion_events is true
wilson-seok Dec 5, 2025
9fe7c84
fix unit test about optimized out need_completion_event=true
wilson-seok Dec 9, 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
4 changes: 4 additions & 0 deletions src/plugins/intel_gpu/src/graph/impls/ocl/primitive_base.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -244,6 +244,10 @@ struct typed_primitive_impl_ocl : public typed_primitive_impl<PType> {
typed_primitive_inst<PType>& instance) override {
stream& stream = instance.get_network().get_stream();
if (instance.can_be_optimized()) {
if (instance.needs_completion_event()) {
stream.wait_for_events(events);
return instance.is_output() ? stream.create_user_event(true) : nullptr;
}
Copy link
Contributor

@isanghao isanghao Dec 5, 2025

Choose a reason for hiding this comment

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

Did you review existing places to handle needs_completion_event? If I remember correctly, we already handled such case and it is surprising that it is not working as expected..
I found this comment in primitive_inst.cpp and this seems to be relevant.

        // Prepare dependencies events in case of OOO queue, CPU implementation,
        // or optimized_out impl which has CPU users (needs_completion_event() && !is_output() condition)

Copy link
Contributor Author

@wilson-seok wilson-seok Dec 5, 2025

Choose a reason for hiding this comment

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

@isanghao Yes, the dep_events is created from prepare_primitive() as concat is optimized out/needs_completion_event()/!is_output(). And this dep_events is not waited in execute() because BMG has SyncMethods=none. That's why this accuracy happens.

return stream.aggregate_events(events, false, instance.is_output());
}
std::vector<event::ptr> tmp_events(events);
Expand Down
5 changes: 1 addition & 4 deletions src/plugins/intel_gpu/tests/unit/module_tests/network_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -141,24 +141,21 @@ TEST(network_test, has_proper_event_for_in_order_queue_optimized_out) {
net.set_input_data("input1", input_mem);
net.execute();

// reshape is optimized out with need_completion_event=true. So it doesn't have event.
ASSERT_TRUE(net.has_event("concat"));
ASSERT_TRUE(net.has_event("reshape"));
ASSERT_TRUE(net.has_event("reorder"));
ASSERT_TRUE(net.has_event("activation"));

auto concat_ev = net.get_primitive_event("concat");
auto reshape_ev = net.get_primitive_event("reshape");
auto reorder_ev = net.get_primitive_event("reorder");
auto activation_ev = net.get_primitive_event("activation");

OV_ASSERT_NO_THROW(downcast<ocl::ocl_base_event>(concat_ev.get()));
OV_ASSERT_NO_THROW(downcast<ocl::ocl_base_event>(reshape_ev.get()));
OV_ASSERT_NO_THROW(downcast<ocl::ocl_base_event>(reorder_ev.get()));
OV_ASSERT_NO_THROW(downcast<ocl::ocl_base_event>(activation_ev.get()));

// Check if we have real underlying OpenCL events
ASSERT_TRUE(downcast<ocl::ocl_base_event>(concat_ev.get())->get().get() != nullptr);
ASSERT_TRUE(downcast<ocl::ocl_base_event>(reshape_ev.get())->get().get() != nullptr);
ASSERT_TRUE(downcast<ocl::ocl_base_event>(reorder_ev.get())->get().get() != nullptr);
ASSERT_TRUE(downcast<ocl::ocl_base_event>(activation_ev.get())->get().get() != nullptr);
}
Expand Down
Loading