[GPU] Optimize Condition operation by better integrating its subnetwo…

…rk primitives with the main network
sshlyapn · Nov 14, 2023 · 2bf4ca1 · 2bf4ca1
1 parent c451a94
commit 2bf4ca1
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 9 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/condition.cpp b/src/plugins/intel_gpu/src/graph/condition.cpp
@@ -257,7 +257,7 @@ void condition_inst::postprocess_output_memory(network::ptr executed_net, cldnn:
     for (auto out_mem_map : branch.output_map) {
         auto out_mem_idx = out_mem_map.first;
         auto inner_out_id = out_mem_map.second;
-        auto mem_ptr = executed_net->get_output(inner_out_id).get_memory();
+        auto mem_ptr = executed_net->get_output_memory(inner_out_id);
         if (mem_ptr) {
             auto layout = _impl_params->get_output_layout(out_mem_idx);
             GPU_DEBUG_LOG << "Reshape output from " << mem_ptr->get_layout().to_short_string()

diff --git a/src/plugins/intel_gpu/src/graph/impls/common/condition.cpp b/src/plugins/intel_gpu/src/graph/impls/common/condition.cpp
@@ -30,13 +30,15 @@ struct condition_impl : typed_primitive_impl<condition> {
     }
 
     event::ptr execute_impl(const std::vector<event::ptr>& events, condition_inst& instance) override {
-        for (auto& a : events) {
-            a->wait();
-        }
-        auto ev = instance.get_network().get_stream().create_user_event(false);
+        // Wait for condition statement event only, and pass all other events to sub-network directly
+        if (!events.empty())
+            events[0]->wait();
+
+        auto& stream = instance.get_network().get_stream();
+        auto ev = stream.create_user_event(false);
         set_node_params(instance.get_node());
 
-        auto pred = condition_inst::get_pred_from_memory(instance.pred_memory_ptr(), instance.get_network().get_stream());
+        auto pred = condition_inst::get_pred_from_memory(instance.pred_memory_ptr(), stream);
         network::ptr executed_net = pred ? instance.get_net_true() : instance.get_net_false();
         auto branch = pred ? instance.get_branch_true() : instance.get_branch_false();
         executed_net->set_shape_predictor(instance.get_network().get_shape_predictor());
@@ -62,16 +64,22 @@ struct condition_impl : typed_primitive_impl<condition> {
             }
         }
 
-        executed_net->execute({});
+        // Ignore condition statement event
+        std::vector<event::ptr> sub_net_events(events.begin() + 1, events.end());
+        auto sub_net_results = executed_net->execute(sub_net_events);
 
         // Update output layout of impl_param in condition_inst
         instance.update_output_layout();
 
         // Set output memory of condition_inst to inner network output memory after inner network execution
         instance.postprocess_output_memory(executed_net, branch);
 
-        ev->set();
-        return ev;
+        std::vector<event::ptr> output_events;
+        for (auto& output : sub_net_results)
+            if (output.second.get_event() != nullptr)
+                output_events.push_back(output.second.get_event());
+
+        return stream.group_events(output_events);
     }
 
     static std::unique_ptr<primitive_impl> create(const condition_node& arg, const kernel_impl_params&) {