Dont' need to create new deleter, add extra test case

Signed-off-by: Bogdan Pereanu <bogdan.pereanu@intel.com>
pereanub · Jan 31, 2025 · f71a8be · f71a8be
1 parent acb0b4f
commit f71a8be
Show file tree

Hide file tree

Showing 2 changed files with 55 additions and 13 deletions.
diff --git a/src/plugins/intel_npu/src/utils/src/zero/zero_wrappers.cpp b/src/plugins/intel_npu/src/utils/src/zero/zero_wrappers.cpp
@@ -252,20 +252,8 @@ std::shared_ptr<CommandQueue> CommandQueuePool::getCommandQueue(
         }
     }  // otherwise create a new object
 
-    // Create shared_ptr with a deleter
     _log.debug("Create Command Queue");
-    auto new_obj = std::shared_ptr<CommandQueue>(
-        new CommandQueue(init_structs, desc, group_ordinal),
-        [this, hash](CommandQueue* ptr) {
-            std::lock_guard<std::mutex> lock(_mutex);
-            if (_pool.at(hash).lock()) {
-                _log.debug("Don't destroy the command queue in case the shared ptr is in use!");
-                return;
-            }
-            _pool.erase(hash);
-            _log.debug("Destroy Command Queue");
-            delete ptr;
-        });
+    auto new_obj = std::make_shared<CommandQueue>(init_structs, desc, group_ordinal);
 
     auto pair = std::make_pair(hash, new_obj);
     _pool.emplace(pair);

diff --git a/src/plugins/intel_npu/tests/functional/internal/overload/compile_and_infer.hpp b/src/plugins/intel_npu/tests/functional/internal/overload/compile_and_infer.hpp
@@ -208,6 +208,60 @@ TEST_P(OVCompileAndInferRequest, CompiledModelWorkloadTypeUpdateAfterCompilation
     }
 }
 
+TEST_P(OVCompileAndInferRequest, CompiledModelWorkloadTypeUpdateAfterCompilationWithMultipleInfers) {
+    if (isCommandQueueExtSupported()) {
+        OV_ASSERT_NO_THROW(execNet = core->compile_model(function, target_device, configuration));
+
+        auto secondCompiledModel = core->compile_model(function, target_device, configuration);
+
+        ov::InferRequest req1, req2, req3;
+        OV_ASSERT_NO_THROW(req1 = execNet.create_infer_request());
+        OV_ASSERT_NO_THROW(req3 = secondCompiledModel.create_infer_request());
+        bool isCalled = false;
+        OV_ASSERT_NO_THROW(req1.set_callback([&](std::exception_ptr exception_ptr) {
+            ASSERT_EQ(exception_ptr, nullptr);
+            isCalled = true;
+        }));
+        OV_ASSERT_NO_THROW(req1.start_async());
+        OV_ASSERT_NO_THROW(req1.wait());
+        ASSERT_TRUE(isCalled);
+
+        OV_ASSERT_NO_THROW(req3.infer());
+
+        req1 = {};
+
+        ov::AnyMap modelConfiguration;
+        modelConfiguration[workload_type.name()] = WorkloadType::EFFICIENT;
+        OV_ASSERT_NO_THROW(execNet.set_property(modelConfiguration));
+        ASSERT_EQ(execNet.get_property(workload_type.name()).as<WorkloadType>(), WorkloadType::EFFICIENT);
+        OV_ASSERT_NO_THROW(req2 = execNet.create_infer_request())
+        OV_ASSERT_NO_THROW(req2.infer());
+
+        modelConfiguration[workload_type.name()] = WorkloadType::DEFAULT;
+        OV_ASSERT_NO_THROW(execNet.set_property(modelConfiguration));
+        ASSERT_EQ(execNet.get_property(workload_type.name()).as<WorkloadType>(), WorkloadType::DEFAULT);
+        isCalled = false;
+        OV_ASSERT_NO_THROW(req2.set_callback([&](std::exception_ptr exception_ptr) {
+            ASSERT_EQ(exception_ptr, nullptr);
+            isCalled = true;
+        }));
+        OV_ASSERT_NO_THROW(req2.start_async());
+        OV_ASSERT_NO_THROW(req2.wait());
+        ASSERT_TRUE(isCalled);
+
+        req2 = {};
+        req3 = {};
+
+        OV_ASSERT_NO_THROW(req1 = execNet.create_infer_request());
+        OV_ASSERT_NO_THROW(req2 = secondCompiledModel.create_infer_request());
+        OV_ASSERT_NO_THROW(req1.infer());
+        OV_ASSERT_NO_THROW(req3 = execNet.create_infer_request());
+        OV_ASSERT_NO_THROW(req2.infer());
+        OV_ASSERT_NO_THROW(req3.infer());
+        OV_ASSERT_NO_THROW(req3.infer());
+    }
+}
+
 using OVCompileAndInferRequestTurbo = OVCompileAndInferRequest;
 
 TEST_P(OVCompileAndInferRequestTurbo, CompiledModelTurbo) {