[GNA] Support caching properties for model cache (#14731)

* [GNA] Enable model caching

* [GNA] Enable model caching tests

* Fix: Use 'removeDir' instead of 'std::remove', on Windows std::remove has no effect

* Extract properties impacting model compilation into a separate method

* Add caching prop for supprted prop list

src/plugins/intel_gna/src/gna_plugin_config.cpp

@@ -268,6 +268,8 @@ OPENVINO_SUPPRESS_DEPRECATED_START
 OPENVINO_SUPPRESS_DEPRECATED_END
         } else if (key == CONFIG_KEY(LOG_LEVEL) || key == ov::log::level) {
             gnaFlags.log_level = ov::util::from_string(value, ov::log::level);
+        } else if (key == ov::cache_dir) {
+            cacheDir = value;
         } else {
             IE_THROW(NotFound)
                 << "[GNAPlugin] in function " << __PRETTY_FUNCTION__<< ": "
@@ -350,6 +352,7 @@ OPENVINO_SUPPRESS_DEPRECATED_END
     keyConfigMap[ov::enable_profiling.name()] =
             gnaFlags.performance_counting ? PluginConfigParams::YES: PluginConfigParams::NO;
     keyConfigMap[ov::log::level.name()] = ov::util::to_string(gnaFlags.log_level);
+    keyConfigMap[ov::cache_dir.name()] = cacheDir;
 }
 
 Parameter Config::GetParameter(const std::string& name) const {
@@ -381,9 +384,23 @@ Parameter Config::GetParameter(const std::string& name) const {
     }
 }
 
-const Parameter Config::GetSupportedProperties(bool compiled) {
+const Parameter Config::GetImpactingModelCompilationProperties(bool compiled) {
     ov::PropertyMutability model_mutability = compiled ? ov::PropertyMutability::RO : ov::PropertyMutability::RW;
     const std::vector<ov::PropertyName> supported_properties = {
+        {ov::intel_gna::scale_factors_per_input.name(), model_mutability},
+        {ov::intel_gna::firmware_model_image_path.name(), model_mutability},
+        {ov::intel_gna::execution_target.name(), model_mutability},
+        {ov::intel_gna::compile_target.name(), model_mutability},
+        {ov::intel_gna::pwl_design_algorithm.name(), model_mutability},
+        {ov::intel_gna::pwl_max_error_percent.name(), model_mutability},
+        {ov::inference_precision.name(), model_mutability},
+        {ov::hint::num_requests.name(), model_mutability},
+    };
+    return supported_properties;
+}
+
+const Parameter Config::GetSupportedProperties(bool compiled) {
+    std::vector<ov::PropertyName> supported_properties = {
         { ov::supported_properties.name(), ov::PropertyMutability::RO },
         { ov::available_devices.name(), ov::PropertyMutability::RO },
         { ov::optimal_number_of_infer_requests.name(), ov::PropertyMutability::RO },
@@ -391,19 +408,20 @@ const Parameter Config::GetSupportedProperties(bool compiled) {
         { ov::device::capabilities.name(), ov::PropertyMutability::RO },
         { ov::device::full_name.name(), ov::PropertyMutability::RO },
         { ov::intel_gna::library_full_version.name(), ov::PropertyMutability::RO },
-        { ov::intel_gna::scale_factors_per_input.name(), model_mutability },
-        { ov::intel_gna::firmware_model_image_path.name(), model_mutability },
+        { ov::caching_properties.name(), ov::PropertyMutability::RO},
         { ov::intel_gna::execution_mode.name(), ov::PropertyMutability::RW },
-        { ov::intel_gna::execution_target.name(), model_mutability },
-        { ov::intel_gna::compile_target.name(), model_mutability },
-        { ov::intel_gna::pwl_design_algorithm.name(), model_mutability },
-        { ov::intel_gna::pwl_max_error_percent.name(), model_mutability },
         { ov::hint::performance_mode.name(), ov::PropertyMutability::RW },
-        { ov::inference_precision.name(), model_mutability },
-        { ov::hint::num_requests.name(), model_mutability },
         { ov::log::level.name(), ov::PropertyMutability::RW },
         { ov::execution_devices.name(), ov::PropertyMutability::RO },
+        { ov::cache_dir.name(), ov::PropertyMutability::RW },
     };
+
+    const std::vector<ov::PropertyName> impacting_model_compilation_properties =
+        GetImpactingModelCompilationProperties(compiled);
+
+    supported_properties.insert(supported_properties.end(),
+                                impacting_model_compilation_properties.begin(),
+                                impacting_model_compilation_properties.end());
     return supported_properties;
 }
 

src/plugins/intel_gna/src/gna_plugin_config.hpp

@@ -45,11 +45,13 @@ struct Config {
         gnaFlags = r.gnaFlags;
         std::lock_guard<std::mutex> lock(r.mtx4keyConfigMap);
         keyConfigMap = r.keyConfigMap;
+        cacheDir = r.cacheDir;
     }
     void UpdateFromMap(const std::map<std::string, std::string>& configMap);
     void AdjustKeyMapValues();
     InferenceEngine::Parameter GetParameter(const std::string& name) const;
     std::vector<std::string> GetSupportedKeys() const;
+    static const InferenceEngine::Parameter GetImpactingModelCompilationProperties(bool compiled);
     static const InferenceEngine::Parameter GetSupportedProperties(bool compiled = false);
 
     ov::hint::PerformanceMode performance_mode = ov::hint::PerformanceMode::UNDEFINED;
@@ -73,8 +75,8 @@ struct Config {
 
     mutable std::mutex mtx4keyConfigMap;
     std::map<std::string, std::string> keyConfigMap;
-
     static const uint8_t max_num_requests = 127;
+    std::string cacheDir;
 };
 
 }  // namespace intel_gna

src/plugins/intel_gna/src/gna_plugin_query_api.cpp

@@ -6,6 +6,7 @@
 #include "gna_plugin.hpp"
 #include "gna/gna_config.hpp"
 #include "openvino/runtime/intel_gna/properties.hpp"
+#include <cpp_interfaces/interface/ie_internal_plugin_config.hpp>
 
 #include <string>
 #include <map>
@@ -61,6 +62,10 @@ Parameter GNAPlugin::GetMetric(const std::string& name, const std::map<std::stri
         return decltype(ov::execution_devices)::value_type {GetName()};
     } else if (ov::model_name == name) {
         return _network_name;
+    } else if (name == ov::caching_properties) {
+        std::vector<ov::PropertyName> cachingProperties = Config::GetImpactingModelCompilationProperties(true);
+        cachingProperties.push_back(ov::PropertyName(ov::log::level.name(), ov::PropertyMutability::RO));
+        return decltype(ov::caching_properties)::value_type(cachingProperties);
     } else {
         const std::unordered_map<std::string, std::function<Parameter()>> queryApiSupported = {
             {METRIC_KEY(AVAILABLE_DEVICES), [this]() {return GetAvailableDevices();}},

src/plugins/intel_gna/tests/functional/shared_tests_instances/skip_tests_config.cpp

@@ -94,12 +94,9 @@ std::vector<std::string> disabledTestPatterns() {
         R"(.*CompileModelCacheTestBase.*(SingleConv|NestedSplitConvConcat).*)",
         R"(.*CompileModelCacheTestBase.*(Bias|ReadConcatSplitAssign).*)",
         R"(.*OVClassLoadNetworkTest.*LoadNetwork.*)",
-        // TODO: Issue: 95234
-        R"(.*smoke_CachingSupportCase_GNA.*)",
+        // does not work due to GNA 3.0 convolution and other primitives limitations, partially can be resolved by switching GNA library to GNA3.5
         R"(.*CachingSupportCase.*LoadNet.*(Bias|Split|Concat|KSO|SingleConv).*)",
         R"(.*CachingSupportCase.*LoadNet.*(ConvPoolRelu|TIwithLSTMcell1)_f32_batch2.*)",
-        // TODO: Issue: 95234
-        R"(.*CachingSupportCase_GNA.*)",
         R"(.*IEClassLoadNetworkTest.*LoadNetwork(HETERO|MULTI|WithDeviceIDNoThrow|WithInvalidDeviceIDThrows).*)",
     };
 }

src/plugins/intel_gna/tests/unit/gna_executable_network_metrics_test.cpp

@@ -61,9 +61,9 @@ TEST_F(GnaExecutableNetworkMetricsTest, TestNetworkName) {
 TEST_F(GnaExecutableNetworkMetricsTest, TestSupportedProperties) {
     std::string supportedProperties =
         "SUPPORTED_PROPERTIES AVAILABLE_DEVICES OPTIMAL_NUMBER_OF_INFER_REQUESTS RANGE_FOR_ASYNC_INFER_REQUESTS "
-        "OPTIMIZATION_CAPABILITIES FULL_DEVICE_NAME GNA_LIBRARY_FULL_VERSION GNA_SCALE_FACTOR_PER_INPUT "
-        "GNA_FIRMWARE_MODEL_IMAGE GNA_DEVICE_MODE GNA_HW_EXECUTION_TARGET GNA_HW_COMPILE_TARGET "
-        "GNA_PWL_DESIGN_ALGORITHM GNA_PWL_MAX_ERROR_PERCENT PERFORMANCE_HINT INFERENCE_PRECISION_HINT "
-        "PERFORMANCE_HINT_NUM_REQUESTS LOG_LEVEL EXECUTION_DEVICES";
+        "OPTIMIZATION_CAPABILITIES FULL_DEVICE_NAME GNA_LIBRARY_FULL_VERSION CACHING_PROPERTIES "
+        "GNA_DEVICE_MODE PERFORMANCE_HINT LOG_LEVEL EXECUTION_DEVICES CACHE_DIR "
+        "GNA_SCALE_FACTOR_PER_INPUT GNA_FIRMWARE_MODEL_IMAGE GNA_HW_EXECUTION_TARGET GNA_HW_COMPILE_TARGET "
+        "GNA_PWL_DESIGN_ALGORITHM GNA_PWL_MAX_ERROR_PERCENT INFERENCE_PRECISION_HINT PERFORMANCE_HINT_NUM_REQUESTS";
     Run(ov::supported_properties.name(), supportedProperties);
 }

src/plugins/intel_gna/tests/unit/gna_plugin_config_test.cpp

@@ -13,6 +13,7 @@ using namespace ov::intel_gna;
 
 IE_SUPPRESS_DEPRECATED_START
 const std::map<std::string, std::string>  supportedConfigKeysWithDefaults = {
+    {CONFIG_KEY(CACHE_DIR), ""},
     {GNA_CONFIG_KEY(SCALE_FACTOR), "1.000000"},
     {GNA_CONFIG_KEY(SCALE_FACTOR) + std::string("_0"), "1.000000"},
     {GNA_CONFIG_KEY(FIRMWARE_MODEL_IMAGE), ""},

src/tests/functional/plugin/shared/src/behavior/plugin/caching_tests.cpp

@@ -176,7 +176,7 @@ void LoadNetworkCacheTestBase::SetUp() {
 
 void LoadNetworkCacheTestBase::TearDown() {
     CommonTestUtils::removeFilesWithExt(m_cacheFolderName, "blob");
-    std::remove(m_cacheFolderName.c_str());
+    CommonTestUtils::removeDir(m_cacheFolderName);
     core->SetConfig({{CONFIG_KEY(CACHE_DIR), {}}});
     APIBaseTest::TearDown();
 }