Enable mmap only for gpu

openvinotoolkit · Oct 7, 2024 · 09904dc · 09904dc
1 parent 385d657
commit 09904dc
Show file tree

Hide file tree

Showing 7 changed files with 35 additions and 8 deletions.
diff --git a/src/inference/dev_api/openvino/runtime/iplugin.hpp b/src/inference/dev_api/openvino/runtime/iplugin.hpp
@@ -212,6 +212,12 @@ class OPENVINO_RUNTIME_API IPlugin : public std::enable_shared_from_this<IPlugin
      */
     const std::shared_ptr<ov::threading::ExecutorManager>& get_executor_manager() const;
 
+    /**
+     * @brief Check if plugin support mmap for cached model reading. Returns false is the method is not overrided by plugin.
+     * @return true if mmap is supported, false otherwise
+     */
+    virtual bool support_mmap_for_caching() const;
+
     virtual ~IPlugin() = default;
 
 protected:

diff --git a/src/inference/src/dev/core_impl.cpp b/src/inference/src/dev/core_impl.cpp
@@ -1354,7 +1354,7 @@ bool ov::CoreImpl::device_supports_internal_property(const ov::Plugin& plugin, c
 }
 
 bool ov::CoreImpl::device_supports_model_caching(const ov::Plugin& plugin) const {
-    return plugin.supports_model_caching();
+    return plugin.supports_model_caching() == ov::Plugin::CachingMode::unsupported ? false : true;
 }
 
 bool ov::CoreImpl::device_supports_cache_dir(const ov::Plugin& plugin) const {
@@ -1409,7 +1409,7 @@ ov::SoPtr<ov::ICompiledModel> ov::CoreImpl::load_model_from_cache(
     try {
         cacheContent.cacheManager->read_cache_entry(
             cacheContent.blobId,
-            coreConfig.get_enable_mmap(),
+            coreConfig.get_enable_mmap() && plugin.supports_model_caching() == ov::Plugin::CachingMode::mmap,
             [&](std::istream& networkStream) {
                 OV_ITT_SCOPE(FIRST_INFERENCE,
                              ov::itt::domains::LoadTime,

diff --git a/src/inference/src/dev/iplugin.cpp b/src/inference/src/dev/iplugin.cpp
@@ -67,6 +67,10 @@ std::shared_ptr<ov::ICore> ov::IPlugin::get_core() const {
     return m_core.lock();
 }
 
+bool ov::IPlugin::support_mmap_for_caching() const {
+    return false;
+}
+
 const std::shared_ptr<ov::threading::ExecutorManager>& ov::IPlugin::get_executor_manager() const {
     return m_executor_manager;
 }

diff --git a/src/inference/src/dev/plugin.cpp b/src/inference/src/dev/plugin.cpp
@@ -101,10 +101,16 @@ ov::Any ov::Plugin::get_property(const std::string& name, const AnyMap& argument
     return {m_ptr->get_property(name, arguments), {m_so}};
 }
 
-bool ov::Plugin::supports_model_caching() const {
-    bool supported(false);
-    supported = util::contains(get_property(ov::supported_properties), ov::device::capabilities) &&
-                util::contains(get_property(ov::device::capabilities), ov::device::capability::EXPORT_IMPORT) &&
-                util::contains(get_property(ov::internal::supported_properties), ov::internal::caching_properties);
+ov::Plugin::CachingMode ov::Plugin::supports_model_caching() const {
+    ov::Plugin::CachingMode supported = ov::Plugin::CachingMode::unsupported;
+    if (util::contains(get_property(ov::supported_properties), ov::device::capabilities) &&
+        util::contains(get_property(ov::device::capabilities), ov::device::capability::EXPORT_IMPORT) &&
+        util::contains(get_property(ov::internal::supported_properties), ov::internal::caching_properties)) {
+        bool support_mmap = false;
+        OV_PLUGIN_CALL_STATEMENT(
+            support_mmap = m_ptr->support_mmap_for_caching();
+        );
+        supported = support_mmap ? ov::Plugin::CachingMode::mmap : ov::Plugin::CachingMode::legacy;
+    }
     return supported;
 }
diff --git a/src/inference/src/dev/plugin.hpp b/src/inference/src/dev/plugin.hpp
@@ -74,7 +74,13 @@ class Plugin {
     T get_property(const ov::Property<T, M>& property, const AnyMap& arguments) const {
         return get_property(property.name(), arguments).template as<T>();
     }
-    bool supports_model_caching() const;
+
+    enum class CachingMode {
+        legacy,
+        mmap,
+        unsupported
+    };
+    CachingMode supports_model_caching() const;
 };
 
 }  // namespace ov

diff --git a/src/plugins/intel_gpu/include/intel_gpu/plugin/plugin.hpp b/src/plugins/intel_gpu/include/intel_gpu/plugin/plugin.hpp
@@ -64,6 +64,7 @@ class Plugin : public ov::IPlugin {
                                     const ov::AnyMap& properties) const override;
     ov::SoPtr<ov::IRemoteContext> create_context(const ov::AnyMap& remote_properties) const override;
     ov::SoPtr<ov::IRemoteContext> get_default_context(const ov::AnyMap& remote_properties) const override;
+    bool support_mmap_for_caching() const override;
 };
 
 }  // namespace intel_gpu

diff --git a/src/plugins/intel_gpu/src/plugin/plugin.cpp b/src/plugins/intel_gpu/src/plugin/plugin.cpp
@@ -219,6 +219,10 @@ ov::SoPtr<ov::IRemoteContext> Plugin::get_default_context(const AnyMap& params)
     return get_default_context(device_id);
 }
 
+bool Plugin::support_mmap_for_caching() const {
+    return true;
+}
+
 void Plugin::set_property(const ov::AnyMap &config) {
     auto update_config = [](ExecutionConfig& config, const ov::AnyMap& user_config) {
         config.set_user_property(user_config);